Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Social agent modeling and simulation : an aid to pre-adapting populations to serious societal disruptions Conroy, Patrick Francis 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2016_may_conroy_patrick.pdf [ 5.4MB ]
Metadata
JSON: 24-1.0228887.json
JSON-LD: 24-1.0228887-ld.json
RDF/XML (Pretty): 24-1.0228887-rdf.xml
RDF/JSON: 24-1.0228887-rdf.json
Turtle: 24-1.0228887-turtle.txt
N-Triples: 24-1.0228887-rdf-ntriples.txt
Original Record: 24-1.0228887-source.json
Full Text
24-1.0228887-fulltext.txt
Citation
24-1.0228887.ris

Full Text

Social Agent Modeling and Simulation: An Aid to Pre-Adapting Populations to Serious Societal Disruptions  by  Patrick Francis Conroy  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Electrical and Computer Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  April 2016  © Patrick Francis Conroy, 2016   ii Abstract  Serious societal disruptions can be expected in our future, and policy makers need better tools to help populations pre-adapt to them, in particular tools that incorporate internal-subjective behavior drivers and the means by which to model the behaviors they create. The Intelligent Agent paradigm developed in the Computer Science discipline is a powerful technology that enables population modeling at the level of the individual, and could have by now been demonstrably useful in creating tools to support policy makers’ work on this challenge. However, the Rational Agent instantiation of this paradigm, the focus of most Intelligent Agent work to date, is unsuitable for modeling the behavior of real human populations in a major societal disruption, due to avoidance of the internal-subjective bases of human sociality and so-called ‘irrational’ behaviors, exactly those that will dominate decision-making in such disruptions. There is growing understanding of such behavior drivers at the level of detail needed to support the modeling of significant-size populations. We propose a Social Agent instantiation of the Intelligent Agent paradigm for bottom-up modeling that explicitly incorporates these drivers, and we analyze the results of an implementation of this model in ‘EnergyWorld’, an abstract simulation of population behavior when resources needed for well-being are abruptly, significantly and persistently made scarce. Formal validation and verification in this research are limited due to the lack quantitative data on the internal-subjective nature of human decision-making; instead, we argue that the credibility of comparative policy analysis based on differential model parameter sets makes this approach useful for scenario-based policy analysis as a complement to other tools. We believe this to be true even if the disruptions we expect do not arrive in the intensity or form that seem likely today.   iii Preface  The research reported here was defined and executed solely by the author, under the general guidance of Dr. Philippe Kruchten of the Department of Electrical and Computer Engineering in the Faculty of Applied Science at the University of British Columbia. No part of the author’s research reported here has been previously published in any venue, nor was it part of any collaborative effort. No part of the research required Ethics Board approval.   iv Table of Contents Abstract .......................................................................................................................................... ii	Preface ........................................................................................................................................... iii	Table of Contents ......................................................................................................................... iv	List of Figures ............................................................................................................................. viii	List of Abbreviations ................................................................................................................... ix	Acknowledgements ........................................................................................................................ x	Dedication ..................................................................................................................................... xi	Chapter 1: Introduction ................................................................................................................1	1.1	 The Problem: A Critical Gap in Our Knowledge .............................................................. 1	1.2	 Our Research Question: Can the Gap be Bridged? ............................................................ 6	Chapter 2: Toward a Solution: The Intelligent Agent Paradigm .............................................7	2.1	 A Brief History: From Novelty to Go-To Technology ...................................................... 8	2.2	 Maturity Rising: Institutions, Standards, Tools ............................................................... 11	2.2.1	 Institutions, Conferences, Consortia, Collaborations ............................................... 12	2.2.2	 Standards, Guidelines, Frameworks, Taxonomies .................................................... 12	2.2.3	 Tools ......................................................................................................................... 15	2.3	 Application of the Paradigm: A Summary Across Domains ........................................... 19	2.3.1	 ‘Exclusively Rational’ Intelligent Agent Initiatives ................................................. 19	2.3.2	 ‘Boundedly Rational’ ABM Initiatives ..................................................................... 21	2.3.3	 ‘Socially Dominant’ ABM Initiatives ....................................................................... 28	2.4	 Assessment of the Work to Date: Enough to Fill the Gap? ............................................. 28	Chapter 3: Overlooked: Internal-subjective Human Behavior Drivers .................................32	  v 3.1	 Human Sociality: In a Major Societal Disruption, Emotion Dominates Thought ........... 32	3.2	 Human ‘Irrationality’: Important Features as Opposed to Flaws .................................... 34	3.3	 Other Elements to Consider in Our ABM ........................................................................ 39	3.3.1	 Needs and Wants ....................................................................................................... 39	3.3.2	 ‘Free Rider’ Response ............................................................................................... 40	3.3.3	 ‘Bounce’ .................................................................................................................... 41	3.3.4	 System 1 / System 2 .................................................................................................. 41	3.3.5	 Management Theory ................................................................................................. 43	3.3.6	 Behavior Modification .............................................................................................. 44	3.3.7	 Cognitive Behavior Therapy ..................................................................................... 45	3.3.8	 Modeling with Missing or Uncertain Empirical Data ............................................... 45	3.4	 Summing Up: We Know What Is Needed ....................................................................... 46	Chapter 4: ‘EnergyWorld’, A Context for Policy Exploration ...............................................47	4.1	 An Abstraction for Feasibility and Clarity ....................................................................... 47	4.2	 The ‘Social Agent’: An Instantiation of the Intelligent Agent Paradigm ........................ 48	4.3	 Functional Implementation .............................................................................................. 61	4.4	 Technical Implementation ............................................................................................... 63	Chapter 5: Explorations with EnergyWorld .............................................................................66	5.1	 Taking The Policy Maker Perspective ............................................................................. 66	5.2	 ‘Objective Function’ Candidates ..................................................................................... 67	5.3	 Energy Supply Disruption Candidates ............................................................................. 69	5.4	 Sensitivity Analysis ......................................................................................................... 70	5.5	 Simulation Scenarios ....................................................................................................... 72	  vi 5.5.1	 Extreme Rational Agent vs Extreme Social Agent ................................................... 73	5.5.2	 50/50 Mix of Extreme Social and Extreme Rational Agents .................................... 80	5.5.3	 Equilibrium Analysis ................................................................................................ 83	5.5.4	 Affinity Proportional To Power ................................................................................ 89	Chapter 6: Discussion ..................................................................................................................93	6.1	 EnergyWorld Abstraction and Simplicity vs Veridicality ............................................... 93	6.2	 Did We Inadvertently ‘Cook the Books’? ....................................................................... 95	6.3	 Interpretation Bias ............................................................................................................ 96	6.4	 Formal Research Validity: Applicability and Limitations ............................................... 98	6.5	 How Realistic is EnergyWorld’s Architecture? ............................................................. 101	6.6	 Credibility: A Better Measure… .................................................................................... 102	Chapter 7: Conclusions .............................................................................................................104	7.1	 Summary of Results ....................................................................................................... 104	7.2	 Have We Answered Our Research Question? ............................................................... 106	7.3	 What Does This Imply? ................................................................................................. 108	7.4	 Recommendations for Future Research ......................................................................... 108	Bibliography ...............................................................................................................................110	Appendices ..................................................................................................................................115	Appendix A Research Materials Database ............................................................................. 115	Appendix B EnergyWorld State Variables ............................................................................. 116	Appendix C Cognitive Bias Detail ......................................................................................... 121	Appendix D EnergyWorld Code Base .................................................................................... 123	Appendix E EnergyWorld Sensitivity Analysis ..................................................................... 124	  vii Appendix F EnergyWorld Scenario Exploration Results ....................................................... 135	Appendix G Additional Resources ......................................................................................... 136	     viii List of Figures Figure 1: EnergyWorld Process Flow .................................................................................... 61 Figure 2: Example PDF for Assigning Agent Parameter Values ........................................... 62 Figure 3: SocialRational Simulation Scenario Cases ............................................................. 76 Figure 4: Selected Charts from SocialRational Simulation Scenario .................................... 77 Figure 5: Selected Charts from Mix of SocialRational Simulation Scenario ........................ 81 Figure 6: One-Bump Recovery Simulation Scenario Cases .................................................. 85 Figure 7: Selected Charts from One-Bump Recover Simulation Scenario ............................ 86 Figure 8: Further Selected Charts from One-Bump Recover Simulation Scenario ............... 86 Figure 9: AffinityProportionalToPower Simulation Scenario Cases ..................................... 90 Figure 10: Selected Charts from AffinityProportionalToPower Simulation Scenario .......... 91        ix List of Abbreviations  ABM: Agent-based modeling, Agent-based model, Agent-based modeling and simulation. BDI: Beliefs-Desires-Intentions framework for ABM. PDF: Probability Density Function.    x Acknowledgements  I am indebted to Dr. Philippe Kruchten, my supervisor, who understood and believed in what I was trying to do when most others did the eyebrow thing.  He also paid me the compliment of giving me complete freedom to explore the previously unexplored; I guess he judged that my relative maturity would keep me from wasting my time and his. He was mostly right, though he was always there with a few illuminating words when I had boxed myself into a darkened corner.  My thanks also to the University of British Columbia, specifically the Department of Electrical and Computer Engineering, the Faculty of Applied Science, and the Faculty of Graduate Studies for their understanding in allowing me the time I needed to complete this ever-expanding project and for providing much-appreciated financial support. These thanks extend also to Dr. Lawrence Ward in the University’s Department of Psychology, Faculty of Arts, and Dr. Martino Tran in the School of Community and Regional Planning, Faculty of Applied Science, who provided key guidance to the work reported here.  I also acknowledge the vision and financial support of Mitacs Canada, in the form of three ‘Accelerate’ grants and one ‘Commercialization Voucher’. The organization and individuals involved could easily have brushed off my crazy idea as tilting at windmills.  And finally, my desire to contribute generally to the human adventure via this research comes from the fact of my children Kimberley and Evan, who at first thought I was nuts to take on a challenge like this at my post-retirement age, but seeing me energized and engaged, approved.   xi Dedication  I dedicate this work to all those who are driven to paint outside the lines by values-based beliefs and contribution-focused commitments, as opposed to blindly adhering to orthodoxy.  I love school; so I wasn’t that surprised when, after retiring from a long professional career in the software industry, I found myself pursuing an idea in a restarted University journey I thought had been completed over 35 years earlier. This idea – one that galvanized me into a level of action suitable to 20 year olds when I should have been on the golf course enjoying late middle age - occurred to me while contemplating a sunset from my deck on Pender Island, British Columbia. Perhaps I should have just helped myself to another beer and shrugged off that crazy idea, but being a man of enthusiasms, I let this one propel me into what has been a seven-year quest to at last resolve a problem that had bested me in my professional career. As will be seen, it has taken me much further than that, but that’s a story that will unfold over time.  Little did I know that this quest would also bring me squarely up against a truism of the academic research world, namely that empirical researchers wanting a career that includes research grants and ultimately tenure go where data of statistically good quality is ready to hand or can be acquired, and craft their investigative efforts around that data so as to produce results deemed worthy of publication, i.e. replete with hypotheses, tests, correlations, explanatory power indicators, etc. intended to both validate the analysis mechanism and to verify predictions against objective reality. The darker side of this truism is that areas where such data   xii is not ready to hand and the necessary validation and verification are not a slam-dunk, are much less well investigated, despite the possibly significant findings such research might generate.  This feature of academe means that some important research questions, such as many in the area of human population behavior, are left uninvestigated. I’m speaking in particular of research that requires access to individuals’ interior lives, including as it does values, beliefs, desires, affinities, thought leadership and highly context-sensitive behaviors, not to mention those pesky ‘cognitive biases’ that make humans reliably not act as ‘rational agents’, or as they are sometimes referred to, homo economicus or homo sociopathicus. It’s no wonder that the bulk of the research into internal-subjective behavior drivers has consisted almost entirely of highly constrained lab experiments focused on atomic, isolated elements of human reasoning and decision – this is about as far as researchers can go in human decision-making research without bumping up against the reality that our internal-subjective lives are much too complex and too obscured to yield to experimental methods that demand the ability to control all but that which is under study.  So what am I really saying here? In bald terms, I’m saying that it is tantamount to academic suicide to conduct empirical human population research if the approach requires a focus on the subjective interiors of people in their real-life context, because there is very little objective data to be gathered, analyzed, and used to validate or invalidate hypotheses. Yet that is precisely what I have chosen to do, in service to the vision granted to me that day on my deck on Pender Island. I can make a respectable case for the credibility of my project results, but it probably falls short   xiii of the expectations of reviewers that demand validated models and verified predictions. Good thing I’m not intent on a standard academic career!  The past few decades of my life have been dominated by a growing intentionality that I make a difference, particularly so that my children and theirs will not suffer as much as they might otherwise do as a result of the dysfunctionality of the world we are leaving them. And that, in my judgment, is about taking chances that others - especially the guardians of current academic orthodoxies - may not be comfortable with, and leaving a body of work that others might choose to build on if they care to paint outside the lines a bit. Like I say in one of my email sign-off tags: “Life is more tryings than gettings, so the key to happiness is to value the tryings.” This project is my latest trying, and I have valued it a lot.   1 Chapter 1: Introduction  1.1 The Problem: A Critical Gap in Our Knowledge  Consider the following imaginary conversation of discovery, in which a thoughtful guide engages you in a series of questions: QUESTION: Are we likely to experience abrupt, significant, persistent disruptions in our economic, social or political systems within, say, the next 50 years? ANSWER: Certainly within 50 years, but I expect some such upheaval well before that, and it won’t be a short-lived event like a major storm, a tsunami or even an earthquake, the physical effects of which are over in weeks or months. No, we would be foolish not to expect some kind of disruption, maybe several, that will certainly affect our way of life for years, probably for decades and quite possibly even longer. And though this may not start with global finance it will certainly spread there - and then the chaotic interdependence of all our systems will affect all areas of life (think war, plague, resource depletion, power outages, water shortages, health services scarcity, food insecurity, and more). This will be much, much worse than the crisis of 2008, and more like Cuba’s loss of economic support from the USSR in the early 1990’s, the 1973 ‘oil shock’ in the US, the 1947 partition of India, and a host of other major societal disruptions that could in theory have been prepared for, but weren’t.  QUESTION: Do humans act mostly rationally in times of such disruptions?   2 ANSWER: Some do, but most folks would be so much in denial about the future possibility of such a crisis that when it actually occurred they would very quickly react primarily out of emotion, instinct, intuition and other ‘non-rational’ bases.  QUESTION: Can policy makers now accurately predict non-rational behaviors in such disruptions? ANSWER: It’s a no-brainer that non-rational behavior will happen, but exactly what, when, where, and how much is another story. I’m aware of research into very narrow aspects of non-rational behavior, but the experimental environment of that research bears no relationship to real life, so no I don’t believe that policy makers can at this point make accurate predictions of this sort.  QUESTION: Do they know how to reduce human propensity for undesirable behaviors once such disruptions occur? ANSWER: No, about all they would be able to do is try to enforce rules and regulations made ahead of time, and the in-denial public and their politicians wouldn’t permit more than vague, inadequate legislation so those rules and regulations wouldn’t be very effective. In any case, there has been virtually no research into the mitigation of non-rational behaviors, so any such efforts would be ad hoc and most likely uncoordinated across any significant population.  QUESTION: What about mitigation efforts ahead of time?   3 ANSWER: If you mean programs coordinated over whole populations in advance of such disruptions so that when a crisis arises, because folks would have already adjusted their assumptions, attitudes and personal circumstances, they can be counted on as individuals to avoid non-rational behavior, that would definitely be a good idea, for sure. But humans are very difficult to change, and the adjustments would not be simple things like have extra water on hand, gas in your car, some cash at home, etc.  QUESTION: Are the adjustments required mostly objective-external, or internal-subjective, in nature? ANSWER: It’s certainly easier to think about objective-external adjustments, but a second look shows it’s obvious that the internal-subjective adjustments are both more fundamental and much trickier to deal with. If you can change someone’s attitudes and assumptions, the external-objective adjustments will more or less follow along, so it seems clear that internal-subjective adjustments would be the place to focus.  QUESTION: What do most policy makers currently analyze most - objective-external or internal-subjective drivers of behavior? ANSWER: Ah, I see where you’re going. Most policy makers study what they can get the most data on, and that’s obviously the objective-external drivers of behavior. I bet I can anticipate your next question...    4 QUESTION: Does it make sense to make analysis of internal-subjective drivers more present in our analyses, and in a crucial sense more important than analysis of objective-external drivers? ANSWER: I pretty much answered that myself, didn’t I? Of course both kinds of behavior drivers need attention, but without getting the internal-subjective drivers right, the focus on objective-external drivers could largely be wasted.  QUESTION: But isn’t internal-subjective research really difficult to do? ANSWER: Yes, and that’s primarily because it’s more difficult to get the hard data needed to make testable hypotheses, support the formation of simulation models that can be properly validated and verify the predictions of those models. I suppose that there might come a time when some future brain scanning device can help us understand these drivers, which would be valuable and necessary because humans are very poor at providing normative data on their internal-subjective states. One thing, though - our policy makers are trained in the scientific method - a great tool for things that can be measured and that don’t change too fast - and are equipped with tools for applying that method, but are much less well equipped to deal with problems where internal-subjective data is key to resolving them.  QUESTION: What will the scientific orthodoxy make of having limited testable hypotheses, no model validation, no prediction verification? ANSWER: I would expect the good men and women of our scientific elite to be suspicious of analysis methods that lack the key elements that define the scientific method. Not that   5 they would reject it outright, but it will take some persuasion to win the general support of the scientific establishment which would in turn be key to persuading the policy analyst and planning communities to employ such methods and tools.  QUESTION: Would policy makers nonetheless welcome the ability to compare differential effects of policy alternatives on human behaviors? ANSWER: Hmm. You mean not predictions per se, but the ability to evaluate the expected effectiveness of one policy over another? Or even the ability to rank a suite of policies by some criteria other than verified predictive capacity? Yes, I think they would welcome such a capability, to be used in conjunction with all their other tools. Come to think of it, they already do something similar when they take polls - polls don’t predict, but properly done they give some clue as to what attitudes and assumptions are out there, which can influence the choice of policies and programs that eventually get implemented.  QUESTION: Might this help them select programs that, over time, mitigate undesirable behaviors in future societal disruptions through pre-adaptation of internal-subjective behavior drivers? ANSWER: Yes, I think so. It’s one thing to just make up rules and regulations and pass laws about objective-external behaviors that are imposed by fiat instead of by agreement, it’s another to expose people to the desirability of alternative attitudes, assumptions and other internal-subjective behavior drivers over time and with some subtlety. That sounds like propaganda, and I allow that it is that, but in service to a goal that justifies it. There’s a philosophical and ethical challenge in this idea, isn’t there?   6  QUESTION: Can we build on existing concepts and technologies to create a mechanism to support policy makers this way? ANSWER: I do believe that is possible, and I’ll bet you’re about to tell me in some detail just how to do that.  And so, with the aid of an imaginary Socrates, we arrive at the focus of this research project: to formulate a concept and construct an instantiation of it in model/simulation form, aimed at helping human population behavior investigators craft and recommend programs that have a chance of pre-adapting populations' internal-subjective behavior drivers to the coming realities of abrupt, significant and persistent disruptions.  1.2 Our Research Question: Can the Gap be Bridged?  Reflecting the above, our Research Question is twofold: (1) Can a human behavioral model based on internal-subjective drivers be sufficiently specified, calibrated, validated and verified to reliably produce credible predictions? (2) If not, can such models still be useful to policy makers tasked with pre-adapting human populations to abrupt, significant, persistent disruptions, given that these will demand human behavioral responses based on internal-subjective behavior drivers?    7 Chapter 2: Toward a Solution: The Intelligent Agent Paradigm  The research framework we have chosen for answering the Research Question above is the Intelligent Agent paradigm, designed as it is to model behaviors of individual, autonomous, communicating agents capable of gathering data in their environment and making behavioral decisions in attempting to meet specified goals. These characteristics make it theoretically possible to emulate real human population behavior by modeling the motives for their actions in a bottom-up fashion that avoids the statistical obfuscations built into top-down modeling.  In this section we explore what makes the Intelligent Agent paradigm particularly useful for answering our Research Question. This exploration starts with a brief history - meant to indicate that it was not a foregone conclusion that it would become, if not yet ubiquitous, an approach that almost every system designer must consider as a candidate for their system’s architectural framework. After identifying its tentative beginnings we then sketch out the trajectory of this paradigm as it gained traction in first the computer science discipline, but very quickly in the disciplines that had been starved for a suite of tools more suited to their challenges than the top-down, statistically-driven methods that dominated their analytical toolkits for so long.  One need only glance at the dates on the citations from this section to grasp the growing acknowledgment of this paradigm, for some because of its promise of more efficient development and maintenance effort over the system life cycle, and for others – like us – because it permits system designs that reflect in structural terms the way the topics of our systems actually operate in the real world. This latter opportunity allows researchers to avoid the   8 obfuscation of statistical aggregation and the artificiality of the tools that simply cannot peer through that obfuscation. On the other hand, it also demands that we strive for more veridicality in what we attempt to do, i.e. researchers must achieve much more consonance between their models and the way their subjects actually live both their internal and external lives.   The level of activity in the field makes an exhaustive review a practical impossibility, so we have limited our attentions in what follows to only the primary trends in adoption of the Intelligent Agent paradigm. We have attempted to sample a broad variety of disciplines to avoid the inadvertent self-sampling effects of the ‘academic silo’ phenomenon.  2.1 A Brief History: From Novelty to Go-To Technology One chronicler of the historical evolution of the Intelligent Agent paradigm (Faltings, 2000) recounts the emergence of the Intelligent Agent paradigm in a general context. Highlights: • The first Intelligent Agent, arguably, was ELIZA (Weizenbaum, 1966), a software program designed to conduct what appeared to be a psychologist’s side of a conversation with a human user/patient. The program’s conversational rules were simple-minded in the extreme. For example, “If the patient’s utterance contains the keyword ‘Mother’, say one of the variants in your lexicon like ‘Tell me more about your Mother’“, “If all generic responses have been used, repeat a randomly-selected leading question without any of the last 10 keywords in it“, etc. Clearly, this effort was meant to inspire support for future initiatives, not to provide a useful tool in the real world. What was unexpected was the effect the program had on even sophisticated ‘patients’ (some of whom knew how   9 ELIZA was constructed) who found themselves responding with honest, thoughtful, private answers meant to elicit professional-grade dialogue and advice from the ‘psychologist’! Obviously, our anthropomorphizing and empathic tendencies apply to technology-based artifacts as well as others. • The first multi-agent system was one proposed in 1987 (Minsky, 1987) but had to wait for its implementation until the practical need was in favorable relationship to the technology and effort required for its realization. In this early work, Minsky envisioned systems constructed of multiple, relatively simple, special-purpose, independent components that could activate and deactivate each other. Not a bad guess! • A 1991 re-imagining of Minsky’s work (Brooks, 1991) described, in Falting’s words “an architecture in which intelligent and complex behavior would be emergent in the interplay of many simple behaviors [each produced by] a simple agent whose activation is decided by a control architecture [such that] one could very easily build robust autonomous robots that had not been possible otherwise.” This exemplified the early ‘arms race’ between visionary architectures and the technologies for realizing them that is still in full throat.  • The initial commercial impetus for Agent technology was the rapidly increasing complexity of many software systems, for example operating systems. Some, like Microsoft Windows 2000, contain tens of millions of lines of code, making design and implementation errors an increasingly untenable technical challenge and financial burden. Faltings observes that “complex homogeneous software systems will be replaced by networks of communicating agents [which] can be written independently as long as they conform to a standard communication language.“   10 • Faltings also says, perhaps inviting a partisan challenge: "When Adam Smith published his classic work The Wealth of Nations, he introduced the argument that local decision making by groups of individuals was more likely to lead to fruitful results than central planning by governments – thus laying the foundations of modern market economies and unprecedented wealth. Agents will bring about a similar revolution in computer software.” Strong words, and perhaps not entirely welcome to those who decry the excesses of extreme individualism and market capitalism, and see in this description the possibility of a rogue robotic state. Still, the main message is that the Intelligent Agent paradigm will have a transformative and enduring effect on the way computing is done, particularly in its search for efficiency and effectiveness amid relentless innovation. • The Intelligent Agent paradigm relies on a different system architecture than those evolved for large systems to date. Current systems exhibit an integrated, layered architecture representing progressive abstraction from Resource Management at the lowest level to User Experience at the highest, and require careful and exquisite coordination of all the components of each design, and thus necessarily produce considerable complexity and error-proneness. In contrast, agent-based systems rely only on standard communication interfaces between agents, each of which can be constructed using individually-optimized design and implementation technologies independent of other agents, and in principle can achieve faster development, lower error rates, easier maintenance cycles, perhaps higher performance, and likely lower costs. • Researchers in the Artificial Intelligence community have employed and advanced the Intelligent Agent paradigm in a variety of ways, such as equipping agents with Bayesian network mechanisms for decision-making, reinforcement learning and other machine   11 learning algorithms, progressive abstraction memory mechanisms, and other increasingly arcane methods, largely in support of applications seeking to optimize some ‘objective function’. Others have emphasized methods by which gangs of agents could execute coordinated activities in search of a shared goal, and even formulate those goals, reminiscent of the manner in which data mining enables the discovery of hidden patterns and their exploitation. • Similarly, frameworks for coordination and negotiation among agents have been proposed, for example (Rosenschein, 1994).  Other works that would add breadth and depth to a history of the Intelligent Agent paradigm are (Jennings, 1998) and (E. H. Durfee, Lesser, V. R., Corkill, D. D., 1989).  2.2 Maturity Rising: Institutions, Standards, Tools  The formation of an international standards body dedicated to the subject of Intelligent Agents in its various forms is an indication of its growing maturity and acceptance. The Foundation for Intelligent Physical Agents (FIPA, www.FIPA.org), initially founded in 1996 as a Switzerland-based entity, was incorporated into the IEEE family of standards bodies in 2005 and currently shows membership in the 60+ range, primarily organizations in academe and the commercial sphere. Where this particular entity will nudge the paradigm given its leaning toward robotics is unclear, but this development is evidence of broadening support in the technology community and that it has transcended its original status as a novelty technology.    12 2.2.1 Institutions, Conferences, Consortia, Collaborations  Another indicator of the increasing maturity of the Intelligent Agent paradigm is the number of conferences, consortia and collaborations focused on it, such as ICMAS (www.icmas.org), AGENTLINK (http://www.ocopomo.eu/), OCOPOMO (http://www.ocopomo.eu/) and a host of other initiatives such as those centered around Dr. Kevin Leyton-Brown at my Alma Mater, the University of British Columbia (http://www.cs.ubc.ca/~kevinlb/). Attention from commercial technology leaders in the form of significant investment in core programs is an indicator of their belief in its future potential, exemplified by the CORBA (IBM) and JINI (Sun) architectures vying for dominance in the high-end commercial market. Some thought leaders even believe that traditional Operating Systems will eventually be rendered obsolete over time by such initiatives.  2.2.2 Standards, Guidelines, Frameworks, Taxonomies  Outpacing, and not always coordinating with, the above institutional developments are attempts to accelerate the emergence and application of the Intelligent Agent paradigm through standards guidelines and tools, such as the following: • A case has been made (Collins, 2015) for standards in ABM to become as ubiquitous as those in other, more established disciplines, and though Collins stops short of proposing specific standards, he identifies several existing standards bodies that could be appropriate stewards for these. • Another standards-related work (Hare M., 2004) suggests an initial taxonomy of ABM models, and though it has its origins in the environmental science domain, it is suggestive   13 of broader application. He attempts to “disentangle” the goals, approaches and vocabulary of the growing number of disparate disciplines utilizing, and in doing so, influencing the characterization of what ABM is and how it can be used, and proposes a 6-element taxonomy, some with subdivisions: (1) those that couple social and environmental models, (2) those that focus on micro-level decision making (knowledge-based inference rules vs goal-based rules vs action-based rules), (3) those that engage in social interaction (individuals vs groups, local vs external), (4) those that employ intrinsic signals vs extrinsic signals to adapt decision making and behavior modes (imitation vs social comparison vs repetition vs deliberation vs combinations thereof), (5) those that, in the course of simulations, adapt individual agent strategies vs those that do not, and (6) those that exhibit single vs multiple levels of decision-making (individuals, socially-linked groups, etc.). • The Bunge-Wand-Weir (BWW) meta-standard provides a basis for guiding and verifying the ontological consistency of model formulations (of all kinds, not just instantiations of the Intelligent Agent paradigm), by ensuring consistent treatment of the various kinds of objects and processes making up the domain being modeled and its internal representation. This meta-standard is an adaptation of a philosophical ontology (M. Bunge, Mahner, M., 1997), (M. Bunge, 1999) extended to information systems (Wand, 2008); illustrations of its application in ABM include one focused on object-oriented implementations (Kiwelekar, 2010), another focusing on domain-specific modeling (Becker, 2010), and one specifically focused on Agent-oriented modeling (Monu, 2005).   14 • Emphasizing the broad range of interests in ABM, (Hamilton, 2015) suggests an Integrated Assessment Framework for organizing the full range of organizational and social considerations involved. • The Beliefs-Desires-Intentions (BDI) framework (Rao, 1995), a useful expansion (Cohen, 1990) of BDI, and the distributed Multi-Agent Reasoning System (dMARS) framework (d'Inverno, 2004) have become strong guidelines, if not de facto design standards for the core reasoning mechanisms of Intelligent Agents. • The Observe-Orient-Decide-Act (OODA) process framework, attributed to USAF Colonel John Boyd, has been applied to Agile programming processes (Adolph, 2006). • The Overview-Design-Details (ODD) framework and its extension ODD+Decision (ODDD) (Muller, 2013), are examples of guidelines for ABM process architectures. • A framework (Shafaei, 2008) for managing system complexity arising from, for example, object-oriented design concepts (particularly inheritance), intermixed with agent-focused architectures has been suggested. It is too early to tell if this approach to hybrid architectures will attract attention; it is included here due to the conceptual attractiveness of the ‘holon’ concept as a means of managing complexity, originally conceived by Arthur Koestler in his 1967 book, The Ghost in the Machine as the core concept for managing complexity. • A ‘descriptor standard’ (S. Wolf, Bouchaud, J.-P., Cecconi, F., Cincotti, S., Dawid, H., Gintis, H., van der Hoog, S., Jaeger, C. C., Kovalevksy, D. V., Mandel, A., Paroussos, L., 2013) came out of the 100th Dahlem Conference “New Approaches in Economics after the Financial Crisis” (Berlin Dahlem, August 28-31, 2010) to promote consistency in the concepts and terms used in the ABM world.   15  Essentially a structured list of questions, the totality of answers to which form a consistent and complete description. It has been previously used at least once (S. Wolf, Furst, S., Mandel, A., Lass, W., Lincke, D., Pablo-Marti, F., Jaeger, C., 2013), and we have used it to describe our ABM; see the Applications section below.   The current state of the art in ABM is such that many researchers, determining that their particular domain and research questions require more specialized processes and decision-making mechanisms than most domain-agnostic versions comfortably support, adapt the above and other assets as needed or create tailor-made mechanisms. Some of the domain-specific applications reviewed below exhibit this phenomenon.  2.2.3 Tools  Powering the maturation process of the Intelligent Agent paradigm is a myriad of development tools, many initially authored by individual researchers needing new and stronger tools to perform ABM in their specific domains. Programming languages, support libraries, and design support tools are continuously arriving on the scene, most developed in support of domain-specific applications, some as projects in their own right that then fade into a rather sedate, secluded life except for the occasional flurry of attention by graduate students passing through the nexus of their development. Some, however, attract widespread collaboration and a rare few find their way into the development of end-user systems doing work in the hardcore commercial and regulatory world. As with the standards described above, many researchers find that the   16 functionality demands of their domains require extensions of generic languages, libraries and design tools, and even greenfield development of highly specialized tools; these decisions are not taken lightly, as such efforts can demand whole teams of collaborators working over several years, even decades; some of the applications reviewed below exhibit such decisions. An alphabetically ordered sampling of the most notable domain-agnostic programming languages employed in ABM initiatives: • AgentSpeak, created by Anand Rao in 1996, is an object-oriented programming language based on the BDI architecture and logic programming, aimed at aiding the understanding of the relation between practical implementations of the BDI architecture and its modal logics. Its use was greatly accelerated by the availability of the Jason interpreter of AgentSpeak programs. See http://jason.sourceforge.net/wp/description. • AnyLogic is a multi-method simulation modeling tool developed by The AnyLogic Company (former XJ Technologies). It supports agent-based, discrete event, and system dynamics simulation methodologies. See www.anylogic.com. • Buzz is a programming language specifically designed for simulating the action of small, relatively unintelligent ‘swarms’ of robotic devices; the primitives of the language permit control of individual or groups of such devices. This language has not been employed in a sufficiently large number of applications to judge how useful it will be in ABM applications. Its promotional literature suggests that its primary application will be with so-called drones, the applications for which are not likely to require much in the way of internal-subjective behavior drivers, so it is also not likely that its developers have not put much emphasis on this in its core architecture. See http://the.swarming.buzz.   17 • NetLogo is the current version of a series of multi-agent modeling languages that includes StarLogo and StarLogoT. NetLogo was authored by Uri Wilensky in 1999, and has been in continuous development since then at the Center for Connected Learning and Computer-Based Modeling (https://ccl.northwestern.edu/netlogo/). Its website refers primarily to teaching techniques for simulating natural and social phenomena. NetLogo runs on Java virtual machines and is capable of supporting “hundreds or thousands of independent agents”, suggesting that it may not scale well to larger simulations. • Simula is considered the first object-oriented programming language. The original and the current version, known as Simula 67, were created in the 1960s by Ole-Johan Dahl and Kristen Nygaard. As its name suggests, it was designed for doing simulations, and has been used in a wide range of applications such as simulating computer chip designs, modeling processes, protocols, algorithms, and other applications where design failures exposed by simulation are far preferable to the expensive business of building and testing. While highly influential in many areas of computer science for its underlying innovations, its application to domains in the natural world have been limited. See www.simula67.info. • SmallTalk is an object-oriented programming language, first implemented in the 1990s by Alan Kay and others, featuring message-passing as its only method calling mechanism. While not expressly focused on ABM, this feature’s implementation has been the inspiration for some of the communication implementations in ABM-focused languages. See http://www.smalltalk.org. • Repast is a family of ABM and simulation platforms of various computational power levels that has evolved over the past 15 years and today enjoys considerable success in   18 numerous disciplines. Some 70 papers, journal articles, books, thesis dissertations and conference papers are listed on their website (http://repast.sourceforge.net).  In addition to the above general programming platforms, other frameworks and tools with features expressly created to support ABM (though still largely domain-agnostic): • KIF (Genesereth, 1992) or Knowledge Interchange Format, is a tool aimed at providing a robust means for agents to request, offer and exchange information about their state, their environment, and their progress toward (possibly mutual) goals. • KQML (Finin, 1995), or Knowledge Query and Manipulation Language, is another language and protocol for communication among software agents and knowledge-based systems. • (Edmonds, 1998) has suggested a design framework for agents meant to learn and adapt via the Genetic Programming paradigm. • Platforms that support group decision making in multi-agent systems (E. H. Durfee, Kenny, P. G., Kluge, K. C., 1998),  (d'Inverno, 2004), and (Shoham, 1993). • ONTO-LINGUA (Gruber, 1993) is a distributed collaborative environment to browse, create, edit, modify, and use ontologies, used primarily by teams of (often geographically dispersed) researchers and developers working on a common project.  A Note to Avoid Confusion The term “Socially Intelligent Agents” (SIA) (Dautenhahn, 1998) has been used in ABM applications in the Sociology discipline, and at first glance appears to share our focus on internal-subjective human behavior, but this is not the case. SIA is situated within the Embodied   19 Artificial Life (EAL) framework, a particularization of the Artificial Life (AL) discipline. While AL distinguishes itself from the ‘intelligence as numeric computation’ stream within Artificial Intelligence (AI), the core goal in SIA is believability of Agents as perceived by humans interacting with them, as opposed to our focus of incorporating verified internal-subjective drivers of human behavior. A somewhat frivolous but illustrative example of SIA’s meaning of sociality is how CyberPets (an actual commercial line of lifelike ‘pets’) interact with their ‘owners’. Customer research confirms that it is the perception of the owners that provides the desired believability as opposed to external manifestations of internal characteristics. We considered distinguishing our Social Agents from SIA’s by adopting a different label, but in the end decided that readers, suitably alerted, would not have any difficulty with our use of the term.  2.3 Application of the Paradigm: A Summary Across Domains  There are thousands of research initiatives utilizing the Intelligent Agent paradigm in domain-specific applications; this volume of work clearly demonstrates enthusiastic embrace of the Intelligent Agent paradigm and its possibilities. In this section we review what we believe is an indicative sampling of these initiatives, grouped by whether they are primarily focused on ‘exclusive-rational’, ‘bounded-rational’, or ‘social-dominant’ techniques in an ABM framework.  2.3.1  ‘Exclusively Rational’ Intelligent Agent Initiatives  We start with initiatives in the ‘exclusively rational’ category. The typical aim of initiatives in this category is to develop methods that optimize some objective-external metric, without   20 concern for the method’s veridicality, i.e. the degree to which methods correspond to those humans actually use. This lack of concern for how humans actually analyze their daily life situations and how they respond to societal disruptions makes this category of ABM initiatives mostly irrelevant to our Research Question. We nonetheless present below a small number of such initiatives to give the reader some appreciation for why this is so.  For further clarification, initiatives in the ‘exclusive-rational’ category tend to employ highly mathematical concepts and constructs sufficient to enable proofs of existence, convergence and optimization - often a requirement for their commissioning due to the competitive aspirations of their (often) commercial sponsors. The analysis and decision making mechanisms of such applications vary, but the key point here is that they bear virtually no resemblance to how humans make decisions, as the agents employed are usually equipped with distinctly non-human methods and resources such as perfect long-term memory, mathematically arcane but powerful analytical and predictive capabilities, as well as super-computer level resources to make full use of these extra-human faculties.  In the following, we do not attempt to illuminate the research questions and their method of resolution in terms the average human would understand. Instead, we list representative citations that hint at the kind of analysis methods and tools used in these initiatives:  • Jiang, A. X., Leyton-Brown, K. (2013). "Polynomial-time computation of exact correlated equilibrium in compact games." Games and Economic Behavior 91 (2015): 347-359.   21 • Camerer, C. V., Ho T-H., Chong, J-K. (2002). “A Cognitive Hierarchy Theory of One-shot Games.” • Aumann, R. (1987). “Correlated equilibrium as an expression of Bayesian rationality.” Econometrica 55 (1) 1-18.  • Raghavan, P. (1998.) “Probabilistic construction of deterministic algorithms: Approximating packing integer programs.” Applied Mathematics 15, 1328-1343.  We trust that the above, as brief as it is, is sufficient to persuade the reader that these initiatives do not, and do not attempt to, represent internal-subjective human behavior drivers, particularly not those related to human sociality and ‘irrational’ behaviors, the drivers most important to policy makers in working out how best to pre-adapt populations to abrupt, significant, persistent societal disruptions. We do not consider this category of ABM initiatives further.  2.3.2  ‘Boundedly Rational’ ABM Initiatives  Our definition of ‘boundedly rational’ in this context is that agents possess primarily objective-external behavior drivers in their analytical and decision-making mechanisms, but these are limited to those that procedurally simple, require modest environmental data, and require limited computing resources.  The concept of ‘bounded rationality’ has been around for some time (Simon, 1991), (G. Gigerenzer, 2006), (Arthur, 1994) and many others, and was a point of departure from researchers in economics arising from the acknowledgment that ‘rational agent’ behavioral   22 models block our attempts to fully understand the way humans analyze and decide, which really matters when individuals and groups are, or will be, faced with situations that automatically and forcefully bring out our sociality and our ‘irrational’ behaviors.  The applications described below represent a small fraction of current efforts of the ‘bounded rationality’ type. They represent an impressive range of domains and share the same primary motivation, i.e. to produce tools useful in real life problems. In each case, the primary method of analysis and decision-making by agents is indicated. • Behavioral Economics. One source of resistance to the ‘rational agent’ approach to ABM from within the economics discipline, (Gowdy, 2008) challenges the “narrowly rational, self-regarding responses to monetary incentives” typical of ‘rational agent’ research, of the ABM variety or not, and promotes the newly-emergent field of Behavioral Economics as a means to better accommodate real human reasoning while not giving up an economics mindset. However, the examples he suggests are in game theory experimental settings. The focus is on individual agents, but solely on external-objective behavioral drivers. • Air Traffic Control: In a demonstration of how the BDI ABM framework might be applied to a complex application such as air traffic control, (Rao, 1995) explores the theoretical and practical challenges of configuring a BDI framework to deal with the complexity of calculating expected time of arriving aircraft, sequencing them according to optimality criteria, reassigning ETAs on that basis, monitoring conformance, etc. In particular, this application focuses on how formal logical reasoning mechanisms such as Decision Theory, and in particular Decision Trees, may be meshed with the BDI   23 framework. The focus is on individual agents, but solely on external-objective behavior drivers. • Memory-Mediated Rational and Biased Reasoning: The notion of belief abstraction, a quintessential human trait, has been explored (Heuvelink, 2008) in a process that involves a degree of ‘forgetting’ of aspects of fact-type beliefs as they are abstracted. The researchers argue that this models humans’ rational and ‘biased’ reasoning. Their focus is on individual-level agents but solely on external-objective behavior drivers. • Inductive Reasoning: Another strongly human reasoning characteristic, inductive reasoning as opposed to deductive reasoning, is explored (Arthur, 1994) as a series of limited, local deductive reasoning steps (that are amenable to formal logic techniques to make low-level decisions) where the method of parsing inductive thinking into a series of deductive steps are opportunities to adjust beliefs based on how well reality matches prior guesses using a ‘distance’ metric based on factual differences. Focus is on individual-level agents, but solely on external-objective behavior drivers. • Emissions Trading: In a typical ‘rational agent’ application, (Zhang, 2011) examines economic transaction costs associated with emissions trading, using an ABM model to simulate dynamic interactions between firms, and utilizes traditional economic metrics within a BDI-like framework. Focus is on firm-level agents and on external-objective behavior drivers. • Self-Defeating Systems: In a demonstration of typical behavior in many social situations (The ‘Bar Model’, well-known in ABM research circles), (Batten, 2007) examines how absence of information sharing among agents, especially when an ABM is designed as a   24 competitive ‘game’, can lead to outcomes that no Agent wants. Focus is on individual agents but solely on external-objective behavior drivers. • Mental Models. In a focus on modeling human reasoning as an exercise in group goal discovery, (Edmonds, 1998) explores how ‘communities of agents’ can arise when Agents are equipped with ‘flexible, open-ended’ analysis and decision-making processes as opposed to ‘reactive’ processes, and communicate robustly via an internal, inspectable ‘language’. His agents continuously select their strategies from a catalog of available strategies, and modify them based on a ‘fitness’ measure on factual aspects of outcomes. He mentions ‘emergence’ but does not come to any conclusions about this. Focus is on individual agents but solely on external-objective drivers. • Adaptation to Climate Change Effects. In a community-level planning project, (Lieske, 2015) describes a manual technology for assembling expert human inputs to a structured mechanism for constructing what is essentially a manual inference engine that could possibly be transformed into an automated ABM using an extended BDI framework. A similar initiative, (Richards, 2012) employed a manual technology for achieving much the same result as Lieske, employing ‘systems thinking’ and a Bayesian Networks tool in a structured series of interactions with experts in a broad range of disciplines. Though neither Lieske’s nor Richards’ technologies employ Agents per se, nor are they implemented as simulation systems, they are included here as possible guides to defining future ABM systems incorporating (extended) BDI frameworks. No focus on individual Agents, though internal-subjective behavior drivers are in evidence in the human participants.   25 • Regional Water Management. A policy-focused initiative focused on regional water management (S. Wolf, Furst, S., Mandel, A., Lass, W., Lincke, D., Pablo-Marti, F., Jaeger, C., 2013) describes Lagom regiO, an ABM designed to exploit Agent learning in a defined micro-economic setting, i.e. one with economics-based assumptions for making decisions at the household level (e.g. Deaton rule for household saving, Taylor rule for interest rate, etc.). Learning is via imitation and mutation. Focus on household-level agents and solely on external-objective behavior drivers. • Energy Demand. In another exploration of a policy-focused ABM initiative (Gaube, 2013) employs a BDI-like framework for an ABM designed to predict actual energy demand throughout Vienna, the capital of Austria, under assumptions regarding population growth (births, deaths, immigration and emigration based on census data), demographic shifts (family structure, age, ethnicity, moves out of the parental home, etc., again based on census data), and elasticity of demand for various environment elements (schools, bodies of water, parks, public transport, etc.). Agents represent households (approximately 770,000), and their analysis and decision-making mechanism is primarily one of making weighted assessments of inputs such as price, proximity to amenities, culturally-based movement patterns, etc. Once initialized, the model proceeds with constrained randomness of decision variables; it does not appear that the model would be capable of reflecting the effects of abrupt, significant, persistent change, but perhaps it could be modified to do so. Focus is on household-level agents, and solely on external-objective behavior drivers. • Technology Adoption: In an application exploring the Theory of Planned Behavior (TPB), (Rai, 2015) includes ‘attitude’ in an otherwise ‘rational agent’ approach to   26 decision making, mediated by the actions of neighbors, which when combined with price and economic payback time, trigger purchase decisions when thresholds are exceeded. Focus is on household agents, and mostly external-objective behavior drivers, with the possible exception of ‘attitude’. • Transportation: An ABM application (Wang, 2005) is of a networked, hierarchical traffic management system consisting of a variety of Agents performing dynamic central dispatch, routing, monitoring, actuating, etc., that utilizes stepwise protocols, algorithms on measurements on the traffic system, and knowledge bases of previous behaviors, for learning and decision making implemented in ‘behavior programming’. Wang suggests that CORBA and the Open Services Gateway Initiative would be the core of the required platform. Focus is on units of traffic, and solely on external-objective behavior drivers. • Land Use: An ABM initiative utilizing a ‘fitness’ measurement of goals to desires (Evans, 2004) models land-use decisions where households make decisions on portfolios of cells in a raster-based programming environment, and where these decisions are driven by spatially-based composition and assembly patterns. Focus is on household-level agents, and solely on external-objective behavior drivers. • Market Structure: A typical application of external-objective behavior drivers, (Heppenstall, 2006), models an actual retail petrol market based on poll-based behavior data to calibrate and validate the model. Focus on consumer-level agents, and solely on external-objective drivers. • Emissions Transaction Costs: An ABM application in which agents anticipate the decisions of others (Zhang, 2011) in the context of emissions trading utilizes a ‘rational   27 agent’ reasoning mechanism. Focus is at the firm level, and incorporates only external-objective-external drivers. • Energy Transition: A similar initiative (Brede, 2013) takes this to another level, with the search for Nash Equilibria driving the decision making mechanism in which Agents (firms, individuals, households) make decisions about carbon-based vs renewables-based energy acquisition, with differential capital, operating and purchase costs, interest rates and other economic factors. Competitive and collaborative scenarios are explored based on Nash equilibria. Focus is on individual-level agents, and solely on external-objective behavior drivers. • Adoption of Alternative Energy Sources. Exemplifying an ABM approach to a typical human conundrum, (Schwoon, 2006) describes a model to capture the main dependencies in the supply and demand markets for fuel-cell vehicles, which exhibit the classic chicken-and-egg dilemma, i.e. how to persuade purchasers to buy when support for them in the traffic grid is inadequate, or its counterpart, how to persuade taxpayers to build up the grid to welcome early adopters. Focus is at the consumer/producer agent level, and solely on external-objective behavior drivers. • Environmentally-Friendly Technologies: A similar initiative (Cantonon, 2008) explores more fully the considerable challenge of new technology diffusion into reluctant markets by considering non-market interventions such as subsidies, tax rebates and other incentives for purchasers, but also consumer-level responses such as the contagion effect. Prices, subsidies, tax and other objective-external behavior drivers form the decision-making mechanism. Focus is on consumer-level agents, and solely on external-objective drivers.   28  For further reading, (Jennings, 1998) has performed a similar but more extensive exercise than the above.  2.3.3 ‘Socially Dominant’ ABM Initiatives  Our definition of ‘socially dominant’ in this context is that Agents and the way they interact possess a mix of external-objective and internal-subjective behavior drivers in their analytical and decision-making mechanisms, with the internal-subjective drivers unconscious and dominant in by far the majority of situations, and the external-objective behavior drivers consciously enabled when there is sufficient motivation, time and skill available, not frequent in average daily life, and almost never in times of crisis. This category represents the ‘sweet spot’ for guidance that could help us answer our Research Question.  We found no ABM research initiatives suitable for inclusion in this category. The reader will understandably wish to suspend judgment until the meaning of ‘internal-subjective’ and its interpretation in our research context is made clear; see the ‘Overlooked So Far’ section below.  2.4 Assessment of the Work to Date: Enough to Fill the Gap?  Virtually all the instantiations of the Intelligent Agent paradigm reviewed incorporated external-objective analysis and decision mechanisms within a ‘bounded rationality’ framework. Only one instance (Rai, 2015) expressly included a recognizably internal-subjective behavior driver   29 (‘attitude’), and given the range of disciplines represented in the papers reviewed, we have no reason not to assume that this scarcity in our review sample is representative of the whole corpus of ABM research. This did not surprise us – recall our earlier observation that most researchers go where model validation and prediction verification are assured, and that tendency strongly favors models based primarily on external-objective drivers of human behavior.  Clarifying Note  The Beliefs, Desires and Intentions components of the much-adopted BDI framework may appear to qualify as internal-subjective behavior drivers, but the consensus use of these components in the material we reviewed is firmly rooted in external-objective experience. Despite the connotation of the terms, in virtually every instance, Beliefs actually refer to facts thought by an Agent to be true about the external world, Desires refer to intended adjustments in the external world around the Agent, and Intentions are actions in the external world for fulfilling those Desires. Since none of these, at least the current use of them, are associated with sociality and ‘irrational’ decision making, they are not representative of the kind of internal-subjective constructs we described in our Problem and Research Question descriptions above. To be fair, one initiative (Heuvelink, 2008) attempted memory abstraction and ‘forgetting’ as representing ‘biased reasoning’, a phenomenon that could be construed as internal-subjective in nature.  Also, though we did not call this out on a case-by-case basis, we note that the majority of ABMs reviewed defined ‘Agents’ at the household, firm, or institution level (consistent with modeling practice grounded in the Economics discipline), as opposed to the individual level that we   30 believe is necessary to adequately to answer our Research Question. Note that this does not negate the intended value of the research findings, but does give guidance to those who choose to examine issues involving internal-subjective behavior drivers, i.e. keep in mind that firms, institutions and the like definitely do not have such motivations, and even the household should be treated carefully in this regard.     31   The conclusion we derive from the above is that, though we have identified a strong suite of technologies and a rich array of conceptual approaches for doing ABM research in general, a significant gap remains between what exists and what is needed to construct ABM systems enabling policy makers to guide populations in pre-adapting to the abrupt, significant, persistent societal disruptions based on internal-subjective behavior drivers. That is, the current state of the ABM art sheds virtually no light on the first part of our Research Question, i.e. ‘Can a human behavioral model based on internal-subjective drivers be sufficiently specified, calibrated, validated and verified to reliably produce credible, normative predictions?’. (We could, alternatively, interpret the nearly complete absence of internal-subjective human behavior triggers in the research reviewed to mean that most researchers believe the answer to our Research Question would be ‘No’, and simply avoid research that would be a natural, and very fertile ground, follow-on if the answer were ‘Yes’.)     32 Chapter 3: Overlooked: Internal-subjective Human Behavior Drivers  As is evident from the above, there is very little evidence of internal-subjective behavior drivers being represented, even if not the focus, in all the decades of ABM research, regardless of discipline. It therefore falls to us to specify what we mean by ‘internal-subjective’, and we do so in what follows.  3.1 Human Sociality: In a Major Societal Disruption, Emotion Dominates Thought  Though Sociologists remain engaged in a vigorous debate about what best accounts for human sociality, we have chosen key elements of Relationship Theory (Fiske, 1992) to both highlight the deficiencies of ABM work done to date in respect of our needs, and guide the development of an alternative instantiation of the Intelligent Agent paradigm to meet those needs. Other theories, such as Resource Theory (Foa, 2012) and Communal vs Exchange Theory (Clark, 2011) have their strong points and should be consulted by anyone wishing to inform their research efforts with tenable approaches to incorporating human sociality in their ABM initiatives.  Before discussing Fiske, consider your personal experience of the following human behaviors, and contrast the reality of these with the abstractness of those exhibited by virtually all of the ABM initiatives reviewed above. Further, consider how much difference these human behavior drivers would make to human decisions under a major societal disruption, compared to the ‘utility-maximizing’ drivers in Exclusive-rational ABMs and even the drivers in ‘Bounded-Rational’ ABMs. We expect that some of these would have been at the core of any ‘Social-  33 Dominant’ ABM initiatives in the previous section, if there had been any to find. Further, imagine how a policy analyst or planner, charged with the responsibility to pre-adapting whole populations to the abrupt, significant, persistent societal disruptions expected in our future, would value a credible instantiation of the Intelligent Agent paradigm incorporating the most salient of these. (Note: these are non-specialist terms, meant only to jog your thoughts, not define terms.)  Altruism/selfishness, affinity/misanthropy, volunteering/monetizing, charity/cheating, sharing/hoarding, fairness/cheating, conforming/rebelling, censuring/tolerating, shaming/supporting, shunning/forgiving, boasting/honoring, (dis)respecting, (dis)engaging, other, eg jealousy, laziness, narcissism, belligerence, litigiousness, etc.   Now we turn to Fiske for specialist guidance. We can do no better than have this leading Sociologist speak for himself, here from an online essay The Inherent Sociality of Homo Sapiens (http://www.sscnet.ucla.edu/anthro/faculty/fiske/relmodov.htm). • “We can appropriately say that a social relationship exists when any person acts under the implicit assumption that they are interacting with reference to imputedly [sic] shared meanings.” • “Calculative, competitive models of "success" and "achievement" are no more natural and no more fundamental than cultural models of altruistic caring; all are socially defined and validated.” • “It is rare for social interaction to be primarily a means to extrinsic asocial ends; the only people who persistently organize their lives this way are sociopaths.”    34 • “The inherent sociability of Homo Sapiens must stem from the adaptive advantages to our ancestors of socially organized production, exchange, consumption, decision-making, moral judgment, and sanctioning. Our unique communicative abilities, complex technical capacities, and delayed maturation resulted in unique opportunities for kin selection and reciprocal altruism to generate ultra-social adaptations. These adaptations involve extraordinarily strong social motives, such that humans need to engage in relationships—and are strongly disposed to judge and sanction others.” • “[P]eople use just four fundamental models for organizing most aspects of sociality most of the time in all cultures (Fiske 1991a, 1992). These models are Communal Sharing, Authority Ranking, Equality Matching, and Market Pricing.”  We are not suggesting here that every ABM initiative should attempt to achieve the currently-impossible goal of mimicking the full length and breadth of the analysis and decision-making mechanisms of real humans in what are often complex situations. But we definitely are suggesting that leaving out completely an entire class of human behavior drivers that obviously inform some of the most important things we do is a hard position to defend, and, fortunately for us, an opportunity to make a strong contribution.  3.2 Human ‘Irrationality’: Important Features as Opposed to Flaws   The common assumption for the better part of the 20th century in disciplines concerned with human reasoning was that it was flawed by being ‘irrational’ some of the time, with the term ‘cognitive biases’ attached to the mechanism by which such ‘irrational’ behavior was produced.   35 The experimenters were very thorough in their exploration of the ways in which we were ‘flawed’ creatures, and in innumerable experiments could reliably demonstrate a large number of ‘cognitive biases’ where humans would act counter to the commonly-assumed ‘rational agent’ paradigm for correct human reasoning, i.e. individuals were routinely observed acting so as to not maximize the ‘utility’ of every situation to their personal interests. This was nearly unthinkable to the orthodoxy of the day, which is a puzzle, reflecting on Fiske’s injunction above that the orthodoxy seemed to be assuming that all humans were sociopaths. Which begs the question “Including themselves?”.  Probably the most prominent and accomplished researcher into ‘cognitive biases’ is Daniel Kahneman, a Nobel Laureate in Economics (and the author of perhaps thousands of research papers, numerous articles, and a popular book, Thinking Fast and Slow, the latter describing the phenomenon in layman’s terms), but the lineup of accomplished researchers in this area is extensive. Some 100 ‘cognitive biases’ have been firmly established by researchers in the field. Here, we put forward a representative list of such and refer readers to a Wikipedia page providing a near-definitive list that establishes the existence of the phenomena: https://en.wikipedia.org/wiki/List_of_cognitive_biases: • Confirmation Bias: the tendency to seek out only that information that supports one's preconceptions, and to discount information which does not. For example, fairly absorbing only one side of a political debate, or, failing to accept the evidence that one's job has become redundant. • Framing Effect: the tendency to react to how information is framed, beyond its factual content. For example, choosing no surgery when told it has a 10% failure rate, where   36 others opt for surgery if told it has a 90% success rate, or, opting not to make the effort to choose organ donation as part of driver's license renewal when the default is 'No'. • Anchoring Bias: the tendency to produce an estimate near a cue amount that may or may not have been intentionally offered. For example, producing a quote based on a manager's preferences, or, negotiating a house purchase price from the starting amount suggested by a real estate agent rather than an objective assessment of value. • Gambler’s Fallacy (also known as the Sunk Cost Bias): the failure to reset one's expectations based on one's current situation. For example, refusing to pay again to purchase a replacement for a lost ticket to a strongly desired entertainment, or, refusing to sell a long stock position in a rapidly falling market. • Representativeness Heuristic: the tendency to judge something as belonging to a class based on a few salient characteristics without accounting for base rates of those characteristics. For example, the belief that one will not become an alcoholic because one lacks some characteristic of an alcoholic stereotype, or, that one has a higher probability to win the lottery because one buys tickets from the same vendor as several big winners. • Halo Effect: the tendency to attribute unverified capabilities in a person based on a different but observed capability. For example, believing an Oscar-winning actor's assertions regarding the harvest of Atlantic seals, or, assuming that a tall, handsome man is intelligent and kind. • Hindsight Bias: the tendency to assess one's previous decisions as more efficacious than they were. For example, 'recalling' one's prediction that Vancouver would lose the 2011 Stanley Cup, or, 'remembering' the proximate cause of the 2008 Recession.   37 • Availability Heuristic: the tendency to estimate that what is easily remembered is more likely than that which is not. For example, estimating that a meeting on municipal planning will be boring because the last such meeting you attended (on a different topic) was so, or, not believing your Member of Parliament's promise to fight for women's equality because he didn't show up to your home bake sale fundraiser for him. • Bandwagon Effect: the tendency to do or believe what others do or believe. For example, voting for a political candidate because your father unfailingly voted for that party, or, not objecting to a bully's harassment because the rest of your peers don't.  It is not possible to mention all of the luminaries who led teams of researchers in this field, but here we cite those who in our view contributed most to evolving our understanding of the significance of their work to ours: (D. Kahneman, Slovic, P., Tversky, A., 1982), (D. Kahneman, Tversky, A., 1972), (G. Gigerenzer, 2006), (G. Gigerenzer, Goldstein, D.G., 1996), (Simon, 1991) (Mercier, 2011), (Ariely, 2008), (Gilovich, 1991), (Goldstein, 2001), (Kida, 2006), and (Lerher, 2009).  In the late 1980s, the obvious question was finally asked: Given the relentless logic of evolution, where virtually every weakness shared within a species is inexorably eradicated or the species goes extinct if the weakness if severe enough, how is it that some 100 ‘cognitive biases’, which appear to have been around for a really long time (long enough so that virtually every human exhibits them with high consistency), snuck through the evolutionary bottleneck apparently unnoticed by this winnowing force? Evolutionary Psychology (Cosmides, 1989) proposed an alternative interpretation of this phenomenon; that ’cognitive biases’ were more properly viewed   38 as ‘cognitive heuristics’, i.e. shortcuts in the human reasoning mechanism that enhanced survival. But what about the ‘errors’ that are reliably produced in modern experiments?  According to Evolutionary Psychology, such ‘errors’ were produced by difficulties in the unconscious part of the human reasoning mechanism trying to match modern reasoning challenges on the one hand, and on the other, finely calibrated ‘heuristics’ fully capable of dealing with long-gone challenges, but not so much the modern challenges. For example, how would a heuristic laid down say 400,000 years ago, developed to distinguish between a branch and a snake in the bush, deal with non-linear differential equations, or navigating a racecar at high speed, or making investment decisions in the derivatives market? Essentially the Evolutionary Psychologists’ explanation for other researchers’ discovered ‘flaws’ is that our built-in heuristics working perfectly, just not on the new kinds of problems presented today. If you are familiar with Confirmation Bias, the poster-boy of ‘cognitive biases’, you may agree that previous researchers had exhibited a strong version of this bias.  One of us (Conroy) is investigating whether it is possible to mitigate cognitive biases/heuristics when this is called for; a Wikipedia summary of relevant work to date is at https://en.wikipedia.org/wiki/Cognitive_bias_mitigation .  In this work, the decision has been taken to take cognitive biases/heuristics as an integral part of the human decision-making mechanism, i.e. they must be reflected in the design of the ABM at the heart of our work, as is, and that our instantiation of the Intelligent Agent paradigm should, against the instincts of the ‘rational agent’ specialists, be equipped with mechanisms that cause   39 our agents to make ‘irrational’ decisions. Of course, our heuristics must conform in their triggers and their effects to those observed in the body of extensive and still useful work dealing with ‘cognitive biases’.  3.3 Other Elements to Consider in Our ABM  3.3.1 Needs and Wants  EnergyWorld treats Agent Needs as logically correlated with their Values, Beliefs and Desires, and categorizes them in a manner similar to Maslow’s Hierarchy of Needs (see, for example, http://www.simplypsychology.org/maslow.html) consisting of Physiological, Safety, Social, Esteem and Self-Actualization. Maslow suggested that most humans allocate most of their resources to the first of these and progressively less to the others, in order. We have adapted this idea and made this progression more flexible, with Agent populations emphasizing Needs without regard to the order Maslow stipulated, with the exception of Physiological Needs, which are the most important for all members of the population.  However, in times of rapid disruption, we can expect the magnitude of the resources allocated to meeting these Needs categories to change, according to an Order Of Consideration reflecting each Agent’s priorities. The EnergyWorld model permits such re-allocation in an order of priority, so that some Agents may indeed scurry to the bottom of the pyramid to ensure bodily survival, but like some humans, other Agents may choose to maintain a philosophically   40 sophisticated view of their lives despite its disruptions, and maintain resources for meeting Needs at higher levels in the diagram for as long as possible (think Gandhi).  In EnergyWorld we have deemed Physiological Needs as the minimum required for bodily survival and are therefore non-adjustable, while all other Needs are adjustable in the order specified for each Agent.  3.3.2 ‘Free Rider’ Response  The ‘Free Rider’ phenomenon highlights one of the weaknesses of the Rational Agent view of human behavior, which posits that Agents maximize individual payoffs by choosing to not cooperate with the rest of a population in generating a benefit that is shared with all members of the population, including the Free Rider. Numerous researchers, such as (Heinrich, 2006), Bowles (2002) and Stigler (1974), demonstrate that cooperating members will willingly incur a cost to punish Free Riders, though this seems to require a stable, ubiquitous institutional framework to administer such punishment, and that Free Riders generally respond by attenuating their asocial behavior, presumably to avoid such punishment. This phenomenon clearly qualifies as an ‘irrational’ response from the perspective of the ‘rational agent’ instantiation of the Intelligent Agent paradigm, and we considered incorporating internal-subjective drivers to manifest both sides of this phenomenon.  However, we chose not to do so, as the Affinity Group notion in EnergyWorld can be interpreted as the outcome of previous responses to the Free Rider phenomenon, i.e. Free Riders have been   41 sufficiently chastened to qualify for membership in Affinity Groups, or have been excluded from Affinity Groups altogether. EnergyWorld does not incorporate any mechanism for adjusting Affinity Group membership during a simulation (except via the death of member Agents), so the model in effect assumes that, while one can argue that there is some notion of ‘altruistic punishment’ carryover from times prior to such disruptions, the urgency and chaos that attend them overwhelms any alterations made while in the disruption period.  3.3.3 ‘Bounce’  Our species is able to learn, with sufficient repetition, when our conscious assumptions produce outcomes that are deleterious to our well being. When the duration and rate of such outcomes exceeds some level, we recognize the dissonance and adjust our assumptions, though often reluctantly. The interesting aspect of this adjustment is that it is usually greater, and sometimes far greater, than the initial adjustment that would have prevented the long run of poor outcomes prior to the eventual adjustment.  We have adapted this phenomenon, which we call ‘Bounce’, from the Overshoot Effect developed for economic phenomena (https://en.wikipedia.org/wiki/Overshooting_model).  3.3.4 System 1 / System 2  Though not directly associated with the Intelligent Agent Paradigm, the ‘System 1 / System 2’ concept of human decision making, popularized in Thinking Fast, Thinking Slow (Kahneman,   42 2001) has some relevance to the work reported here. Essentially, System 1 is characterized as the unconscious, rapid and highly parallel toolkit of reasoning shortcuts, evolved over evolutionary time, that enhanced survival for our distant ancestors. By contrast, System 2 is characterized as the conscious, slow and largely serial mechanism that acts as a check or validation of System 1 decisions when there is time to do so. Brain scientists such as (Damasio, 2005) describe a similar dichotomy, with a similar provenance, i.e. the ‘ancient’ limbic and ‘modern’ cerebral systems. In very loose terms, human sociality and ‘irrational’ decision making are associated with System 1, and rational agent decision making is associated with System 2, though there is only suggestive evidence of this to date.  The ‘System 1 / System 2’ concept is related to our distinction between subjective/objective and internal/subjective behavior drivers in that System 2 is the likely center for ‘rational’ decision making, whereas System 2 is the likely center for ‘social’ decision making. Note that while both System 1 and System 2 are ‘internal’ experiences as opposed to ‘external’ behaviors, the individual is conscious of System 2 processes and can reasonably accurately report on them, whereas this is not true of System 1 processes, which resist introspection.  We do not provide here a comprehensive citation list to support this discussion as we do not draw on the body of research associated with System 1 / System 2.      43 3.3.5 Management Theory  This is another example of a (group of) concepts that have a relationship to our internal-subjective, external-objective concept. Management Theory abounds with strongly opposing approaches to how best to manage people, the successes and failures of which can indicate something about what motivates humans in various situations.  In particular, the group of concepts that includes Theory X, Theory Y, Culture Management Theory, and Systems Theory tells us that others have grappled with the internal-subjective, external-objective dichotomy, while not generally basing their concepts on anthropology (as we have done) or physiology (as neuroscientists have done).  In brief: • ‘Theory X’ focuses on the individual worker and the individual product in an objective, scientific, exterior, analytic fashion, and employs individual, tangible reward and punishment motivation. • ‘Theory Y’ focuses on the interiors of the individual worker, in particular on what makes employees (and leaders) happy, how they can find meaning in their work, how their jobs can provide value and purpose in their lives, and how the workplace can become a source of joyful engagement. • ‘Culture Management Theory’ asserts that all individuals exist in many various groups or collectives such as family, friends, colleagues, religious or political affiliations, tribes, states, nations, collective humanity, and that every organization has a specific culture, an   44 interior set of values and meanings, rules and roles that hold the group together from within. • ‘Systems Theory’ looks at the group from the outside, in an overall objective fashion and frames them as interwoven into networks of interrelated systems, structures, physical meshworks, etc., and maintains that the dominant reality is that every individual is set in networks and systems of mutually interdependent and interwoven processes.  In the work reported here, we reflect (but did not draw on) a combination of ‘Theory Y’ and ‘Culture Management Theory’, and we actuate the notions involved in the whole community as opposed to individual, separate organizations. This body of work demonstrates that the concepts of internal-subjective and external-objective have resonated with others.  Again, we do not provide here a comprehensive citation list to support this discussion as we do not draw on the body of research associated with Management Theory.  3.3.6 Behavior Modification  Largely discredited as an effective way to modify how an individual performs as a viable group member, Behavior Modification demonstrated the need to take into account the internal motivations of individuals if one’s goal is to avoid the results of compelling obedience to authority through punishment, such as social dissonance and even social disintegration.    45 3.3.7 Cognitive Behavior Therapy  Not related to Behavioral Modification, this discipline is gaining ground in the psychology world due to its success at helping the individual modify thinking patterns and thus decision making and external behaviors without the catastrophic outcomes observed in Behavioral Modification. However, the relatively high rate of recidivism in troubled individuals suggests that its effectiveness in altering fundamental patterns of thought is yet to be firmly established.  In our view, the at least partial success of this treatment protocol reinforces the potential efficacy and the need for more robust internal-subjective research and implementation.  3.3.8 Modeling with Missing or Uncertain Empirical Data  A study commissioned by the Club of Rome and funded by the Volkswagen Corporation resulted in a book published in 1972, The Limits to Growth, as well as several followup publications (Beyond the Limits in 1993, Limits to Growth: 30-Year Update in 2004, A Comparison of `The Limits to Growth` with Thirty Years of Reality in 2007, as well as several conferences on the subject.) This initiative adopted a methodology that inspired the approach to the work reported here. The inspiration is not so much the content (exploring the effect of global resource decline), but rather how the authors dealt with an insufficiency of empirical data that forced them to develop a wide range of scenarios, represented by parameter sets in their simulation model, as opposed to outright predictions on a narrow empirical base. As time passed from the initial publication, better and more complete datasets resulted in narrower sets of scenarios, and after   46 35-odd years, the ‘standard’ model parameter set was demonstrated to have ‘predicted’ reality rather well. This is what we hope for regarding the work reported here, i.e. that our scenarios, currently based largely on parameters with values not formally validated by previous empirical research, will become more representative of reality over time if sufficient empirical research produces more robust parameter values.  3.4 Summing Up: We Know What Is Needed  The instantiations of the Intelligent Agent paradigm employed in almost all the Agent-based work done to date, i.e. the ’Exclusive-Rational’ and ‘Bounded-Rational’, fall far short of what is needed by policy makers in guiding the pre-adaptation of populations to the abrupt, significant and persistent societal disruptions we expect in our future. What is needed is a robust version of the ‘Social-Dominant’ instantiation of the Intelligent Agent paradigm, incorporating behavior drivers behind human sociality and ‘irrational’ decision making.  We label our intended instantiation of the Intelligent Agent paradigm the ‘Social Agent’, and describe it in the next section.   47 Chapter 4: ‘EnergyWorld’, A Context for Policy Exploration  4.1 An Abstraction for Feasibility and Clarity  The population modeling and simulation initiatives we reviewed earlier focused on human interactions with their environments in relatively narrow domains, e.g. water, interest rates, land use, derivatives, etc.). We have chosen not to model individual aspects of daily life and instead account for all domains at once, since in the kind of crisis we anticipate will involve virtually all such domains. To accomplish this we elected to create an abstracted community of Agents in which each Agent’s activities in each simulation Tick are cast as an aggregate consumption of energy. In this abstraction, which we call EnergyWorld, there is one time-varying Energy Supply to the community to be allotted against the Needs-driven energy demands of Agents in that community, with shortfalls expressing themselves as unmet Needs. EnergyWorld individuals generate their next simulation Tick’s energy demands by first adjusting (or not) their internal-subjective behavior drivers based on the current simulation Tick’s allotment versus demand, particularly Cognitive Biases (which may influence and even delay such adjustment) but also their Values and the ‘opinions’ of the members of their Affinity Groups. This is as opposed to what a ‘Rational Agent’ might do in an immediate scarcity, such as calculating Nash equilibria, instead of consulting and even helping Affines and being consulted and helped in return. In this formulation, there is no substitute source of supply for satisfying demand, there are no hidden tradeoffs between demanded services and products, and there are no ‘externalities’ as assumed in economic modeling; what you see is you what get, without unicorns, butterflies, and market distortions.  It is obvious that, with primarily internal-subjective behavior drivers in play in crisis   48 situations, the manner in which individuals in a community make adjustments will not be obvious, and certainly quite different to what Exclusive-Rational behavior drivers would produce. Having said this, note that there will likely be sociopathic tendencies in some Agents, due to the probabilistic manner in which Values, Beliefs, Desires, Cognitive Biases, etc. are generated.  We have adopted this abstraction because it avoids the obfuscation of proliferating domains of Agent action, while focusing attention on the internal-subjective behavior drivers at the heart of all such Agent action, which we believe will provide unique value to policy makers.  4.2 The ‘Social Agent’: An Instantiation of the Intelligent Agent Paradigm  As we have seen in the Introduction, the Exclusively-Rational instantiation of the Intelligent Agent paradigm is inappropriate for supporting policy makers’ attempts to pre-adapt human populations to the abrupt, significant and persistent societal disruptions we expect in our future; this is due to the reliance on analysis and decision-making mechanisms foreign or unavailable to ordinary human beings. The Bounded-Rational instantiation is more appropriate than the Exclusively-Rational instantiation in this regard, in that the analysis and decision-making mechanisms granted to ABM agents are more human-scale; however, these mechanisms still represent only objective-external behavior drivers, exactly those drivers that will take a back seat to internal-subjective behavior drivers in the intense, challenging times which the expected disruptions will trigger.    49 We thus concluded that, of the three instantiations described in this work, only a Social-Dominant instantiation of the Intelligent Agent paradigm, featuring a suite of internal-subjective behavior drivers, has a chance of providing policy makers with what they need to pre-adapt populations to the societal disruptions we expect. We also concluded that our best strategy was to build on the core ABM properties (autonomy, memory, communication) and the Beliefs-Desires-Intentions framework, following (Rao, 1995), and provide our Social Agents with the following: • Sociality, following (Fiske, 1992) and others, in particular Values, Affinities, Sharing and Thought Leadership. • ‘Irrational’ behaviors, following (Cosmides, 1997), (D. Kahneman, Slovic, P., Tversky, A., 1982), (G. Gigerenzer, Goldstein, D.G., 1996) and others, in particular a selection of the most relevant Cognitive Biases (see Appendix B: Cognitive Bias Detail). • Needs, following Maslow and others (see Chapter 3 Section 3.3.1 Needs and Wants). • A ‘bounce’ mechanism, reflecting a suitably-reframed version of the Economics-based Overshoot Effect, in which humans tend to delay needed adjustments, then over-adjust when they finally realize that such adjustment is necessary.  On the other hand, we did not endow our Social Agents with certain features central to other ABM initiatives, including: • Extreme learning. Our Social Agents do not proactively scour their environment for data in pursuit of some optimal goal. Instead, our Social Agents react to events - but reluctantly, jerkily, selectively, probabilistically, and incrementally. That is, more like humans than machines.   50 • Extreme competitiveness. Our Social Agents do not act as if they are in direct competition with other Agents. Instead, they offer to share, to a certain extent, with members of their Affinity Groups even though their own well-being may deteriorate as they do so. That is, more like members of a typical community, than a gang of narcissists or sociopaths.  To more formally describe EnergyWorld, we elected to follow the ABM documentation specification (S. Wolf, Bouchaud, J.-P., Cecconi, F., Cincotti, S., Dawid, H., Gintis, H., van der Hoog, S., Jaeger, C. C., Kovalevksy, D. V., Mandel, A., Paroussos, L., 2013) that came out of the Dahlem 100 conference. This question-based narrative-style material, organized into topics below, and is followed by a conceptual schematic. • Rationale o What is the object under consideration? EnergyWorld is an abstraction of a typical community of humans where all activities and objects in the community are framed as consumption of a single, generalized form of energy representing all activities and objects in the Agents’ environment. o What is the intended usage of the model? The model is intended to explore how to incorporate into our Social Agents internal-subjective behavior drivers that guide agent’s responses to abrupt, significant, persistent societal disruptions, in EnergyWorld represented by abrupt, significant, persistent decreases in EnergySupply.   51 o Which issues can be investigated? The primary domain-related issue under investigation is how credibly our treatment of internal-subjective behavior drivers reflects human-like behavior. • Agents o What kind of agents are considered in the model? Our ABM expects a population of Agents demographically representative of the human population of an average Canadian community. o Is there a refined taxonomy of agents? The Power characteristic of our Agents is a strong differentiator among them, as it dictates the order in which Agents’ energy demands are satisfied from a (perhaps inadequate) EnergySupply. o In particular, are there agent groupings that are considered relevant? Each Agent is assigned, during simulation initiation or at its creation during a simulation, subsets of existing Agents we call Affines with which they have affinity, and who are a source of advice and support, particularly as they form their energy demands and deal with energy shortfalls. In the EnergyWorld context we have created ‘PersonalRelationship’, ‘CulturalBinding’ and ‘SharedPrinciple’ affinity groups for each Agent, with different (probabilistic) parameters by which affinity group members may provide support in the form of energy transfers. • Other Entities o What are the other entities that are time-evolving but not decision-making? None.    52 • Boundaries o What additional inputs are provided to the model at run time? The Energy Supply is considered in EnergyWorld to be an externality from the community’s point of view; it is specified at simulation initialization but not revealed to the Agents. o Which outside influences on the model are hence represented? Abrupt, significant, persistent energy supply disruptions to the community. • Relations o What kind of relationships structure the agents’ interactions? Three ways: (1) Agents’ Power levels dictate the order in which Agents have their Needs-driven energy demands fulfilled (or not) in each simulation period; (2) Agents discuss with their Affines their satisfaction with the energy allocated, adjust their satisfaction levels based on these discussions, and take their satisfaction values into account in making energy demands in the next simulation period; and, (3) Agents provide support to Affines whose energy shortfalls are threatening to exceed their CoreNeeds, in the form of ‘affordable’ energy transfers. o To which extent do these represent institutions? None. • Activities o What kinds of actions and interactions are the agents engaged in? Agents are created, age, and die, and while part of a simulation make Needs-driven demands in each simulation period on the total EnergySupply flowing into their community in that period. They discuss with their Affines their degree of Satisfaction   53 with their allocations and may transfer to, or receive from, their Affines an energy ‘gift’ to avoid hardship and death due to persistently unmet CoreNeeds. • Time, activity patterns and activation schemes. o What is the basic sequence of events in the model? Once a simulation initiation step has completed, each step in a simulation proceeds as follows: ! Agents independently decide on the 3-part demand (current needs, personal savings, community contribution) they will make on that period’s EnergySupply, based on their current Desire level. ! Agents, in decreasing order of Power (randomly within each discrete Power level), are allocated their energy demands until the available EnergySupply has been depleted. ! Agents determine their Satisfaction with their energy allocation as the numeric difference between demand and allocation. ! Agents share their satisfaction level with each of their Affines and receive replies from their Affines; each such message contributes to a simple rolling average by each Agent. This process repeats for all Agents until message count limits are exceeded. Given the many-to-many relationship between Agents and Affines, this in practice results in a consensus among Affines, if sufficient messages are exchanged. Thought Leader Affines carry more weight in these discussions. ! Agents adjust their next Needs-driven energy demand toward the level of the Satisfaction level they end up with after the above discussions, the actual match mediated by the strengths of each Agent’s CognitiveBiases, e.g. a DenialEffect   54 bias may delay completely any adjustment, whereas a MemoryEffect bias may skew the actual value. ! The above adjustment to energy demands will carry an Agent for some time, depending on the model’s parameters. However, if an Agent’s energy allocations fall short of Needs, a more complex process is engaged; see below for the item describing how Beliefs and Desires are adjusted. ! Once all Agents have gone through the information sharing stage described above, the population is adjusted in preparation for the next simulation period. This involves creating new Agents (based on the population’s overall birth rate, the age of female Agents, and their current pregnancy status), aging, and deaths; see below for details. ! A new simulation cycle is initiated, until the required number of simulation cycles has been completed, or all Agents have died. o Are activities by agents triggered by a central clock, or by actions by other agents? A central clock initiates all events. o What is the interpretation of one time unit in the model? One simulation unit is equivalent to one calendar year. • Interaction protocols and information flows. o What are the general properties of the protocols governing the actions between agents? Agents interact only with their Affines for advice and support, but due to the many-to-many relationship between Agents and Affines, this can result in a rather robust communication network.    55 o How is it determined which agents can interact with each other? Agents interact directly only with their Affines. o What kind of information is available to each agent? Other than their own state variables, the only information available to Agents is the satisfaction of their Affines with their energy allocations in the current period (and that information is discarded at the end of each simulation period). o If agents act within institutional frameworks like firms or markets, what are the main properties of these institutions? The Agents do not operate within an institutional framework. • Forecasting. o Are Agents in the model forward looking or purely backward looking? Purely backward looking, and only for the number of Ticks defined by a memory duration parameter. o If Agents are forward looking, what is the basic approach to modeling forecasting behavior? Not applicable. • Behavioral Assumptions and Decision Making. o Based on which general concepts is the decision making behavior of the different types of agents modeled? All Agents use the same decision making algorithm, the core of which is that Beliefs drive Desires which in turn drive Intentions to express Needs-driven energy demand; Beliefs, Desires and Needs (except for CoreNeeds) are adjusted, based on energy allocation shortfalls against energy demands.   56 o If the decision making of certain agents is influenced by their beliefs, how are the beliefs formed? In EnergyWorld, Beliefs drive Desires and Desires drive Intentions (to fulfill Needs) which in EnergyWorld are synonymous with the energy demands each Agent makes make against each period’s EnergySupply. Higher-level Beliefs drive higher-level Desires and higher-level Desires create higher-level Intentions (i.e. demands), so scaling Desires (within Beliefs) and Beliefs up or down results in higher or lower demands. Initial Beliefs, Desires and the Needs they stipulated during simulation initialization.  Beliefs are adjusted when adjustments to Desires force the situation, and Desires are adjusted in response to persistent failure for energy allocations to meet Needs. The specifics are conceptually straightforward but somewhat complex in their implementation. The essence is that persistence results in adjustment, though Cognitive Bias effects probabilistically delay that adjustment, and when adjustment is actually made, the adjustment is usually more, via the BounceEffect, than what would have been sufficient if the adjustment had been made at first detection of allocation shortfalls. • Learning. o Are decision rules of agents changed over time? The decision of what energy demands to make is based on Desire and Belief levels (plus CognitiveBias treatments), which are initially set during simulation initialization. During simulation, Desires and Beliefs are modified in response to   57 persistent energy allocation shortfalls versus demands. While this phenomenon could be termed ‘learning’, we prefer to frame it as ‘structured adaptation’. o If yes, what types of algorithms are used to do this? See the previous item. • Population Demography. o Can agents drop out of the population and new agents enter the population during a simulation run? Agents can be removed from the simulated population in three ways: (1) old age, when their age, incremented every simulation Tick, exceeds a MaximumAge; (2) accident, a probabilistic occurrence driven by a DeathRate; and (3) persistently unmet CoreNeeds (mediated by CognitiveBias and Bounce treatments). New Agents are added to the simulated population by a female Agent giving birth after being earlier impregnated at a rate reflecting a BirthRate, subject to age being between a MaturityAge and a MaximumPregnancyAge and a PregnancyDelayTime has lapsed since her last birth; newborn Agents acquire their mother’s DemographicIdentity, WorldviewFilter and ActivationProfile, except for Age and Gender. o If yes, how are exit and entry triggered? See above. • Levels of Randomness. o How do random events and random attributes affect the model? The EnergySupply in each simulation period is in effect a random event from the point of view of the Agents. Most of the decision-making parameters are Agent-specific and are probabilistically selected from a range.   58 • Description of Agents and Other Entities, action and interaction. o What, in detail, are the agents and other entities in the model? Each Agent’s specification includes a DemographicIdentity defining its factual makeup (Age, Gender, Pregnancy Status, Needs, Affinities, and Thought Leader Status), a WorldviewFilter defining its internal-subjective behavior drivers (Values, CognitiveBiases, BounceUp, BounceDown), and an ActivationProfile to translate their internal-subjective behavior drivers into actions (Beliefs, Desires, Power). Note that only the ActivationProfile can change during an Agent’s lifetime (except for Age and Pregnancy Status in DemographicIdentity). There are no other entities than Agents. o What agent/entity does what, and in what order? All Agents do the same things at each step in the simulation process, though Agents’ Power levels determine the order in which they have their energy demands fulfilled, possibly leaving some Agents’ energy demands partially or wholly unfulfilled. o For each kind of agent/entity, what are the state model variables and parameters, their types, dimensions, ranges of values, what they represent, units of measurement, how often they are updated, how they are initialized? There is a sizeable number and variety of state variables in the EnergyWorld ABM. Due to length, we have located these in an Appendix: EnergyWorld State Variables. o What information, and with whom, does each kind of agent exchange information for decision making? Agents provide directly to their Affines, and receive directly from Agents for whom the Agent is an Affine, one signed message (per messaging cycle within each   59 simulation Tick) containing the sending Agent’s satisfaction level with the most recent energy allocation; note that, due to Agents having multiple Affines and being Affines to multiple Agents, this can involve, in effect, a cascade of indirect communication. Also note that this process is repeated as many time as is allowed by a MaximumMessagesPerAgentPerTick. o When are state variables updated? State variables are updated as the final step in each simulation Tick. o How are state variables updated? This varies by state variable. Satisfaction is updated as a rolling average of Agent’s own Satisfaction and those of the Agent’s Affines; recall from the above that this is a repeated process in which an Agent’s Affines have themselves updated their Satisfaction in the same manner vis a vis other Agents for whom they are Affines. • Initialization. o How is the model initialized? Currently, the model initialization is by declaration statements of structured data within the EnergyWorld code body, as well as a suite of algorithms for generating Agents’ personae from these data and a pre-simulation history of energy demands, energy allocations and satisfaction levels for all Agents. No other modeling entities exist, so no other initialization is required. See details below. o What kind of input is needed? Base demographic data (age/gender % of total, birth and death rates) for an actual community, total number of Agents to be modeled, probability density functions for   60 all WorldviewFilter and ActivationProfile variable values, and parameters governing the simulation process (see Appendix: EnergyWorld State Variables for details. o How is the initial state obtained from the input? An Agent population is determined so that age and gender proportions in the final population match that of the external data provided, and is no less than the target population specified. Once the Agent population has been generated, sufficient female Agents are made pregnant to reflect a BirthRate in the first simulation periods, Affines are generated for all Agents, and WorldviewFilter and ActivationProfile variables (see Appendix: EnergyWorld State Variables) are initialized by sampling probability density functions describing their distribution over all age/gender combinations. o Are the initial values based on data or chosen arbitrarily? The only initial values based on external data are those derived from Statistics Canada demographic data, including: Gender ratios at each Age to generate an initial Agent Population comparable to that of Canada circa 2013; BirthRate data (unfortunately only a single value for the whole population is available, also for 2013) from which we derived the number of pregnant female Agents needed to reflect that BirthRate in the initial simulation Ticks; and, DeathRate data (similar to BirthRate data). o In the latter case, what guided the selection of arbitrary values? Values were arrived at through a combination of intuition, lay knowledge and impressions from the work of other researchers.    61 • Run-Time Input. o Does the model use input from external sources that drive the model? Not during simulations. See the previous item for use of external sources during model initialization. o Are there data files or other models that represent these external processes? Not applicable. o If so, what kind of data is needed to feed the model at runtime? Not applicable.  4.3 Functional Implementation  A concise way to understand the core of our functional implementation of Social Agents is in the EnergyWorld State Variables Appendix. See the description of the AgentData item, a list of lists that shows how an Agent’s persona is built up, including a DemographicIdentity, a WorldviewFilter and an ActivationProfile, with all their components (themselves lists of lists). We draw attention to the WorldviewFilter in particular; to Needs Met? Power Action Outcomes Affinity Group Members Intentions Desires Beliefs Values Core Needs Other Needs Consult Adjust No Environment Biases, Bounces Yes  = Permanent Element Figure 1: EnergyWorld Process Flow   62 emphasize that we adopted the Beliefs-Desires-Intentions (BDI) formulation described in the Standards, Guidelines, Framework, Taxonomies section of Chapter 2, and encapsulated it in a broader set of human behavior drivers. In our instantiation of the Intelligent Agent paradigm, Beliefs, Desires and Intentions are adjustable based on the level of satisfaction of Needs, but a number of key adjuncts, especially Values and Cognitive Biases reflect deep human behavior drivers and are not adjustable.  The diagram is a somewhat simplified depiction of EnergyWorld’s key processes, and reflects several of the OODA and OOD-D concepts and constructs described in the Standards, Guidelines, Framework, Taxonomies section of Chapter 2. Note that simulation initialization processes have not been included here. Also note that ‘Environment’ equates to ‘Energy Supply’.  A key concept we employed in EnergyWorld was the use of manually-specified Probability Density Functions (PDFs) as the basis for providing parameter values to each Agent. In total, we utilized 460 PDFs to specify the 23 Agent parameters used in EnergyWorld, for each of 10 age categories and 2 genders. Each PDF was specified by a series of integer values representing the frequency with which parameter values are likely to occur. Example:  Figure 2: Example PDF for Assigning Agent Parameter Values   63 4.4 Technical Implementation  We identified several requirements for the technologies to implement EnergyWorld: • Ability to handle simulations of Agent populations large enough to demonstrate human population behaviors without small-number granularity effects in simulations with durations in the tens of simulation cycles. • Ability to operate on our development/test platforms, i.e. simulations with relatively small Agent populations and attenuated ‘discussions’, as well as on exploration platforms capable of running larger simulations in terms of populations and durations. • Level of functional abstraction that translates into fewer lines of EnergyWorld code at the possible cost of longer simulation durations due to the amount of hidden overhead code. • Availability of robust code libraries to reduce the amount of original code we would have to produce, with the possible disadvantage of awkward coding to conform to the input/output characteristics of library functions. • The learning curve that we would need to master the technologies selected.  We first assessed the agent-focused ABM platforms in the Tools section, hoping to select one that would meet all the above criteria. We decided that, given our knowledge at the time, none of these platforms was suitable for our needs. The principal reasons for this decision: (1) It appeared to us that most of the tools did not have sufficient support to adequately complement the (modest and dated) programming skills of the programmer (Conroy). Of the remainder, Repast appeared to be the best choice from this point of view. (2) While there appeared to be no design limit to the platforms’ size-related characteristics, we noted that most examples of   64 platform usage we could locate were of very limited scenarios and concluded (perhaps erroneously) that the hidden overhead code in most of the platforms was significant. Again, Repast stood out as much more capable than the rest in this regard, with a high-performance version advertised as capable of very large scenario simulations. (3) The ability to operate on our development/test platforms, and then port all the required code, libraries and configuration specific to a suitable exploration platform appeared, for some of the platforms, to be burdensome to the point that it could compromise our progress. Again, Repast’s multiple offerings seemed to offer a reliable way to mitigate these difficulties, though we were unable to verify this. (4) The learning curve for our programmer (Conroy) appeared to us to be a barrier to progress, for all the ABM Tools we reviewed.  Having eliminated the set of agent-focused ABM platforms described in the Tools section of Chapter 2, we then turned our attention to Agent-unaware programming languages. Several programming languages were suggested by other researchers, some working with Agents. C++ was the choice of strong programmers due to its potential for higher performance, and Python was the choice of those with more modest programing skills, its ‘list’ variable type permitting very flexible Agent characterization, large number of libraries, and strong network of support. Both C++ and Python are supported at minimal cost and effort on a wide variety of computing platforms, including our development/test hardware platform (MacBook Air) and the WestGrid super-computer network that we expected to use for EnergyWorld explorations. The final decision came down to accommodating the modest and dated skills of our programmer (Conroy). Clearly, it had to be Python.    65 A complete listing of EnergyWorld’s code is included in the EnergyWorld Code Base Appendix. If you view it, you will grasp the modest and dated skills of our programmer (Conroy); the obvious preference for procedural coding techniques and the near-absence of object-oriented techniques are dead giveaways of an ideas guy, not a strong designer or programmer. Still, the job got done, and the 3,000-odd lines of code and comments are, we are proud to say, much less than we thought would be required to construct a platform for having a chance to answer our Research Question.      66 Chapter 5: Explorations with EnergyWorld  EnergyWorld is a software simulator that provides the context for exploring our Research Question, repeated here in its two parts: (1) Can a human behavioral model based on internal-subjective drivers be sufficiently specified, calibrated, validated and verified to reliably produce credible predictions? (2) If not, can such models still be useful to policy makers tasked with pre-adapting human populations to abrupt, significant, persistent disruptions, given that these will demand human behavioral responses based on internal-subjective behavior drivers?  We emphasize that EnergyWorld is intended as an exploratory platform, with the above Research Question as the primary focus of our explorations. It is not intended to be a platform for actual policy analysis, though we hope that others might see its possibilities in this regard and move this research toward meeting this goal.  5.1 Taking The Policy Maker Perspective  As noted elsewhere, we do not have sufficient empirical data to quantitatively validate EnergyWorld against real world research findings, nor to verify its outcomes against actual experience, and as a result we do not pursue the goal of prediction. Instead, we focus on how a policy maker might use the tool as a way to conduct ‘differential policy analysis’, i.e. operate EnergyWorld on different sets of internal-subjective behavior drivers - reflecting populations with distinct differences - in order to facilitate comparison of their outcomes. Our general idea   67 for how that might occur is that a policy maker would specify an Objective Function representing the means to judge different policies regarding community well-being, develop a suite of EnergyWorld parameter sets representing policy alternatives for moving the community toward a target state of preparedness to cope with the kinds of social disruptions we have described above, and then rank these according to how well their simulation outcomes satisfy the selected Objective Function. We make no claim that the degree to which outcomes match the policy-maker’s Objective Function represents what is achievable in real life, only that their relative ranking is to some extent a meaningful input to policy analysis.  5.2 ‘Objective Function’ Candidates  In what follows, we describe a number of aspirational goals for community pre-adaptation to serious social disruptions that might suggest specific, quantifiable Objective Function candidates for use with EnergyWorld. We also demonstrate how each of these is so fraught with ambiguity and measurement difficulties as to make the search for a single, simple, all-purpose candidate a nearly hopeless task. A few examples to illustrate this point: • Community Sustainability. One might assume the desirability, in crafting a pre-adaptation strategy, of ensuring that the resultant community retains its essential character throughout the kinds of societal disruptions we have described. But how does one measure ‘essential character’? Is it the method of governance, total acreage of parks, percentage of high school graduates, community-level GDP, crime rate, cultural makeup, poverty rate, number of libraries, teen pregnancy rate, adequacy of snow removal   68 equipment, ratio of dogs to cats, etc.? And perhaps more important, would every community member want their community to stay the same as it is now? • Community Well-Being. Similar to Community Sustainability, but with value-tinged components such as suicide rate, divorce rate, mortgage default rate, number of well-attended dances per month, favorable/unfavorable votes in national polls, etc.? And again, how consistent must these measures be across the individual members of the community? In this category, at least one major community has made a serious attempt at an Objective Function – Bhutan, with its Gross Happiness Product companion to GDP as a measure of national well-being. But this measure, with its 72 measures, each with multiple components, is well beyond the modeling detail of EnergyWorld, so alas we can only look on enviously. • Community Improvement. Similar to both Community Sustainability and Community Well-Being, except this measure would focus on aspirational improvement goals, but as is obvious by now, which goals, how are these defined, how are they measured, and how are such improvements valued by the members of the community? • Negative Goals. Given the dire nature of the social disruptions we have described, we thought perhaps a more realistic approach to selecting an Objective Function would be to state negative goals, i.e. those outcomes that a policy maker would wish to avoid. For example, decimation or even complete annihilation of the community, or gross inequities among its members (wealth, power, status, needs fulfillment, connection), etc. Again, we found that specifying candidates in this category was bedeviled with ambiguity, measurement challenges and difficulties in breadth of applicability across the modeled community.   69 • Radically Re-Engineered Society. Though probably not feasible on a community scale in advance of serious social disruption, radical social engineering is mentioned here mostly to assure the reader that EnergyWorld is not sophisticated enough to help bring about an aspiration on the scale of Plato’s vision in which Philosopher Kings rule, or a society resembling that portrayed in H.G. Well’s The Time Machine, or that depicted in the recent movie Hunger Games. Many other utopian and dystopian visions have been contemplated over the ages, and some were even attempted.  We concluded that it is simply not possible to craft a single measureable quantity adequate to serve as the Objective Function for a situation as complex as pre-adapting a whole community to the social disruptions we postulate. As result, we chose to use ‘data-driven intuition’ as our guide to interpreting the outcomes of EnergyWorld simulations. While this opens us to the criticism of personal values, preferences and biases, we suggest that policy makers operating in the messiness of the real world (as opposed to highly-controlled lab situations) would probably take this stance when confronted with the uncertainties and impacts of the kinds of social disruption we describe in this world.  5.3 Energy Supply Disruption Candidates  We considered several energy supply disruption profile candidates for EnergyWorld. Happily, this task was much simpler, and we settled on two main energy supply disruption candidates: • ‘Gradual Decline’, in which successive modest declines in EnergySupply are presented to our Agent community, and various parameter setting scenarios are explored to   70 determine what settings are needed to prevent a given ‘Gradual Decline’ profile from killing off the Agent community due to Agent-level failure to respond with sufficient Needs adjustments. • ‘Crash’, in which a single step down in EnergySupply lasts for the duration of the EnergyWorld simulation period, and again various parameter setting scenarios are explored to determine the settings needed to prevent a specific ‘Crash’ profile from killing off the Agent community. • ‘Decline and Recovery’ in which either ‘Gradual Decline’ or ‘Crash’ profiles are reversed at some point in the simulation. While interesting, we decided that this kind of EnergySupply disruption profile was out of scope for the work reported here. • Variations on the ‘Gradual Decline’ and ‘Crash’ profiles above.  5.4 Sensitivity Analysis  EnergyWorld is not a large computer program by simulation standards (~3,000 Python code statements), but the work it may be asked to do can require arbitrarily large amounts of computing power, due largely to several O^n processes at its heart, particularly those related to population size and Agent interaction ‘intensity’. While we had access to the WestGrid network of ‘supercomputers’ for dealing with the most demanding scenarios we intended to explore, we decided to determine if we could avoid accessing WestGrid, since our desire was to have EnergyWorld produce results on computing platforms that policy makers would likely have access to. To this end, we ran a suite of simulations on our development/test platform, a MacBook Air (1.3GHz, 8GB memory, 256GB ‘disk’) to determine how sensitive EnergyWorld   71 run times was to key order-of-magnitude parameters of the model. We focused on the sensitivity of Final Population as a % of Starting Population against each of Starting Population, Message Intensity and Adjustment Acceleration over a representative range of other parameters, especially those associated with Beliefs/Desires/Needs/Sociality/Bias, as well as different Energy Supply profiles. See Appendix E: EnergyWorld Sensitivity Analysis for details.  This sensitivity analysis involved 108 simulation runs, reflecting: • 3 Starting Population levels (200, 600, 1,000), • 3 Message Intensity levels (2, 5, 10), • 2 ‘PDF Shape’ options (‘flat’ = Values/Beliefs/Desires/Needs/Sociality/Bias parameter values for all Age/Gender combinations taken from a flat probability density function over each domain, and ‘shaped’ = non-flat probability density function for each of Values/Beliefs/Desires/Needs/Sociality/Bias values for all Age/Gender combinations), • 3 Beliefs/Desires/Needs Adjustment Acceleration levels, and • 2 Energy Supply profiles (‘Gradual Descent’, ‘Crash’). Each simulation ran for 25 Ticks.  For the totality of our Sensitivity Analysis Cases, our findings indicated low correlations between Final Population as % of Starting Population against Starting Population, and extremely low correlations between Final Population as % of Starting Population against both Message Intensity and Adjustment Acceleration. Since both Energy Supply and PDF Shape are non-numeric, preventing a numerical correlation analysis of Final Population against these scenario options, we performed additional correlation analyses on the 4 combinations of these options   72 (each with 27 combinations contributing the numerical correlation calculation) and found low to very low correlations for all parameter combinations involving ‘shaped’ PDFs, and modest correlations for all parameter combinations involving ‘flat’ PDFs. As a result of this analysis we have omitted ‘flat PDFs’ from subsequent simulations, and have selected a single Starting Population (over 900) for all Scenarios explored; simulations with this population size requires between 10 and 30 minutes to run on our MacBook Air development/test platform. See Appendix E: EnergyWorld Sensitivity Analysis for details.  5.5 Simulation Scenarios  We designed several EnergyWorld simulation Scenarios to help answer our Research Question. We expressly do not claim that our Scenarios are viable pre-adaptation strategies for coping with the kinds of social disruptions we postulate in this work; instead we chose Scenarios that we viewed as contributing to answering our Research Question. We of course were not unaware of the possibility that EnergyWorld might be viewed as a starting point for initiatives aimed at evaluating specific Scenarios for actual policy analysis, and therefore attempted to expose within the code the manner in which we explicitly endowed our Agents with internal-subjective decision mechanisms demonstrated elsewhere to be powerful drivers of human behavior. See the Discussion section for more detail.  The simulations described below should be viewed as beginning with a population whose internal-subjective behavior drivers have been shaped through policy initiatives undertaken in a previous period, the actual simulation period starting with abrupt, significant and persistent   73 disruption to the community’s Energy Supply, often followed by a succession of similar events. The outcomes of the simulations therefore should be viewed as giving an indication of whether the pre-adaptation policies (might) have the intended effect should the anticipated disruptions actually occur.  5.5.1 Extreme Rational Agent vs Extreme Social Agent  We have elsewhere characterized the Rational Agent instantiation of the Intelligent Agent paradigm as inadequate to the task of modeling real human behavior, particularly in the social chaos and personal anxieties brought on by rapid, significant, persistent disruptions. With this claim being central to our research, it seemed reasonable that our first exploration with EnergyWorld should be to observe how differences between Rational Agent and Social Agent EnergyWorld populations would affect simulation progression and eventual outcomes.  The method for exploring this scenario was to run pairs of simulations for a variety of parameters sets, in which one of the pair simulated a population made up entirely of Extreme Rational Agents and the other of the pair simulated a population made up entirely of Extreme Social Agents. The Cases we ran are listed in the Excel worksheet imaged below; these reflect the following adjustments to EnergyWorld parameter sets and Agent interactions: • Parameter adjustments to manifest a population of Extreme Social Agents: o Values, Beliefs, Desires parameters set to high values to drive higher sociality.   74 o Bias parameters set to high values to drive more Denial, Risk Avoidance, Belongingness, Thought Leadership, and Memory Dependence and thus increase ‘irrational’, though socially cohesive, decision making by individual Agents. o Attachment, Actualization and Transcendence Needs parameters set to high values, and Safety and Esteem Needs parameters set to low values, to represent more socially sophisticated behaviors. o Message Intensity parameter set to a high value to drive more discussion with Affines, i.e. those in an Agent’s Affinity Groups. • Parameter adjustments to manifest a population of Extreme Rational Agents: o Values, Beliefs, Desires parameters set to low values to drive lower sociality. o Bias parameters set to low values to drive less Denial, Risk Avoidance, Belongingness, Thought Leadership, and Memory Dependence and thus increase ‘rational’, though less socially cohesive, decision making by these Agents. o Attachment, Actualization Needs parameters set to low values, and Safety and Esteem Needs parameters set to high values, to drive more self-interest behaviors. o Prevent Transfers of energy to and from Affines and from Community Storage, i.e. limit Agents to only Personal Usage energy allocations from the community-wide energy supply, complemented only by their Personal Savings.  For both Extreme Social and Extreme Rational simulations, all Power and ‘Bounce’ parameter values were obtained from the Baseline ‘Shaped’ Probability Density Functions for all Age/Gender combinations.    75 Based on the our Sensitivity Analysis (see above), we decided to use one Starting Population for all Cases, a number not so low as to risk small-population granularity and not so high that we would lose the very low correlations between Starting and Final Populations. A Suggested Population of 800 was selected, resulting in a Starting Population of 915 (the number needed to accurately mimic the Population Specification provided to EnergyWorld).  Reminder: This simulation Scenario is not presented as an argument to create either an Extreme Rational or Extreme Social community. Its purpose was to help us determine whether the internal-subjective approach to modeling human behavior does useful work, and thus help us address our Research Question.   76 Figure 3: Social/Rational Simulation Scenario Cases EnergyWorld	Simulations	-	SocialRationalExplorationCode	Versions	2	(FullSocial)	and	3	(FullRational)25	TicksEnergy Adjustment	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor Intensity Population Population PDFs After	20	Ticks* After	20	Ticks1 GradualDecline 1.0 2 800																									 915																									 FullSocial 208																									 23%2 GradualDecline 1.0 0 800																									 915																									 FullRational 353																									 39%3 GradualDecline 1.0 5 800																									 915																									 FullSocial 4																													 0%4 GradualDecline 1.0 N/A 800																									 915																									 FullRational 353																									 39% *	5 GradualDecline 1.0 10 800																									 915																									 FullSocial 15																											 2%6 GradualDecline 1.0 N/A 800																									 915																									 FullRational 353																									 39% *7 GradualDecline 3.0 2 800																									 915																									 FullSocial 63																											 7%8 GradualDecline 3.0 0 800																									 915																									 FullRational 345																									 38%9 GradualDecline 3.0 5 800																									 915																									 FullSocial 209																									 23%10 GradualDecline 3.0 N/A 800																									 915																									 FullRational 345																									 38% *11 GradualDecline 3.0 10 800																									 915																									 FullSocial 196																									 21%12 GradualDecline 3.0 N/A 800																									 915																									 FullRational 345																									 38% *	13 GradualDecline 5.0 2 800																									 915																									 FullSocial 16																											 2%14 GradualDecline 5.0 0 800																									 915																									 FullRational 335																									 37%15 GradualDecline 5.0 5 800																									 915																									 FullSocial 17																											 2% 	16 GradualDecline 5.0 N/A 800																									 915																									 FullRational 335																									 37% *17 GradualDecline 5.0 10 800																									 915																									 FullSocial 26																											 3%18 GradualDecline 5.0 N/A 800																									 915																									 FullRational 335																									 37% *19 Crash 1.0 2 800																									 915																									 FullSocial 1																													 0%20 Crash 1.0 0 800																									 915																									 FullRational 259																									 28%21 Crash 1.0 5 800																									 915																									 FullSocial 1																													 0%22 Crash 1.0 N/A 800																									 915																									 FullRational 259																									 28% *23 Crash 1.0 10 800																									 915																									 FullSocial 1																													 0%24 Crash 1.0 N/A 800																									 915																									 FullRational 259																									 28% *25 Crash 3.0 2 800																									 915																									 FullSocial 178																									 19%26 Crash 3.0 0 800																									 915																									 FullRational 265																									 29%27 Crash 3.0 5 800																									 915																									 FullSocial 6																													 1%28 Crash 3.0 N/A 800																									 915																									 FullRational 265																									 29% *29 Crash 3.0 10 800																									 915																									 FullSocial 196																									 21%30 Crash 3.0 N/A 800																									 915																									 FullRational 265																									 29% *31 Crash 5.0 2 800																									 915																									 FullSocial 2																													 0%32 Crash 5.0 0 800																									 915																									 FullRational 248																									 27%33 Crash 5.0 5 800																									 915																									 FullSocial 5																													 1%34 Crash 5.0 N/A 800																									 915																									 FullRational 248																									 27% *35 Crash 5.0 10 800																									 915																									 FullSocial 106																									 12%36 Crash 5.0 N/A 800																									 915																									 FullRational 248																									 27% *Notes:'Rational'	=	PDFs	skewed	to	very	low	sociality*	=	FullRational	Tick20	Populations	are	the	same	for	all	MessageIntensities	with	otherwise	same	characterizations'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Social'	=	PDFs	skewed	to	high	sociality'Message	Intensity'	=	NumMessages	PerTickPerAgent  77 The following charts present a more detailed look at this scenario’s outcomes, and compare Run015 and Run016, chosen for their representativeness of the Extreme Social and Extreme Rational Agent cases modeled.  The above is only one comparison of Extreme Social and Extreme Rational Agent Cases. All the others demonstrate essentially the same characteristics, with curves incrementally shifted left or right, and bars incrementally taller or shorter, depending on the Case depicted.  Figure 4: Selected Charts from SocialRational Simulation Scenario    78  The simulation outcomes for this Scenario invite us to speculate: • Power matters, indicated by the relative size stability of the highest Power level (blue bars in the top charts showing Population and Energy Supply) and the extinction of Agents with lower Power levels, with the exception of the lowest Power level, which is somewhat maintained by Agent births. See the Discussion section for a more detailed treatment of Power. • Power matters much more in Extreme Rational populations, in that Agents that do not possess the highest Power level are made extinct rather quickly. This is due to the lack of fallback sources of energy for any Agent, since Extreme Rational Agents were not endowed with Affinity Groups. This is truly a dog-eat-dog Case, with all the dogs being sociopaths completely lacking in altruism. Not a society worth aiming for. • Extreme Rational populations react much more quickly to reduced energy supply than do Extreme Social populations, both by immediately adjusting their Needs and quickly dying when these are persistently not being met (rather than being supported by their Affines and Community Storage). Despite this rapid adjustment, at some point in every Case in this scenario Extreme Rational population size always falls below that of their paired Extreme Social population. This is due to the broader base of social support that the latter enjoy, in the form of access to Affine (and Thought Leader) ‘opinions’ on energy-reduced situations, access to the portions of Affines’ Personal Savings that they are willing to share due to membership in a needy Agent’s Affinity Group(s), and finally access to the managed portions of the overall Community’s Store of energy. This support helps to offset the propensities of Social Agents to be in denial about the seriousness of   79 their situations, their tendencies to make assessment errors of what adjustment to make when they finally decide to take action, their tendency to over-compensate (and thus make excessive demands on the Community store of energy), and other ‘irrational’ behaviors. • The above Deficiency/Transfer chart show the marked difference between how an Extreme Social population and its Extreme Rational counterpart cope with a significant reduction in energy supply. An Extreme Social population, by helping each other in times of need (the ‘forest’ of green bars), gives itself time to adjust its Needs sufficiently to cope with reduced energy supply, whereas an Extreme Rational population has no such mechanism; indeed, the higher-Power Agents in such a population essentially cannibalize the rest of the population in order to survive. Can you say 1%? • Note that Deaths in Extreme Social Cases are much delayed compared to Deaths in Extreme Rational Cases (due to Affine and Community Storage energy contributions to needy Agents), but when Affine and Community Storage capacities wane due to contribution-driven depletion, there is always a sudden, significant number of Deaths. After this first ‘partial die-off’, Deaths in Extreme Social Cases resemble more closely those in Extreme Rational Cases. • The above Death outcomes are made more transparent in the Deficiencies/Transfers chart above, where both energy Deficiencies and the Transfers to moderate them occur early and often in Extreme Social Cases, and much less so in Extreme Rational Cases, where Agents are drawing down their Personal Storage of energy until, for some, this supplemental source is fully depleted, upon which they ‘die’.   80 • One can safely conclude that, even with EnergyWorld’s uncertain conformance to reality, a suite of policies aimed at pre-adapting a community for major societal disruption by creating a community with universal Extreme Rational tendencies is a poorer choice than a suite aimed at creating a population with Extreme Social tendencies.  We have other data supplementing that shown, including differentiation of total Births, Deaths, Deficiencies and Transfers by Agent Power, as well as complete Agent descriptions at periodic simulation intervals for a random sample of Agents. We have chosen not to display all this information in the body of this document (the sample Agent data alone takes up 60 pages for each Case), though we occasionally refer to it in support of our higher-level observations. Instead, we have made all our research materials (EnergyWorld code based – all versions, all Scenario Case listings, and all simulation outputs) available in a Google Drive folder with universal access; see Appendix A: Research Materials Database.  5.5.2 50/50 Mix of Extreme Social and Extreme Rational Agents  Given the heterogeneous nature of real human communities, and especially given the strength of the Confirmation Bias (the tendency for humans to seek out and take in information that agrees with their predispositions, and to avoid and reject information that goes against these predispositions), we concluded that, while a useful demonstration of our model’s different responses to different internal-subjective behavior drivers, the previous scenario (Extreme Social vs Extreme Rational) did not represent any realistic suite of policies that might be developed for pre-adaptation to the kind of social disruptions we postulate. On the other hand, we concluded   81 that it might be useful to run a suite of simulations involving mixtures of Extreme Social and Extreme Rational Agents in the same population; while not an ideal society due to its high degree of polarization, it has the ‘benefit’ of pre-adaptation probably being more achievable than the previous scenario, if policy makers deem this a good outcome.    The simulation Cases we ran to explore the characteristics of a suite of policies to achieve a mixed Extreme Social & Extreme Rational community employ the same parameter value assignments as for the previous scenario, with an equal number of Extreme Social and Extreme Rational Agents across all Age/Gender combinations being created at the beginning of each Figure 5: Selected Charts from Mix of Extreme Social/Rational Simulation Scenario   82 simulation. The charts above represent the outcome of a representative Case of the simulation runs for this scenario, and should be compared with those in the previous scenario.  As usual, these simulation outcomes invite speculation: • Comparing this set of charts to those of the previous scenario (Extreme Social vs Extreme Rational), it appears that the presence of Extreme Rational Agents alongside Extreme Social Agents enables the latter to survive longer due to the more rapid decrease in Extreme Rational Agents (caused by the latter’s individualistic tendencies). This was somewhat unexpected, and it was tempting for us to explore more mixes of Extreme Social and Extreme Rational Agents to determine the parameter set ‘boundary’ at which this population behavior might shift to other outcomes. However, having reminded ourselves that EnergyWorld is not (at least yet) a fine-tuned model that reflects reality at high fidelity, we decided to divert our energies to other scenarios in favor of more directly addressing our Research Question. See the Discussion section below. • Examination of sample data at the Agent level confirms that in this scenario, as with the others above, a significant majority of Agents surviving at simulation end are at the highest Power level. Combined with the above observation, we took this to mean that, while Power always matters, perhaps more aggressive levels of Affinity support and less precipitous energy supply declines may result in a broader Power distribution for surviving Agents; see the Affinity Proportional To Power Scenario below for a basic treatment of such a scenario. • Further examination of detailed sample Agent data revealed that Power was more a factor in Agent longevity than sociality, but not by as much as for the previous scenario, i.e.   83 Affine contributions made some difference, but only if an Agent and enough of its Affines survived long enough to contribute to each other. Like fast-track skaters, who have to survive the bulk of the race to be in position to literally push their final skater to the finish line, Agents must survive long enough to contribute to other Agents’ survival, even if it means the ‘death’ of the contributing Agents. • We also noted the apparent retreat, at the end of the simulation, of the Deficiency curve, which is suggestive that the Agents may have arrived at a sort of equilibrium with the Energy Supply profile. However, the amount of data was insufficient to make confident statements that this would indeed by reflected in simulation runs of longer duration.  As indicated earlier, we have made all our research materials (code, Cases, simulation outcomes) freely available in a Google Drive folder; see Appendix A: Research Materials Database.  5.5.3 Equilibrium Analysis  The above Scenarios were useful in establishing that EnergyWorld is capable of producing coherent behaviors based on the values of its parameters, which are proxies for values of internal-subjective behavior drivers. These scenarios, however, did not by themselves give us confidence that the model ‘settles down’ to a new equilibrium with a new Energy Supply. This scenario focuses on that aspect of the model’s parameter set tuning.  As our initial exploration of the equilibrium idea, we chose to run Cases from the scenarios depicted above, as well as the initial Sensitivity Analysis scenario, in simulations where an   84 energy supply factor applied to the first Tick’s supply was reduced from 1.0 at simulation start to, in successive simulations, to 0.9, 0.8, 0.7, 0.6 and 0.5 at Tick 1 and then held constant for the simulation’s duration. Each of these Cases thus represents a single ‘bump’ down in energy supply, to which the population attempts to adjust to through Needs reduction.  The Excel table image below describes the One Bump Recovery Cases for this scenario. The charts following that image depict EnergyWorld convergence behavior for these Cases (note that only the first 4 Cases in each of the above groups are shown):   85   Figure 6: One-Bump Recovery Simulation Scenario Cases     EnergyWorld	Simulations	-	Recovery	Exploration	CasesCode	Version See	'Scenario'	belowTicks 25MessageIntensity 5AdjustmentAcceleration 5.0SuggestedPopulation 800StartingPopulation 915	Energy	SupplyRun	# Scenario Factor	@	Tick	11 FullSocial	V2 0.92 FullSocial	V2 0.83 FullSocial	V2 0.74 FullSocial	V2 0.65 FullSocial	V2 0.56 FullRational	V3 0.97 FullRational	V3 0.88 FullRational	V3 0.79 FullRational	V3 0.610 FullRational	V3 0.511 50/50MixExtreme	V4 0.912 50/50MixExtreme	V4 0.813 50/50MixExtreme	V4 0.714 50/50MixExtreme	V4 0.615 50/50MixExtreme	V4 0.516 Sensitivity	V1 0.917 Sensitivity	V1 0.818 Sensitivity	V1 0.719 Sensitivity	V1 0.620 Sensitivity	V1 0.5  86  Figure 7: Selected Charts from One-Bump Recover Simulation Scenario   87  Figure 8: Further Selected Charts from One-Bump Recover Simulation Scenario     88     Again, we speculate: • The Extreme Rational Cases (Runs 6 through 9, first page right hand column) do not show ‘recovery’ behavior, though extending the simulation period may show slowing of population decline, and perhaps recovery as the Agents continue to adjust their Needs. In any case, the end population for these Cases is substantially lower than that of the others. • The Mix Of Extremes Cases (Runs 11 through 14, second page left hand column) show near-‘recovery’ behavior, but insufficient to establish a stable population. We suggest that this is a result of the strong presence of Extreme Rational Agents and the high mortality rate to which they eventually succumb. • Both the Extreme Social and Baseline Cases (Runs 1 through 4 for the former, Runs 16 through 19 for the latter) show definite ‘recovery’ behavior, though a close look at the Population scales of the associated charts show that the Baseline Case (i.e. a broad mix of sociality and bias parameter values) to be far superior in the sense that 50% more Agents survive by the time ‘recovery’ is complete. (Other measures than mere survival may cast a different light on this assessment, but such considerations are beyond the scope of the work reported here.) • In the Baseline Cases, ‘recovery’ occurs soonest when Message Intensity is low and Adjustment Acceleration is high. This reflects, in the case of Message Intensity, that too much ‘discussion’ with Affines tends to slow the decision to adjust Needs to a lower-  89 energy reality (and/or attenuate the adjustment amount), and in the case of Adjustment Acceleration, that larger adjustments are better than smaller adjustments. • We did not perform further ‘boundary’ testing, despite the theoretical attraction of discovering finer and finer tunings of the model, because EnergyWorld has not been proven to sustain that level of detailed investigation.  We again note that we have uploaded all our research materials to a Google Drive folder; see Appendix A: Research Materials Database.  5.5.4 Affinity Proportional To Power  We contemplated many other Scenarios for EnergyWorld simulation, but resisted the temptation to just blast away, produce an avalanche of charts and greatly expand our full research database. Instead, we have deliberately limited our explorations to the above Scenarios, plus the one described here. This Scenario explores two aspects of EnergyWorld that arose during the other Scenarios, namely how to ‘tone down’ the overwhelming influence that Agent Power has on who survives and who doesn’t, and bring more nuance to how Affine contributions help a population survive as a group rather than as lone individuals.  In this Scenario we arrange for Affinity and Power to complement each other by scaling both the propensity for an Affine to contribute to an Agent in need, as well as the level of such contributions, by the Power value of the Affine. Agents with high Rationality parameter settings will not benefit from this, but those with high Sociality will – especially those whose own low   90 Power is somewhat offset by Affines with high Power. The effect we envisaged for this arrangement was to break down the tendency for high-Power Agents to push low-Power Agents toward extinction; this appeared desirable on its own merits, but also because a pre-adaptation policy suite to achieve this arrangement would likely get less resistance and more uptake by a population not yet alarmed by future social upheaval. To effect this Scenario, we ran the following simulation Cases:  Figure 9: AffinityProportionalToPower Simulation Scenario Cases  Note that we chose ‘medium’ levels of both Message Intensity and Adjustment Acceleration to avoid too much dilution of Agents’ opinions of how well their Needs are being met, and to avoid EnergyWorld	Simulations	-	AffinityProportionalToPower	CasesCode	Version 5Ticks 25SuggestedPopulation 800StartingPopulation 915PDFs ShapedEnergySupply	Mode Crash	Energy	Supply Message AdjustmentRun	# Factor	@	Tick	1 Intensity Acceleration1 0.9 3 3.02 0.8 3 3.03 0.7 3 3.04 0.6 3 3.05 0.5 3 3.06 0.4 3 3.0Affinity	and	Powercomplement	each	other	in	that	both	the	Propensity	and	Level	of	an	Agent's	contriubtion	to	reducing	an	Affine's	PersonalUsageDeficiency	are	calculated	as	a	%	of	the	maximum	values	assigned	to	the	Agent	according	the	Agent's	Power.  91 excessive incremental Needs adjustments. All other parameters are those of the Baseline version of EnergyWorld.  The following charts depict the outcomes of these simulation runs:    Figure 10: Selected Charts from AffinityProporationalToPower Simulation Scenario   92 Again, we speculate on what these charts might tell us re the EnergyWorld model’s usefulness to a policy analyst designing a suite of policies to pre-adapt a population to the kinds of social disruptions we postulate: • All Cases demonstrate ‘recovery’ behavior, with some variation in when Energy Supply shortfalls start to ‘bite’ and when a new equilibrium of Agent Needs to available Energy Supply is achieved. We believe that the variation in when this happens is due largely to the randomness built into the model. • Though each Case establishes a new equilibrium at lower and lower levels as the Energy Supply disruption is increased, applying a simple heuristic (New Equilibrium Population divided by Starting Population divided by Energy Supply factor) suggests that the model’s tuning is such that the Final Population % of Starting Population is roughly inversely proportional to the ‘bump’ down in Energy Supply Factor. We suspect that if the pre-adaptation implied by these outcomes were actually achieved in the real world, the policy suite that produced it would be considered a success.  We again note that we have uploaded all our research materials to a freely-accessible Google Drive folder; see Appendix A: Research Materials Database.   93 Chapter 6: Discussion  In the previous Chapter we adopted the role of Policy Maker so as to both inform, and gain insight from, EnergyWorld simulations for crafting policies aimed at pre-adapting human populations to abrupt, significant, persistent social disruptions. In this Chapter, we take on the role of Critic, and challenge the degree of usefulness EnergyWorld has displayed in this regard, beyond some pretty graphs, a large corpus of EnergyWorld simulation output, and our potentially biased observations. Our goal in this Chapter is to separate wishful thinking about the degree to which EnergyWorld is modeling real human behavior, from reasonable expectations of what might be achievable with our approach, and thus help us answer our Research Question.  6.1 EnergyWorld Abstraction and Simplicity vs Veridicality  We deliberately created EnergyWorld with a high degree of abstraction, i.e. everything that would be in the environment of a real world population was proxied by a single entity, energy, which constituted the sole resource to be consumed by our Agents and the sole asset stored for future consumption. Our model did not attempt to explicitly model the multitude of transactions we experience in our every day lives, nor the political, legal, social and physical systems that mediate those transactions, with the exception of interactions between Agents and the members of their Affinity Groups - and these were limited to simple, one-to-one information exchanges and transfers of energy.    94 Another simplification we adopted in creating EnergyWorld is that virtually all of the model’s parameters were implemented as completely independent of each other. While this is not an issue for parameters defining the kind of simulation to run, it could be an issue for the parameters guiding the assignment of characteristics to Agents, especially those determining the operation of internal-subjective behavior drivers. In particular, we made no attempt to reflect the magnitude of the influence of these drivers relative to each other, nor how environmental conditions might alter these. Nor did we model any cross-influences between these drivers, such as the degree to which one influencer might excite or inhibit the expression of others. This is not a case of laziness on our part, but a constraint due to the fact that there is no empirical research to shed light on these matters. That is, despite a blizzard of research over decades establishing the existence and surprising power of internal-subjective behavior drivers, we could find no studies to guide our decisions in this regard; we also note that virtually all research into internal-subjective behavior drives has been performed in highly controlled lab conditions, a further barrier to creating a decision-making model that reflects the messiness and complexity of real life, especially when in crisis mode.  Yet another important kind of abstraction/simplicity reflected in EnergyWorld is in the specification of how to generate an Agent’s Affines and the propensity and level of support they might offer an Agent in need. While not grossly simplistic, this specification does not draw on research on the various kinds of networks in which humans participate in real life. This, unlike the above kind of abstraction/simplicity, is an area where we believe there is useful information for better guiding how EnergyWorld’s Agents join, interact with, and leave various kinds of   95 networks. We chose not to drive this level of guidance into our model due to the level of expertise required and the time involved in doing this job well.  The question is, then, did we create too abstract or too simple a modeling tool, or did we achieve a balance between abstraction and simplicity that allowed us to answer our Research Question?  Bottom line, we readily assert that EnergyWorld is at best a crude implementation of the much more sophisticated decision making mechanisms that humans routinely employ in real life. We also strongly believe that the level of abstraction and simplicity we employed, along with the other discussion points below, was a good enough balance for answering our Research Question.  6.2 Did We Inadvertently ‘Cook the Books’?  Humans are pattern-seeking beings and tend to see patterns that aren’t really there. We are also equipped with strong biases that prevent us from seeing patterns that are actually there but do not conform to what we expect or want to see. Further, those of us steeped in the Scientific Method often ignore ‘outliers’ in order to uphold a favored hypothesis, when it is sometimes the case that an outlier is a clue to a better theory. Overall, humans have to be very careful experimenters, observers, and interpreters for their efforts to be worthwhile.  In constructing EnergyWorld, and performing the simulation runs described in the previous Chapter, we were very aware of these tendencies and were alert to the possibility of biases built into EnergyWorld’s design and implementation, especially as regards the specification of high   96 and low values for each of its some 60 parameters, and whether or not the order of simulation processes would have a pernicious effect on simulation outcomes. In the end, we believe that there was very little opportunity for us to inadvertently skew the results of the simulations performed by EnergyWorld in this way. However, we were and are acutely aware that there is another way in which unintentional bias could have been introduced into our research: our interpretation of the results produced by EnergyWorld.  6.3 Interpretation Bias  Kaptchuk (2003) identifies several kinds of bias that can affect interpretation of research data (our comments regarding applicability to our research in square brackets): • Confirmation Bias, i.e. evaluating evidence that supports one's preconceptions differently from evidence that challenges these convictions. [We encountered several unexpected indications in EnergyWorld outcomes, notably the negative correlation between the intensity of Affine discussions about ‘Satisfaction’ values (we thought more discussion would result in more coordinated response across Agents and their Affines, but instead it produces more ‘dilute’ evaluations and thus delays in adjustment to real shortages), and separately the early onset and rapidity of population rapid die-off in highly Rational Agents (upon reflection, due to the lack of support from Affines when in need). We had no difficulty in assimilating these and recording such observations as valid, suggesting a low propensity to Confirmation Bias in our interpretation of EnergyWorld outcomes.]   97 • Rescue Bias, i.e. discounting data by finding selective faults in the experiment. [We discounted no outcomes. Also, there does not appear be any bias in our selection of Scenarios, since these were selected more for the ease of their implementation than their expected outcomes.] • Auxiliary Hypothesis Bias, i.e. introducing ad hoc modifications to imply that an unanticipated finding would have been otherwise had the experimental conditions been different. [No such modifications were made due to our fear of irreparably damaging EnergyWorld’s rather complex structure at the code level.] • Mechanism Bias, i.e. being less skeptical when underlying science furnishes credibility for the data. [We plead guilty to being convinced of the general notion of internal-subjective behavior drivers (though not their specific form, strengths or interactions). However, we do not have a vested interest in the creation or promotion of this mechanism, so we are confident that we are innocent of this bias’s effects.] • “Time will tell” Bias, i.e. the phenomenon that different scientists need different amounts of confirmatory evidence. [We do not suffer from this bias, though we expect others will be, especially those committed to the current formal methods of scientific research validity (see below).] • Orientation Bias, i.e. the possibility that the hypothesis itself introduces prejudices and errors and becomes a determinate of experimental outcomes. [This is a bias to which we readily admit; indeed, it forms part of our Research Question. Our peers in academe will judge whether we are biased to the extent that this work should be discounted.]    98 6.4 Formal Research Validity: Applicability and Limitations  Four categories of formal scientific research validity are described in (Wohlin & Wesslen, 2000), which we discuss here in order to identify ways in which our research effort may have been compromised.  Internal Validity. This type of experimental validity relates to how certain it is that the experimental treatment causes the effect observed, i.e. whether there are confounding factors that may create (or obscure) a cause-effect relationship. The authors identify several common ways in which this might happen (our comments in square brackets): • Fidelity, i.e. the degree to which the experimental setup is compatible with the real life environment in which the causes can occur and the effects are observed. [Our setup cannot be termed ‘experimental’; it is instead an ‘exploratory’ environment. Further, our Agent populations and the environment in which they operate are software constructs under the total control of a computer program that knows nothing of the real world. This aspect of Internal Validity is therefore irrelevant.] • Replicability, i.e. the degree to which an experiment can be replicated in terms of setup, operation and observation. [EnergyWorld is equipped with numerous constrained randomization steps intended to represent uncertainties in the modeled environment, so exact replication is not possible, but we expect EnergyWorld to produce, over a large enough number of replications of starting conditions, a constrained range of outcomes implied by the aggregation of these randomizations, sufficient to differentiate between different starting conditions, much like a Monte Carlo type experiment.]   99 • Observational Bias, i.e. the degree of bias brought to the setup, operation or observation of the experiment by the researcher. [See ‘Did We Cook the Books?’ above.] • Illusory Correlation, i.e. the likelihood of false positives or false negatives in the outcomes of the experiment. [See ‘Did We Cook the Books?’ above.] • Causal Error, i.e. the likelihood of creating erroneous, out-of-scope, over-constrained (e.g. sure to create or obviate certain effects), etc. treatments. [EnergyWorld actually exhibits considerable inertia, i.e. small perturbations don’t cause large outcome differences, so it would take a large error or manipulation to create a meaningfully distorted outcome.] • Effect Size, i.e. the degree to which small effects can be overwhelmed by large inputs or experimental error. [Despite our initial Sensitivity Analysis, which we claimed allowed us to run meaningful simulations with relatively small populations, this issue remains to be definitely put to rest. See the Conclusions Chapter.]  External Validity. This type of experimental validity relates primarily to how broadly the conclusions can be applied to other situations.  • Population Generalizability, i.e. the degree to which a cause-effect relationship, established for a given set of subjects from a given population, is applicable to other subject sets and/or populations. [We would expect different populations to produce different outcomes, but with a clear trail of incremental deviation from those of other populations pointing back to the differences in parameters assigned to EnergyWorld’s Agents.]   100 • Ecological Generalizability, i.e. the degree to which an experiment can be applied to a large range of environments other than that used for the experiment, perhaps with different components and interactions among them. [The primary ‘ecological’ element in EnergyWorld simulations is the energy supply over time. So long as one accepts this abstraction, this is not a source of validity deficiency.] • Researcher Bias, i.e. the degree to which the researcher inappropriately claims applicability to other situations, experimental or otherwise. [This is the kind of validity that we are most concerned about, as it is key to answering our Research Question; see ‘Credibility: A Better Measure’ below.] • Hawthorne or Novelty Effects, i.e. the tendency of experimental subjects to behave differently than they would in ‘normal’ situations. [Since our ‘subjects’ are artificially-created software Agents that have no life other than that in an EnergyWorld simulation, this is not a source of validity deficiency.] • Order Bias, i.e. the effect arising from the order of events in the subjects’ experience. [The order of the processes EnergyWorld’s Agents are subjected to may indeed be an issue; see ‘How Realistic is our Software Architecture’ below.]  Construct Validity. This kind of validity relates to ensuring that experimental treatments adequately represent causes, and that outcomes adequately represent effects. [EnergyWorld is both the ‘treatment’ and ‘outcome’ of our research, in that its implementation of internal-subjective behavior drivers is the object of scrutiny. We believe that this is not a source of validity deficiency, but would engage in a debate on this.]    101 Conclusion Validity. This kind of validity is framed as requiring a traditional statistical treatment supporting the claims made by researchers. [We do not have available, nor did we set to have available, a compelling statistical argument for answering our Research Question. However, a variant on this kind of validity must apply for us to make any reasonable claim regarding our Research Question. See “Credibility: A Better Measure” below.]   6.5 How Realistic is EnergyWorld’s Architecture?  EnergyWorld’s Agents are as a group subject to a sequential series of processes during each simulation Tick: they make claims in essentially a random order, energy is allotted to them against these claims in the order of their Power (randomly within Power groups), they discuss this allotment with their Affines, they assess their Satisfaction with this allotment, and finally they adjust their Beliefs/Desires/Needs according to their sociality and ‘irrational’ characteristics.  This strict order could easily be the cause of certain effects that make a difference, e.g. Power may have an outsized influence on simulation outcomes; some Agents could end up with the ‘short straw’ more often than other architectures might produce; ‘generous’ Affines might have their Personal Storage of energy significantly depleted, leading to their own difficulties sooner than with other architectures; etc.     102 6.6 Credibility: A Better Measure…  Even if we worked at a reasonable level of abstraction, did not inadvertently ‘Cook the Books’, and avoided most if not all of the sources of interpretation biases and formal validity described above, how credible will the conclusions of this research be, given that there is almost no empirical data against which to calibrate EnergyWorld’s parameters and assess the outcomes it produces? Note that we are not embarrassed by this lack of empirical support for our research, because as we have argued earlier, other theories and models are open to similar criticisms, plus the questionable reliance on highly controlled lab environments in other research setups, not to mention the dearth of experimental replication that is supposed to be a fundamental tenet of scientific research. Our belief is that we’re in pretty good company, even if our fingers are little stained from painting outside the lines from time to time.  Is there anything positive that can be said, then, about the credibility of our research beyond the declamations made above? Yes, there is, but our core argument for the value of our research to Policy Makers goes a toe-tip or two outside the orthodoxy of the Scientific Method.  In Aumann (2011), an argument is made for ‘credibility’, i.e. adequacy to a particular use, being a more salient way to assess the value of certain kinds of research, as opposed to the formal interpretation and validity criteria above, especially when there is limited empirical data to validate models and verify their outcomes (something that is pretty much baked into much sociological research). Included in the kinds of research to which this argument applies are social policy research initiatives aimed at achieving future outcomes that cannot in principle be   103 specified due to the complex influences of intervening events. Instead, the authors suggest, a high degree of correspondence between the psychological mechanisms implemented in simulation models and the mental models of informed stakeholders speaks to the credibility of such initiatives. While this is no guarantee that model outcomes are rock-solid predictions, it is a compelling argument for careful but optimistic use of the model.  We suggest that our research initiative qualifies for this treatment, because EnergyWorld, as a model of the psychological mechanisms in play by Agents reasoning about meeting their Needs from a declining source of the means to meet them, is easily matched to the mental models of Policy Analysts that would use such a model. This is essentially a ‘fitness for purpose’ argument, and while we make no claim to strong predictive power, we claim that EnergyWorld can be useful for ‘differential policy analysis’ by comparing simulation suites based on one parameterization of the model against those of others.   In summary, we present our research as being ‘Credible’, as a more meaningful measure than ‘Validated’ and ‘Verified’.     104 Chapter 7: Conclusions  In this Chapter we present our conclusions and brief references to the evidence supporting them.   7.1 Summary of Results  Our work has produced several tangible results, listed below without detailed explanation and without listing their implications; see the ‘Have We Answered Our Research Question’ section following this Summary for a consolidated argument for how these results support our answer.  • An Agent-Based Model based on Agents incorporating and expanding upon the characteristics stipulated in the Intelligent Agent paradigm, i.e. autonomy, the ability to sense their environment, make decisions about their individual goals, and communicate with other Agents. Where most other research has emphasized the Rational Agent instantiation of this paradigm, we have based our research on what we term the Social Agent instantiation of the paradigm, the better to model real life human population decision making and the behaviors that result; this applies especially to human sociality and ‘irrational’ behaviors. This model reflects an extension of the much-adopted Belief-Desires-Intentions framework, by specifying that Agents have a Demographic Identity, a Worldview Filter, and an Activation Profile that adds Values, Needs, Affines, Thought, Cognitive and Power to the usual characteristics of Agents in most ABMs.  The ABM we constructed is highly abstract in that there are only two kinds of objects   105 (Agents and an Energy Supply to the Agent population that varies over the simulation period) and only one composite short-term goal, i.e. meet Needs in the current simulation Tick, build up a store of Energy for future use, contribute to a Community Store of Energy to assist Agents in need, and adjust Beliefs, Desires and Needs in response to energy shortfalls in a manner driven by internal-subjective behavior drivers.  This model is highly parameterized: some 60 parameters are utilized in specifying how Agent characteristics are determined at the beginning of each simulation, how Agents make decisions, and how they adjust their Needs in response to unmet Needs. This degree of parameterization reflects a general lack of empirical data that would otherwise determine some Agent characteristics, but provides a rich option space for simulations.  • An implementation of this model (EnergyWorld) using the Python software language, capable of running simulations on a MacBook Air laptop (single 1.3 GHz processor 8GB memory and 256GB storage) in a reasonable timeframe (10-30 minutes for a simulation with 25 Ticks, for an Agent population of over 900). The Age/Gender specification of EnergyWorld’s Agent population reflects the demographics of Canada circa 2015.  • A suite of Scenarios, i.e. sets of simulations, each set aimed at highlighting ways that internal-subjective behavior drivers, implemented via sets of model parameter values, express themselves when abrupt, significant, persistent disruptions to the Energy Supply available to the EnergyWorld Agent population occur. The focus of these Scenarios is on exposing the relationship between different sets of parameter values (representing   106 different pre-adaptation ‘policies’) and the simulation outcomes they produce, i.e. reflecting support for ‘differential policy analysis’. Our simulations represent how an Agent population would behave after their internal-subjective behavior drivers, reflected in the values of their individual parameter sets, had been achieved in a previous period of adaptation.  • A database of simulation outcomes associated with the above Scenarios, and our interpretations of this data, which constitutes the evidence on which we base our Answer to our Research Question (see next).  • A corpus of material containing the totality of the above, in a Google Drive folder freely available to any who wish to access it; see the Research Materials Database Appendix.   7.2 Have We Answered Our Research Question?  The answer to our Research Question is in two parts, reflecting its two-part nature: (1) Can a human behavioral model based on internal-subjective drivers be sufficiently specified, calibrated, validated and verified to reliably produce credible predictions?  The short answer to this part of our Research Question, at least at the present time, is NO. We note that future research could well convert this to YES, but there are significant barriers to such research, including the difficulty of acquiring data on internal-subjective behavior drivers in even highly controlled lab conditions, let   107 alone in the real world, and the courage required of researchers to engage with initiatives that may not produce results conforming to current notions of scientific Validity and Verification. On the encouraging side, we note that the ‘Limits to Growth’ initiative of the Club of Rome started with much the same limitations on empirical data but evolved over time into a predictive model after numerous research initiatives closed the empirical data gap.  (2) If not, can such models still be useful to policy makers tasked with pre-adapting human populations to abrupt, significant, persistent disruptions, given that these will demand human behavioral responses based on internal-subjective behavior drivers?  The short answer to this part or our Research Question is YES, and the degree of utility can be increased. We defend this claim by appealing to the argument that Credibility, closely defined and commensurate with the current norms of experimental science, is a sufficient basis on which to judge the efficacy of research results based on inputs and outputs are not known, and perhaps cannot be known in advance of their occurrence. In addition, we note that practitioners using a model such as EnergyWorld can gain significant insight into the behaviors of ABM populations through ‘differential policy analysis’, i.e. the comparison of results of different mode parameterizations representing the results of alternative approaches to pre-adaptation of the population to rapid, significant, persistent social disruptions.    108 7.3 What Does This Imply?  We suggest that our answer to our Research Question implies that reflecting internal-subjective behavior drivers in models of human responses to abrupt, significant, persistent disruptions has positive utility to human population investigators.  This in turn implies that human behavior investigators could find fertile ground in crafting experimental initiatives to expand the body of empirical data on internal-subjective behavior drivers, especially their relative magnitudes and the ways in which they interact.  Our findings also imply that the Intelligent Agent community would likely find consumers for adaptations to this paradigm and its implementations that support Social Agent instantiations.  Finally, our findings also imply that the research community should consider alternatives to the restrictive notion of formal Validity and Verification, particularly Credibility, as closely defined (see Credibility: A Better Measure…).  7.4 Recommendations for Future Research  We propose several future research initiatives aimed at improving the Social Agent instantiation of the Intelligent Agent paradigm we have introduced with this work: • The Agent Power characteristic as specified in our model is too blunt and simplistic. As a result, high-Power Agents tend to force lower-Power Agents into near extinction. While   109 such an outcome should be permitted by the model, this way of achieving that result masks the more subtle ways in which such outcomes could arise, e.g. the effects of fortuitous Affinity Group assignments to Agents. We suggest a more randomized treatment, one that still favors high-Power Agents but not to the extent in EnergyWorld. • The Affinity Group treatment in our model is dominated by random assignment as opposed to ‘natural’ affinities such as family membership, shared neighborhoods, etc. In addition, our Affinity Groups are static, i.e. each Agent’s affinity relationships remain the same over the entire simulation period. A better way to model affinities can likely be achieved by drawing on the considerable body of Social Science research on how humans create, join, interact with, and leave networks. • Our model contains essentially no excitatory or inhibitory relationships between Values, Beliefs, Desires, Needs, and Cognitive Biases. More robust relationships would improve the veridicality of the Social Agent model. • EnergyWorld’s implementation forces a highly sequential treatment of Agent activity, i.e. each kind of treatment (make energy claim, receive energy allotment, discuss with Affines, decide on adjustments, etc.) is applied to all Agents before moving on the next treatment. A more robust implementation would have Agent treatments occur much more asynchronously, perhaps via some form of process randomization. • Rework the EnergyWorld code base to reflect more modern design and implementation standards; it currently displays the deficiencies of its author in this regard.    110 Bibliography Adolph, S. (2006). What lessons can the Agile community learn form a maverick fighter pilot? Proceedings of AGILE 2005 Conference, 1-6.  Ariely, D. (2008). Predictably Irrational: The Hidden Forces That Shape Our Decisions. New York, USA: Harper Collins. Arthur, W. B. (1994). Inductive Reasoning and Bounded Rationality. The American Review, 84(2), 406-411.  Aumann, C. (2011). Constructing model credibility in the context of policy appraisal. Environmental Modelling and Software, 26, 258-265.  Batten, D. (2007). Are some human ecosystems self-defeating? Environmental Modelling and Software, Elsevier, 22, 649-655.  Becker, J., Bergener P., Breuker, D., Rackers, M. (2010). Evaluating the Expressiveness of Domain Specific Modeling Languages Using the Bunge-Wand-Weber Ontology. Proceedings of the 43rd Hawaii Internation Conference on System Sciences - 2010, 1-10.  Bowles, S., Gintis, H. (2002). Homo reciprocans. Nature, 415, 126-128.  Brede, M., de Vries, B. J. M. (2013). The energy transition in a climate-constrained world: Regional vs. global optimization. Environmental Modeling and Software, 44, 44-61.  Brooks, R. (1991). Intelligence without representation. Artificial Intelligence, 47, 139-159.  Bunge, M. (1999). The Sociology-Philosophy Connection. Piscatawy, USA: Transaction Publishers. Bunge, M., Mahner, M. (1997). Foundations of Biophilosophy. Berlin, Germany: Springer. Cantonon, S., Silverberg, G. (2008). A percolation model of eco-innovation diffusion: the relationship between diffusion, learning economies and subsidies. Retrieved November 7, 2015  from Maastricht, The Netherlands: http://www.merit.unu.edu/publications/working-papers/?year_id=2008 Clark, M. S., Mills, J. R. (2011). Handbook of Theories of Social Pscyhology. Los Angeles, CA: Sage. Cohen, P. R., Levesque, H. J. (1990). Intention is Choice with Commitment. Artificial Intelligence, 42, 213-261.    111 Collins, A., Petty, D., Vernon-Bido, D., Sherfey, S., Vernon-Bido, D., Sherfey, S. (2015). A Call to Arms: Standards for Agent-Based Modeling and Simulation. Journal of Artificial Societies and Social Simulation, 18(3), 12-12.  Cosmides, L., Tooby, J. (1989). Evolutionary Psychology and the Generation of Culture, Part II. Ethiology and Sociobiology, 10, 51-97.  Cosmides, L., Tooby, J. (1997). Evolutionary Psychology: A Primer.   Retrieved November 8, 2015 from http://www.psych.ucsb.edu/research/cep/ d'Inverno, M., Luck, M., Georgeff, M., Kinny, D., Wooldridge, M. (2004). The dMARS Architecture: A Specification of the Distributed Multi-Agent Reasoning Systems. Autonomous Agents and Multi-Agent Systems, 9, 5-53.  Damasio, A. (2005). Descartes' Error: Emotion, Reason and the Human Brain. New York, NY: Putnam. Dautenhahn, K. (1998). The art of designing socially intelligent agents: science, fiction, and the human in the loop. Applied Artificial Intelligence, 12(7-8), 573-617.  Durfee, E. H., Kenny, P. G., Kluge, K. C. (1998). Integrated Permission Planning and Execution for Unmanned Ground Vehicles. Autonomous Robots, 5, 97-110.  Durfee, E. H., Lesser, V. R., Corkill, D. D. (1989). Trends in Cooperative Distributed Problem Solving. IEEE Transactions of Knoweldge and Data Engineering, 1(1), 63-83.  Edmonds, B. (1998). Modeling Socially Intelligent Agents. Applied Artificial Intelligence, 12, 677-699.  Evans, T. P., Kelley, H. (2004). Multi-scale analysis of a household level agent-based model of landcover change. Journal of Environmental Management, 72, 57-72.  Faltings, B. (2000). Intelligent Agents: Software Technology for the New Millennium. Informatik/Informatique, 1, 1-5. Finin, T., Labrou, Y., Mayfield, J. (1995). KQML as an agent communication language. In U. o. M. B. County (Ed.), Software Agents (pp. 22). Cambridge: MIT Press. Fiske, A. P. (1992). The Four Elementary Forms of Sociality: Framework for a Unified Theory of Social Relations. Psychology Review, 99(4), 689-723.  Foa, E. B., Foa, U. G. (2012). Resource Theory of Social Exchange. In A. K. K. Tornblom (Ed.), Handbook of Social Resource Theory: Theoretical Extensions, Empirical Insights, and Social Applications. New York: Springer.   112 Gaube, V., Remesch, A. (2013). Impact of urban planning on household's residential decisions: An agent based simultation model for Vienna. Environmental Modelling and Software, 45, 92-103.  Genesereth, M. R., Fikes, R. E. (1992). Knowledge Interchange Format, Version 3.0 Reference Manual. Retrieved March 2, 2016 from http://www.upv.es/sma/teoria/sma/kqml_kif/kif.pdf Gigerenzer, G. (2006). Bounded and Rational. In R. J. Stainton (Ed.), Contemporary Debates in Cognitive Science (pp. 115-133). Oxford, England: Blackwell Publishing. Gigerenzer, G., Goldstein, D.G. (1996). Reasoning the Fast and Frugal Way: Models of Bounded Rationality. [Reviewed]. Psychological Review, 103(4), 650-669.  Gilovich, T. (1991). How We Know What Isn't So: The Fallibility of Human Reason in Everyday Life. New York, NY: The Free Press. Goldstein, D., Gigerenzer, G., Hogarth, R.M. (2001). Group report: Why and when do simple heuristics work? In G. Gigerenzer & R. Selten (Eds.), Bounded Rationality: The Adapative Toolbox. Cambridge, MA: MIT Press. Gowdy, J. M. (2008). Behavioral economics and climate change policy. Journal of Economics Behavior and Organization, 68, 632-644.  Gruber, T. (1993). A Translation Approach to Portable Ontology Specifications Retrieved March 2, 2016 from http://tomgruber.org/writing/ontolingua-kaj-1993.pdf Hamilton, E., S., Guillaume, J. H. A., Jakeman, A. J., Pierce, S. A. (2015). Integrated assessment and modelling: Overview and synthesis of salient dimensions. Environmental Modelling and Software, 64, 215-229.  Hare M., D. P. (2004). Further towards a taxonomy of agent-based simulation models in environmental management. Mathematics and Computers in Simulations, 64, 25-40.  Heinrich, J. (2006). Cooperation, Punishment, and the Evolution of Human Institutions. Science, 312, 60-61.  Heppenstall, A., Evans, A., Birkin, M. (2006). Using Hybrid Agent-Based Systems to Model Spatially-Influenced Retail Markets. Journal of Artificial Societies and Social Simulation, 9, 1-20.  Heuvelink, A., Klein, M. C. A., Treur, J. (2008). An Agent Memory Model Enabling Rational and Biased Reasoning. Paper presented at the International Conference on Web Intelligence and Intelligent Agent Technology. Jennings, N. R., Sycara, K., Wooldridge, M. (1998). A Roadmap of Agent Research and Development. Autonomous Agents and Multi-Agent Systems, 1, 7-38.    113 Kahneman, D., Slovic, P., Tversky, A. (Ed.) (1982). Judgment Under Uncertainty: Heuristics and Biases. New York, NY: Cambridge University Press. Kahneman, D., Tversky, A. (1972). Subjective Probability: A Judgment of Representativeness. Cognitive Psychology, 3(3), 430-454.  Kaptchuk, T. J. (2003). Effect of interpretive bias on research evidence. British Medical Journal, 326(704), 1453-1455. doi:10.1136/bmj.326.7404.1453 Kida, T. (2006). Don't Believe Everything You Think: The 6 Basic Mistakes We Make in Thinking. New York, NY: Prometheus. Kiwelekar, A. W., Joshi, R. K. (2010). An Object-Oriented Metamodel for Bunge-Wand-Weber Ontology. Computer Science and Engineering. Indian Institute of Technology Bombay.   Lerher, J. (2009). How We Decide. New York, NY: Houghton Mifflin Harcourt. Lieske, D. J. (2015). Coping with climate change: The role of spatial decision support tools in facilitating community adaptation. Environmental Modelling and Software, 68, 98-109.  Mercier, H., Sperber, D. (2011). Why Do Humans Reason? Arguments for an Argumentative Theory. Behavioral and Brain Sciences, 34, 55-74.  Minsky, M. (1987). The Society of Mind. London: Heinemann. Monu, K. (2005). Intelligent Agents as a Modelling Paradigm. (Master of Science in Business Administration), University of British Columbia, Vancouver, British Columbia, Canada.    Muller, B., Bohn, F., Dressler, G., Groeneveld, J., Klassert, C., Martin, R., Schluter, M., Schulze, J., Weise, H., Schwarz, N. (2013). Describing human decisions in agent-based models - ODD+D, an extension of the ODD protocol. Environmental Modelling and Software, 48, 37-48.  Rai, V., Robinson, S. A. (2015). Agent-based modeling of energy technology adoption: Empirical integration of social, behavioral, economic and environmental factors. Environmental Modeling and Software, 70, 163-177.  Rao, A. S., Georgeff, M. P. (1995). BDI agents: From theory to practice. Proceedings of the First International Conference on Multi-Agent Systems, ICMAS'95, 312-319.  Richards, R., Sano, M., Roika, A., Carter, R. W. Bussey, M., Matthews, J., Smith, T. F. (2012). Bayesian belief modeling of climate change impacts for informing regional adaptation options. Environmental Modelling and Software, 44, 113 - 121.  Rosenschein, J. S., Zlotkin, G. (1994). Rules of Encounter: Designing Conventions for Automated Negotiation among Computer. Cambridge: The MIT Press.   114 Schwoon, M. (2006). Simulating the adoption of fuel cell vehicles. Journal of Evolutionary Economics, 16, 435-472.  Shafaei, S., Aghaee, N. G. (2008). Biological Network Simulation Using Holonic Management Systems. Computer Modeling and Simulation, 617-622.  Shoham, Y. (1993). Agent-oriented programming. Artificial Intelligence, 60(1), 51-92.  Simon, H. A. (1991). Bounded Rationality and Organizational Learning. [Reviewed]. Organization Science, 2(1), 125-134.  Stigler, G. J. (1974). Free Riders and Collective Action: An Appendix to Theories of Economic Regulation. The Bell Journal of Economics and Management Science, 5(2), 359-365.  Wand, Y., Weber, R. (2008). On the ontological expressiveness of information systems. Journal of Information Systems, 3, 217-237.  Wang, F. Y. (2005). Agent-Based Control for Networked Traffic Management Systems. Intelligent Systems IEEE, September/Octover 2005, 92-96.  Weizenbaum, J. (1966). ELIZA - A Computer Program for the Study of Natural Language Communication between Man and Machine. Computational Linguistics, 9(1), 36-45.  Wohlin, C., Runeson, P., Host, M.,  Ohlsson, M.S., Regnell, B., & Wesslen, A. (2000). Experimentation in Software Engineering: An Introduction. Norwell, MA: Kluwer Academic Publishers. Wolf, S., Bouchaud, J.-P., Cecconi, F., Cincotti, S., Dawid, H., Gintis, H., van der Hoog, S., Jaeger, C. C., Kovalevksy, D. V., Mandel, A., Paroussos, L. (2013). Describing economic agent-based models - Dahlem ABM documentation guidelines. Complexity Economics, 2, 63-74.  Wolf, S., Furst, S., Mandel, A., Lass, W., Lincke, D., Pablo-Marti, F., Jaeger, C. (2013). A multi-agent model of several economic regions. Environmental Modelling and Software, 44, 25-43.  Zhang, B., Zhang, Y., Bi, J. (2011). An adaptive agent-based modeling approach for analyzing the influence of transaction costs on emissions trading markets. Environmental Modelling and Software, 26, 482-491.     115 Appendices  Appendix A   Research Materials Database  All our research material has been uploaded to a freely accessible Google Drive folder, link https://drive.google.com/folderview?id=0B6gSrcrSNZo6NGFSeERzVjJWcTQ&usp=drive_web&ddrp=1#, and includes: • An ‘EnergyWorld Baseline’ folder containing an Overview .txt file that describes the implementation of both the Social Agent model and the EnergyWorld simulator, and a .txt file containing Version 1.0 EnergyWorld code. Both files are best viewed via a Python-aware text display application. • An ‘EnergyWorld Simulation Scenarios’ folder containing a collection of folders, one for each of the Scenarios described in Section 5.5 of this document, i.e. ‘ExtremeSocial vs ExtremeRational’, ‘Mixed Social/Rational’, ‘Equilibrium Analysis’, and ‘AffinityProportionalToPower’. In turn, each of these folders contains a .exe file listing the simulation runs performed for the Scenario described in terms of parameter settings, a .txt file containing the EnergyWorld code version(s) used for the Scenario (best viewed via a Python-aware text display application, a subfolder of simulation run Outputs in the form of .txt files, and a subfolder of simulation run Charts associated with selected Outputs in the form of .png files.  Please cite this Thesis document as the source of the above information; feel free to alert others to its existence and general availability.   116 Appendix B  EnergyWorld State Variables  This material is a complement to the Chapter 2 section “Social Agent ABM Specification”, and lists EnergyWorld’s state variables in the form developed at the Dahlem 100 Conference “New Approaches in Economics after the Financial Crisis” (Berlin Dahlem, August 28-31, 2010). The elements in the following table are those that would be needed to restart a simulation from a completed simulation step. EnergyWorld State Variables “Type” Legend: I = integer, R = real number, A = alpha, L = logical; () = list Name Type Description Update Initialization # Ticks per simulation I  NA In-code assignment # Ticks Memory Duration I Agents access rolling memory NA In-code assignment Maximum # messages/Agent/Tick I Best if > expected Affine count NA In-code assignment Maximum # message/batch I  NA In-code assignment Maximum Age I Agent automatically dies after NA In-code assignment Maturity Age I If younger, no Discussions NA In-code assignment Retirement Age I If older, no Discussions NA In-code assignment Gender ratios at all Ages 0<R<1 Affects gender at birth NA In-code assignment Birth rate R=I/1000 Affects # pregnancies each Tick NA In-code assignment Death rate R=I/1000 Affect death by accident NA In-code assignment Gestation Ticks I Duration of pregnancies NA In-code assignment Pregnancy Delay Ticks I No pregnancy after last birth NA In-code assignment Maximum Pregnancy Age I If older, no pregnancy NA In-code assignment PDF – Value n (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Value n (female) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Cognitive Heuristic n (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Cognitive Heuristic n (female) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Need n (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Need n (female) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Bounce Up (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Bounce Up (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Bounce Down (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Bounce Down (female) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Belief n (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Belief n (female) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Desire n (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Desire n (female) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Power n (male) R>=0 Relative frequency at each age NA As per PDFSpec* PDF – Power n (female) R>=0 Relative frequency at each age NA As per PDFSpec* Threshold Violation Tolerance Low 0<=R<=1 % under allowance before action NA In-code assignment Threshold Violation Tolerance High 0<=R<=1 % over allowance before action NA In-code assignment Prosperity Trend Threshold 0<=R<=1 % over allowance before action NA In-code assignment Hardship Trend Threshold 0<=R<=1 % over allowance before action NA In-code assignment Power to Savings Conversion Rate R>=0 Convert Power to Energy Savings NA In-code assignment Savings to Power Conversion Rate R>=0 Convert Energy Savings to Power NA In-code assignment Affinities Specification* Spec* For generating Agents’ Affine NA In-code assignment Thought Leader Fraction 0<=R<=1 Increased weight in Discussions NA In-code assignment Thought Leader Weights R>0 See previous NA In-code assignment Max Community Savings Depletion Rate 0<=R<=1 Limit on total aid to Agents per Tick NA In-code assignment   117 EnergyWorld State Variables “Type” Legend: I = integer, R = real number, A = alpha, L = logical; () = list Name Type Description Update Initialization Energy Supply (I) Total for all Agents each Tick NA In-code assignment Simulation Tick I # simulation cycles complete + 1 at end of each cycle Start at 1 AgentData (each Agent) () Next-indented items below        Demographic Identity () Next-indented items below             Age I              Gender “M”/”F”              Pregnancy Status I 0 = not pregnant >0 = months pregnant <0 = months to delay pregnancy = 1 if 0 and need more pregnant agents, and randomly choose this one  = - Pregnancy Delay Ticks if just gave birth + 1 if not 0 = 0 when Agent is generated or born           Needs (I) Prioritized hierarchy, must have Core Needs Each Tick, decrement by shortfall in Personal Usage Claim, cumulatively starting with lowest priority Optional Need As per age/gender Needs PDF           Affines ((I)) IDs of Affine Agents in each Affine group specified Affines removed if die As per Affinities Specification**           Thought Leader Status L These Agents’ messages during Satisfaction Discussions carry more weight NA (permanent) As per Thought Leader Fraction parameter      Worldview Filter () Next0indented items below             Values (I) Guides Agents’ generosity when assisting others in hardship NA (permanent) As per age/gender Values PDFs           Cognitive Heuristics (I) Specific heuristics mediates each decision NA (permanent) As per age/gender Heuristics PDFs           Bounce Up Level I Overcompensation for Heuristics’ delay/reduce/emphasize effect NA (permanent) As per age/gender Bounce Up PDF           Bounce Down Level I Overcompensation for Heuristics’ delay/reduce/emphasize effect NA (permanent) As per age/gender Bounce Down PDF      Activation Profile () Next-indented items below             Beliefs (I) Higher Beliefs drive higher Desires Persistently unmet Claims reduce Belief/Desire level in cascade fashion As per age/gender Beliefs PDFs           Desires (I) Higher Desires drive higher Intentions (Claims, below) Persistently unmet Claims reduce Belief/Desire level in cascade fashion As per age/gender Desires PDFs           Power Level I Power determines order (random within Power level) of allocating Energy Supply to Agent Claims If Agent Savings are exhausted by topping up unmet Needs, Power is converted to Agent Energy Savings As per age/gender Power PDF      Claims Data (aka ‘Demands’) () Next-indented items below             Personal Usage Claims (I) List of previous Claims (length = Memory Ticks) As generated from Desires in each Tick Fictitious ‘history’ at simulation start           Personal Savings Claims (I) List of previous Claims (length = Memory Ticks) As generated from Desires in each Tick Fictitious ‘history’ at simulation start           Community Savings Claims (I) List of previous Claims (length = Memory Ticks) As generated from Desires in each Tick Fictitious ‘history’ at simulation start      Allocations Data () Next-indented items below            Personal Usage Allocations (I) List of previous Allocations (length = Memory Ticks) As allocated from Energy Supply in each Tick Fictitious ‘history’ at simulation start          Personal Savings Allocations (I) List of previous Allocations (length = Memory Ticks) As allocated from Energy Supply in each Tick Fictitious ‘history’ at simulation start          Community Savings Allocations (I) List of previous Allocations (length = Memory Ticks) As allocated from Energy Supply in each Tick Fictitious ‘history’ at simulation start      Transfers Data (I) List of previous Transfers (length = Memory Ticks) from Savings to hardship Affines’ Savings As allocated from Personal Savings to hardship Affines’ Personal Savings Fictitious ‘history’ as simulation start, all 0      Satisfactions Data (I) List of previous Satisfaction Sum of total Allocation Fictitious ‘history’   118 EnergyWorld State Variables “Type” Legend: I = integer, R = real number, A = alpha, L = logical; () = list Name Type Description Update Initialization levels (length = Memory Ticks) minus total Claims as simulation start, all 0      Total Personal Savings I Rolling sum of Personal Savings, i.e. no Memory Tick limit Rolling sum of Personal Savings, i.e. no Memory Tick limit = 0 Power Block Data (I,I) Low and High Agent Indices for each Power level (AgentData sorted by Power) Agent deletions, Agent creations, and Power-to/from-Savings transfers cause rework As per AgentData at simulation start  • PDFSpec The “PDF Specification” is a mechanism for creating user-determined probability density functions for Value/Belief/Desire/Need/Bias/Bounce/Power parameters, each with an arbitrary Domain and Range that is mapped onto High and Low values for each parameter, for each Age/Gender combination. Though the full specification is too voluminous to include here, the following three images represent the specification and how it translates into values for various parameters.   • The above specification fragment represents 20 probability density functions (PDFs); EnergyWorld uses some 23 (parameters) * 2 (Male/Female) * 10 (age groups) = 460 such PDFs. The next image shows the PDF for one such PDF contained in the above fragment   119 (the “Value” parameter “Commitment to Self Development”, for Males between 40 and 49 years of age).            In the case of this parameter, the Range is mapped onto the interval (0.0-1.0) for use in the EnergyWorld simulator. The following 3D image adjoins the PDFs for Males of all ages for the relevant parameter.                 120  **Affinities Specification The following is an image capture from our Python-based implementation of EnergyWorld, showing the Affinities Specification mentioned in the State Variables table above.   In this version of an Affinities Specification, there are three kinds of Affinity Groups (we have chosen “PersonalRelationship”, “CulturalBinding”, “SharedPrinciple”), each of which have three strength levels (“Strong”, “Medium”, “Weak”). Each such level has information indicating how to generate, during simulation initialization, a number of Affines (from a range that is either an absolute “#” or a percentage of Starting Population “%”) for each Agent and how these Affines will respond to the Agent’s hardship (from a range of Propensity to providing some level of aid, and a range of PersonalStoragePortion to provide).    121 Appendix C  Cognitive Bias Detail  We have chosen the following (some are composites of several ‘cognitive biases’ in the literature) to be part of our Social Agent instantiation of the Intelligent Agent paradigm, and have implemented these in EnergyWorld, as key internal-subjective behavior drivers in the Agent community involved in our simulations: • RiskAvoidanceEffect: Used in AssessAdjustActivationProfile in determining whether to convert PersonalStorage to Power or vice versa, depending on whether the Agent is experiencing Hardship or Prosperity; a risk tolerant Agent experiencing Hardship will convert Power to PersonalStorage, and conversely, a risk tolerant Agent experiencing Prosperity, will convert PersonalStorage to Power. A risk averse Agent will adjust Needs up or down depending on whether the Agent is experience Prosperity or Hardship. • BelongingnessEffect: In DiscussionWithAffines, an Agent will accept an Affine’s satisfaction value if he has a sufficiently high Belongingness value, and will reject an Affine’s satisfaction value if he his Belongingness value is sufficiently low; ‘sufficiency’ here is mediated by comparison to a random value drawn from the same range as the range of the Belongingness factor. • HaloEffect: Also used in DiscussionWithAffines, an Agent will put more weight on an Affine that is also a ThoughtLeader. • MemoryEffect: Also known as the Recency Bias and here used in AssessAdjustActivationProfile, previous values of Satisfaction are weighted by how far in the past they are when computing a final Affine-mediated Satisfaction that an Agent has with the most recent energy Allocation.   122 • DenialEffect: The value of this factor determines how many simulation Ticks an Agent waits until ‘accepting’ that energy Allocations will be persistently less than Claims in the future, and thus delays Needs/Desires/Beliefs adjustments to respond to such shortfalls.      123 Appendix D  EnergyWorld Code Base  The EnergyWorld code exists in several versions corresponding to the specific needs for exploring the scenarios outlined in Chapter 5: Explorations with EnergyWorld. While not significantly different from each other, the specific differences are sometimes spread throughout the code. As a result, the totality of the combined code base is too large to include here. Instead, we have uploaded all versions of EnergyWorld, as well as all simulation Case listings and outputs to a Google Drive folder; see Appendix A Research Materials Database.      124 Appendix E  EnergyWorld Sensitivity Analysis  The following Excel file images list various groupings of simulation Cases for our Sensitivity Analysis and the Pearson Coefficient for each such grouping: (1) All Cases. (2) Cases for Energy Supply = “GradualDecline” (3) Cases for Energy Supply = “Crash” (4) Cases for PDFs = “Flat” (5) Cases for PDFs = “Shaped” (6) Cases for Energy Supply = “GradualDecline” + PDFs = “Flat”. (7) Cases for Energy Supply = “Crash” + PDFs = “Flat”.  (8) Cases for Energy Supply = “GradualDecline” + PDFs = “Shaped”. (9) Cases for Energy Supply = “Crash” + PDFs = “Shaped”. Here, “GradualDecline” means that relative Energy Supply declines from 1.0 of that at simulation start by 0.1 every 4th Tick, “Crash” means that relative Energy Supply declines from 1.0 of that at simulation start to 0.5 at the 2nd Tick and remains at that level, “Flat” means that all Beliefs/Desires/Needs/Sociality/Bias parameter values are drawn with equal probability for all Age/Gender combinations, and “Shaped” means that parameter values are drawn according to a probability density function over that domain, specified by the user for each Age/Gender combination.   125  EnergyWorld	Simulations	-	SensitivityExploration	CasesPearson	Coeefficient SumSqauresCode	Version	1 0.41925	Ticks -0.072Energy Adjustment 0.061	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks1 GradualDecline 1.0 Flat 2 200																									 340																									 146																															 43%2 GradualDecline 1.0 Flat 2 600																									 730																									 397																															 54%3 GradualDecline 1.0 Flat 2 1,000																						 1,110																						 542																															 49%4 GradualDecline 1.0 Flat 5 200																									 340																									 145																															 43%5 GradualDecline 1.0 Flat 5 600																									 730																									 299																															 41%6 GradualDecline 1.0 Flat 5 1,000																						 1,110																						 564																															 51%7 GradualDecline 1.0 Flat 10 200																									 340																									 152																															 45%8 GradualDecline 1.0 Flat 10 600																									 730																									 343																															 47%9 GradualDecline 1.0 Flat 10 1,000																						 588																									 175																															 30%10 GradualDecline 1.0 Shaped 2 200																									 340																									 174																															 51%11 GradualDecline 1.0 Shaped 2 600																									 730																									 365																															 50%12 GradualDecline 1.0 Shaped 2 1,000																						 1,110																						 484																															 44%13 GradualDecline 1.0 Shaped 5 200																									 340																									 166																															 49%14 GradualDecline 1.0 Shaped 5 600																									 730																									 412																															 56%15 GradualDecline 1.0 Shaped 5 1,000																						 1,110																						 506																															 46%16 GradualDecline 1.0 Shaped 10 200																									 340																									 169																															 50%17 GradualDecline 1.0 Shaped 10 600																									 730																									 351																															 48%18 GradualDecline 1.0 Shaped 10 1,000																						 1,110																						 657																															 59%19 GradualDecline 3.0 Flat 2 200																									 340																									 172																															 51%20 GradualDecline 3.0 Flat 2 600																									 730																									 342																															 47%21 GradualDecline 3.0 Flat 2 1,000																						 1,110																						 526																															 47%22 GradualDecline 3.0 Flat 5 200																									 340																									 139																															 41%23 GradualDecline 3.0 Flat 5 600																									 730																									 375																															 51%24 GradualDecline 3.0 Flat 5 1,000																						 1,110																						 520																															 47%25 GradualDecline 3.0 Flat 10 200																									 340																									 155																															 46%26 GradualDecline 3.0 Flat 10 600																									 730																									 440																															 60%27 GradualDecline 3.0 Flat 10 1,000																						 1,110																						 580																															 52%28 GradualDecline 3.0 Shaped 2 200																									 340																									 149																															 44%29 GradualDecline 3.0 Shaped 2 600																									 730																									 355																															 49%30 GradualDecline 3.0 Shaped 2 1,000																						 1,110																						 565																															 51%31 GradualDecline 3.0 Shaped 5 200																									 340																									 164																															 48%32 GradualDecline 3.0 Shaped 5 600																									 730																									 370																															 51%33 GradualDecline 3.0 Shaped 5 1,000																						 1,110																						 626																															 56%34 GradualDecline 3.0 Shaped 10 200																									 340																									 130																															 38%35 GradualDecline 3.0 Shaped 10 600																									 730																									 374																															 51%36 GradualDecline 3.0 Shaped 10 1,000																						 1,110																						 634																															 57%37 GradualDecline 5.0 Flat 2 200																									 340																									 143																															 42%38 GradualDecline 5.0 Flat 2 600																									 730																									 405																															 55%39 GradualDecline 5.0 Flat 2 1,000																						 1,110																						 602																															 54%40 GradualDecline 5.0 Flat 5 200																									 340																									 137																															 40%41 GradualDecline 5.0 Flat 5 600																									 730																									 409																															 56%42 GradualDecline 5.0 Flat 5 1,000																						 1,110																						 509																															 46%43 GradualDecline 5.0 Flat 10 200																									 340																									 159																															 47%44 GradualDecline 5.0 Flat 10 600																									 730																									 381																															 52%45 GradualDecline 5.0 Flat 10 1,000																						 1,110																						 559																															 50%46 GradualDecline 5.0 Shaped 2 200																									 340																									 128																															 38%47 GradualDecline 5.0 Shaped 2 600																									 730																									 397																															 54%48 GradualDecline 5.0 Shaped 2 1,000																						 1,110																						 501																															 45%49 GradualDecline 5.0 Shaped 5 200																									 340																									 141																															 41%50 GradualDecline 5.0 Shaped 5 600																									 730																									 364																															 50%51 GradualDecline 5.0 Shaped 5 1,000																						 1,110																						 521																															 47%52 GradualDecline 5.0 Shaped 10 200																									 340																									 139																															 41%53 GradualDecline 5.0 Shaped 10 600																									 730																									 334																															 46%54 GradualDecline 5.0 Shaped 10 1,000																						 1,110																						 527																															 47%55 Crash 1.0 Flat 2 200																									 340																									 116																															 34%56 Crash 1.0 Flat 2 600																									 730																									 359																															 49%57 Crash 1.0 Flat 2 1,000																						 1,110																						 471																															 42%58 Crash 1.0 Flat 5 200																									 340																									 123																															 36%59 Crash 1.0 Flat 5 600																									 730																									 301																															 41%60 Crash 1.0 Flat 5 1,000																						 1,110																						 583																															 53%61 Crash 1.0 Flat 10 200																									 340																									 122																															 36%62 Crash 1.0 Flat 10 600																									 730																									 298																															 41%63 Crash 1.0 Flat 10 1,000																						 1,110																						 551																															 50%64 Crash 1.0 Shaped 2 200																									 340																									 126																															 37%65 Crash 1.0 Shaped 2 600																									 730																									 281																															 38%66 Crash 1.0 Shaped 2 1,000																						 1,110																						 429																															 39%All	CasesCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensity 0.184Population%@20Ticks	vs	AdjustmentAccelerationFactor  126  67 Crash 1.0 Shaped 5 200																									 340																									 127																															 37%68 Crash 1.0 Shaped 5 600																									 730																									 304																															 42%69 Crash 1.0 Shaped 5 1,000																						 1,110																						 465																															 42%70 Crash 1.0 Shaped 10 200																									 340																									 118																															 35%71 Crash 1.0 Shaped 10 600																									 730																									 291																															 40%72 Crash 1.0 Shaped 10 1,000																						 1,110																						 436																															 39%73 Crash 3.0 Flat 2 200																									 340																									 116																															 34%74 Crash 3.0 Flat 2 600																									 730																									 356																															 49%75 Crash 3.0 Flat 2 1,000																						 1,110																						 575																															 52%76 Crash 3.0 Flat 5 200																									 340																									 117																															 34%77 Crash 3.0 Flat 5 600																									 730																									 329																															 45%78 Crash 3.0 Flat 5 1,000																						 1,110																						 479																															 43%79 Crash 3.0 Flat 10 200																									 340																									 123																															 36%80 Crash 3.0 Flat 10 600																									 730																									 291																															 40%81 Crash 3.0 Flat 10 1,000																						 1,110																						 432																															 39%82 Crash 3.0 Shaped 2 200																									 340																									 117																															 34%83 Crash 3.0 Shaped 2 600																									 730																									 304																															 42%84 Crash 3.0 Shaped 2 1,000																						 1,110																						 535																															 48%85 Crash 3.0 Shaped 5 200																									 340																									 125																															 37%86 Crash 3.0 Shaped 5 600																									 730																									 294																															 40%87 Crash 3.0 Shaped 5 1,000																						 1,110																						 462																															 42%88 Crash 3.0 Shaped 10 200																									 340																									 119																															 35%89 Crash 3.0 Shaped 10 600																									 730																									 293																															 40%90 Crash 3.0 Shaped 10 1,000																						 1,110																						 401																															 36%91 Crash 5.0 Flat 2 200																									 340																									 143																															 42%92 Crash 5.0 Flat 2 600																									 730																									 325																															 45%93 Crash 5.0 Flat 2 1,000																						 1,110																						 431																															 39%94 Crash 5.0 Flat 5 200																									 340																									 127																															 37%95 Crash 5.0 Flat 5 600																									 730																									 336																															 46%96 Crash 5.0 Flat 5 1,000																						 1,110																						 610																															 55%97 Crash 5.0 Flat 10 200																									 340																									 110																															 32%98 Crash 5.0 Flat 10 600																									 730																									 331																															 45%99 Crash 5.0 Flat 10 1,000																						 1,110																						 507																															 46%100 Crash 5.0 Shaped 2 200																									 340																									 139																															 41%101 Crash 5.0 Shaped 2 600																									 730																									 354																															 48%102 Crash 5.0 Shaped 2 1,000																						 1,110																						 475																															 43%103 Crash 5.0 Shaped 5 200																									 340																									 136																															 40%104 Crash 5.0 Shaped 5 600																									 730																									 308																															 42%105 Crash 5.0 Shaped 5 1,000																						 1,110																						 426																															 38%106 Crash 5.0 Shaped 10 200																									 340																									 132																															 39%107 Crash 5.0 Shaped 10 600																									 730																									 313																															 43%108 Crash 5.0 Shaped 10 1,000																						 1,110																						 512																															 46%Notes:'Message	Intensity'	=	NumMessagesPerTickPerAgent'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'GradualDecline'	=	1.0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum  127   EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 0.42625	Ticks -0.001Energy Adjustment -0.007	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks1 GradualDecline 1 Flat 2 200 340 146 43%2 GradualDecline 1 Flat 2 600 730 397 54%3 GradualDecline 1 Flat 2 1000 1110 542 49%4 GradualDecline 1 Flat 5 200 340 145 43%5 GradualDecline 1 Flat 5 600 730 299 41%6 GradualDecline 1 Flat 5 1000 1110 564 51%7 GradualDecline 1 Flat 10 200 340 152 45%8 GradualDecline 1 Flat 10 600 730 343 47%9 GradualDecline 1 Flat 10 1000 588 175 30%10 GradualDecline 1 Shaped 2 200 340 174 51%11 GradualDecline 1 Shaped 2 600 730 365 50%12 GradualDecline 1 Shaped 2 1000 1110 484 44%13 GradualDecline 1 Shaped 5 200 340 166 49%14 GradualDecline 1 Shaped 5 600 730 412 56%15 GradualDecline 1 Shaped 5 1000 1110 506 46%16 GradualDecline 1 Shaped 10 200 340 169 50%17 GradualDecline 1 Shaped 10 600 730 351 48%18 GradualDecline 1 Shaped 10 1000 1110 657 59%19 GradualDecline 3 Flat 2 200 340 172 51%20 GradualDecline 3 Flat 2 600 730 342 47%21 GradualDecline 3 Flat 2 1000 1110 526 47%22 GradualDecline 3 Flat 5 200 340 139 41%23 GradualDecline 3 Flat 5 600 730 375 51%24 GradualDecline 3 Flat 5 1000 1110 520 47%25 GradualDecline 3 Flat 10 200 340 155 46%26 GradualDecline 3 Flat 10 600 730 440 60%27 GradualDecline 3 Flat 10 1000 1110 580 52%28 GradualDecline 3 Shaped 2 200 340 149 44%29 GradualDecline 3 Shaped 2 600 730 355 49%30 GradualDecline 3 Shaped 2 1000 1110 565 51%31 GradualDecline 3 Shaped 5 200 340 164 48%32 GradualDecline 3 Shaped 5 600 730 370 51%33 GradualDecline 3 Shaped 5 1000 1110 626 56%34 GradualDecline 3 Shaped 10 200 340 130 38%35 GradualDecline 3 Shaped 10 600 730 374 51%36 GradualDecline 3 Shaped 10 1000 1110 634 57%37 GradualDecline 5 Flat 2 200 340 143 42%38 GradualDecline 5 Flat 2 600 730 405 55%39 GradualDecline 5 Flat 2 1000 1110 602 54%40 GradualDecline 5 Flat 5 200 340 137 40%41 GradualDecline 5 Flat 5 600 730 409 56%42 GradualDecline 5 Flat 5 1000 1110 509 46%43 GradualDecline 5 Flat 10 200 340 159 47%44 GradualDecline 5 Flat 10 600 730 381 52%45 GradualDecline 5 Flat 10 1000 1110 559 50%46 GradualDecline 5 Shaped 2 200 340 128 38%47 GradualDecline 5 Shaped 2 600 730 397 54%48 GradualDecline 5 Shaped 2 1000 1110 501 45%49 GradualDecline 5 Shaped 5 200 340 141 41%50 GradualDecline 5 Shaped 5 600 730 364 50%51 GradualDecline 5 Shaped 5 1000 1110 521 47%52 GradualDecline 5 Shaped 10 200 340 139 41%53 GradualDecline 5 Shaped 10 600 730 334 46%54 GradualDecline 5 Shaped 10 1000 1110 527 47%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentAll	'GradualDecline'	EnergySupply	CasesCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.182  128    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 0.01325	Ticks 0.002Energy Adjustment -0.085	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks55 Crash 1 Flat 2 200 340 116 25%56 Crash 1 Flat 2 600 730 359 7%57 Crash 1 Flat 2 1000 1110 471 10%58 Crash 1 Flat 5 200 340 123 25%59 Crash 1 Flat 5 600 730 301 8%60 Crash 1 Flat 5 1000 1110 583 8%61 Crash 1 Flat 10 200 340 122 16%62 Crash 1 Flat 10 600 730 298 12%63 Crash 1 Flat 10 1000 1110 551 24%64 Crash 1 Shaped 2 200 340 126 9%65 Crash 1 Shaped 2 600 730 281 2%66 Crash 1 Shaped 2 1000 1110 429 21%67 Crash 1 Shaped 5 200 340 127 5%68 Crash 1 Shaped 5 600 730 304 23%69 Crash 1 Shaped 5 1000 1110 465 22%70 Crash 1 Shaped 10 200 340 118 7%71 Crash 1 Shaped 10 600 730 291 20%72 Crash 1 Shaped 10 1000 1110 436 16%73 Crash 3 Flat 2 200 340 116 14%74 Crash 3 Flat 2 600 730 356 8%75 Crash 3 Flat 2 1000 1110 575 6%76 Crash 3 Flat 5 200 340 117 13%77 Crash 3 Flat 5 600 730 329 25%78 Crash 3 Flat 5 1000 1110 479 14%79 Crash 3 Flat 10 200 340 123 14%80 Crash 3 Flat 10 600 730 291 15%81 Crash 3 Flat 10 1000 1110 432 15%82 Crash 3 Shaped 2 200 340 117 14%83 Crash 3 Shaped 2 600 730 304 3%84 Crash 3 Shaped 2 1000 1110 535 4%85 Crash 3 Shaped 5 200 340 125 5%86 Crash 3 Shaped 5 600 730 294 2%87 Crash 3 Shaped 5 1000 1110 462 3%88 Crash 3 Shaped 10 200 340 119 23%89 Crash 3 Shaped 10 600 730 293 5%90 Crash 3 Shaped 10 1000 1110 401 2%91 Crash 5 Flat 2 200 340 143 8%92 Crash 5 Flat 2 600 730 325 8%93 Crash 5 Flat 2 1000 1110 431 6%94 Crash 5 Flat 5 200 340 127 0%95 Crash 5 Flat 5 600 730 336 11%96 Crash 5 Flat 5 1000 1110 610 7%97 Crash 5 Flat 10 200 340 110 14%98 Crash 5 Flat 10 600 730 331 6%99 Crash 5 Flat 10 1000 1110 507 25%100 Crash 5 Shaped 2 200 340 139 6%101 Crash 5 Shaped 2 600 730 354 4%102 Crash 5 Shaped 2 1000 1110 475 3%103 Crash 5 Shaped 5 200 340 136 3%104 Crash 5 Shaped 5 600 730 308 24%105 Crash 5 Shaped 5 1000 1110 426 7%106 Crash 5 Shaped 10 200 340 132 4%107 Crash 5 Shaped 10 600 730 313 6%108 Crash 5 Shaped 10 1000 1110 512 5%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentAll	'Crash'	EnergySupply	CasesCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.007  129    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 0.50125	Ticks -0.115Energy Adjustment 0.157	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks1 GradualDecline 1 Flat 2 200 340 146 43%2 GradualDecline 1 Flat 2 600 730 397 54%3 GradualDecline 1 Flat 2 1000 1110 542 49%4 GradualDecline 1 Flat 5 200 340 145 43%5 GradualDecline 1 Flat 5 600 730 299 41%6 GradualDecline 1 Flat 5 1000 1110 564 51%7 GradualDecline 1 Flat 10 200 340 152 45%8 GradualDecline 1 Flat 10 600 730 343 47%9 GradualDecline 1 Flat 10 1000 588 175 30%19 GradualDecline 3 Flat 2 200 340 172 51%20 GradualDecline 3 Flat 2 600 730 342 47%21 GradualDecline 3 Flat 2 1000 1110 526 47%22 GradualDecline 3 Flat 5 200 340 139 41%23 GradualDecline 3 Flat 5 600 730 375 51%24 GradualDecline 3 Flat 5 1000 1110 520 47%25 GradualDecline 3 Flat 10 200 340 155 46%26 GradualDecline 3 Flat 10 600 730 440 60%27 GradualDecline 3 Flat 10 1000 1110 580 52%37 GradualDecline 5 Flat 2 200 340 143 42%38 GradualDecline 5 Flat 2 600 730 405 55%39 GradualDecline 5 Flat 2 1000 1110 602 54%40 GradualDecline 5 Flat 5 200 340 137 40%41 GradualDecline 5 Flat 5 600 730 409 56%42 GradualDecline 5 Flat 5 1000 1110 509 46%43 GradualDecline 5 Flat 10 200 340 159 47%44 GradualDecline 5 Flat 10 600 730 381 52%45 GradualDecline 5 Flat 10 1000 1110 559 50%55 Crash 1 Flat 2 200 340 116 34%56 Crash 1 Flat 2 600 730 359 49%57 Crash 1 Flat 2 1000 1110 471 42%58 Crash 1 Flat 5 200 340 123 36%59 Crash 1 Flat 5 600 730 301 41%60 Crash 1 Flat 5 1000 1110 583 53%61 Crash 1 Flat 10 200 340 122 36%62 Crash 1 Flat 10 600 730 298 41%63 Crash 1 Flat 10 1000 1110 551 50%73 Crash 3 Flat 2 200 340 116 34%74 Crash 3 Flat 2 600 730 356 49%75 Crash 3 Flat 2 1000 1110 575 52%76 Crash 3 Flat 5 200 340 117 34%77 Crash 3 Flat 5 600 730 329 45%78 Crash 3 Flat 5 1000 1110 479 43%79 Crash 3 Flat 10 200 340 123 36%80 Crash 3 Flat 10 600 730 291 40%81 Crash 3 Flat 10 1000 1110 432 39%91 Crash 5 Flat 2 200 340 143 42%92 Crash 5 Flat 2 600 730 325 45%93 Crash 5 Flat 2 1000 1110 431 39%94 Crash 5 Flat 5 200 340 127 37%95 Crash 5 Flat 5 600 730 336 46%96 Crash 5 Flat 5 1000 1110 610 55%97 Crash 5 Flat 10 200 340 110 32%98 Crash 5 Flat 10 600 730 331 45%99 Crash 5 Flat 10 1000 1110 507 46%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum'Message	Intensity'	=	NumMessages	PerTickPerAgentPopulation%@20Ticks	vs	MessageDensityPopulation%A@20Ticks	vs	AdjustmentAcclerationFactorAll	'Flat'	PDF	CasesCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulation0.289'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that  130    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 0.33425	Ticks -0.026Energy Adjustment -0.042	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks10 GradualDecline 1 Shaped 2 200 340 174 51%11 GradualDecline 1 Shaped 2 600 730 365 50%12 GradualDecline 1 Shaped 2 1000 1110 484 44%13 GradualDecline 1 Shaped 5 200 340 166 49%14 GradualDecline 1 Shaped 5 600 730 412 56%15 GradualDecline 1 Shaped 5 1000 1110 506 46%16 GradualDecline 1 Shaped 10 200 340 169 50%17 GradualDecline 1 Shaped 10 600 730 351 48%18 GradualDecline 1 Shaped 10 1000 1110 657 59%28 GradualDecline 3 Shaped 2 200 340 149 44%29 GradualDecline 3 Shaped 2 600 730 355 49%30 GradualDecline 3 Shaped 2 1000 1110 565 51%31 GradualDecline 3 Shaped 5 200 340 164 48%32 GradualDecline 3 Shaped 5 600 730 370 51%33 GradualDecline 3 Shaped 5 1000 1110 626 56%34 GradualDecline 3 Shaped 10 200 340 130 38%35 GradualDecline 3 Shaped 10 600 730 374 51%36 GradualDecline 3 Shaped 10 1000 1110 634 57%46 GradualDecline 5 Shaped 2 200 340 128 38%47 GradualDecline 5 Shaped 2 600 730 397 54%48 GradualDecline 5 Shaped 2 1000 1110 501 45%49 GradualDecline 5 Shaped 5 200 340 141 41%50 GradualDecline 5 Shaped 5 600 730 364 50%51 GradualDecline 5 Shaped 5 1000 1110 521 47%52 GradualDecline 5 Shaped 10 200 340 139 41%53 GradualDecline 5 Shaped 10 600 730 334 46%54 GradualDecline 5 Shaped 10 1000 1110 527 47%64 Crash 1 Shaped 2 200 340 126 37%65 Crash 1 Shaped 2 600 730 281 38%66 Crash 1 Shaped 2 1000 1110 429 39%67 Crash 1 Shaped 5 200 340 127 37%68 Crash 1 Shaped 5 600 730 304 42%69 Crash 1 Shaped 5 1000 1110 465 42%70 Crash 1 Shaped 10 200 340 118 35%71 Crash 1 Shaped 10 600 730 291 40%72 Crash 1 Shaped 10 1000 1110 436 39%82 Crash 3 Shaped 2 200 340 117 34%83 Crash 3 Shaped 2 600 730 304 42%84 Crash 3 Shaped 2 1000 1110 535 48%85 Crash 3 Shaped 5 200 340 125 37%86 Crash 3 Shaped 5 600 730 294 40%87 Crash 3 Shaped 5 1000 1110 462 42%88 Crash 3 Shaped 10 200 340 119 35%89 Crash 3 Shaped 10 600 730 293 40%90 Crash 3 Shaped 10 1000 1110 401 36%100 Crash 5 Shaped 2 200 340 139 41%101 Crash 5 Shaped 2 600 730 354 48%102 Crash 5 Shaped 2 1000 1110 475 43%103 Crash 5 Shaped 5 200 340 136 40%104 Crash 5 Shaped 5 600 730 308 42%105 Crash 5 Shaped 5 1000 1110 426 38%106 Crash 5 Shaped 10 200 340 132 39%107 Crash 5 Shaped 10 600 730 313 43%108 Crash 5 Shaped 10 1000 1110 512 46%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentAll	'Shaped'	PDF	CasesCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%A@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.114  131    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 0.41325	Ticks -0.077Energy Adjustment 0.304	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks1 GradualDecline 1 Flat 2 200 340 146 43%2 GradualDecline 1 Flat 2 600 730 397 54%3 GradualDecline 1 Flat 2 1000 1110 542 49%4 GradualDecline 1 Flat 5 200 340 145 43%5 GradualDecline 1 Flat 5 600 730 299 41%6 GradualDecline 1 Flat 5 1000 1110 564 51%7 GradualDecline 1 Flat 10 200 340 152 45%8 GradualDecline 1 Flat 10 600 730 343 47%9 GradualDecline 1 Flat 10 1000 588 175 30%19 GradualDecline 3 Flat 2 200 340 172 51%20 GradualDecline 3 Flat 2 600 730 342 47%21 GradualDecline 3 Flat 2 1000 1110 526 47%22 GradualDecline 3 Flat 5 200 340 139 41%23 GradualDecline 3 Flat 5 600 730 375 51%24 GradualDecline 3 Flat 5 1000 1110 520 47%25 GradualDecline 3 Flat 10 200 340 155 46%26 GradualDecline 3 Flat 10 600 730 440 60%27 GradualDecline 3 Flat 10 1000 1110 580 52%37 GradualDecline 5 Flat 2 200 340 143 42%38 GradualDecline 5 Flat 2 600 730 405 55%39 GradualDecline 5 Flat 2 1000 1110 602 54%40 GradualDecline 5 Flat 5 200 340 137 40%41 GradualDecline 5 Flat 5 600 730 409 56%42 GradualDecline 5 Flat 5 1000 1110 509 46%43 GradualDecline 5 Flat 10 200 340 159 47%44 GradualDecline 5 Flat 10 600 730 381 52%45 GradualDecline 5 Flat 10 1000 1110 559 50%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentCases	for	'GradualDecline'	EnergySupply	&	'Flat'	PDFsCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%A@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.269  132    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSqauresCode	Version	1 0.12625	Ticks -0.258Energy Adjustment 0.165	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks10 GradualDecline 1 Shaped 2 200 340 174 0%11 GradualDecline 1 Shaped 2 600 730 365 10%12 GradualDecline 1 Shaped 2 1000 1110 484 25%13 GradualDecline 1 Shaped 5 200 340 166 12%14 GradualDecline 1 Shaped 5 600 730 412 20%15 GradualDecline 1 Shaped 5 1000 1110 506 7%16 GradualDecline 1 Shaped 10 200 340 169 24%17 GradualDecline 1 Shaped 10 600 730 351 7%18 GradualDecline 1 Shaped 10 1000 1110 657 4%28 GradualDecline 3 Shaped 2 200 340 149 24%29 GradualDecline 3 Shaped 2 600 730 355 7%30 GradualDecline 3 Shaped 2 1000 1110 565 24%31 GradualDecline 3 Shaped 5 200 340 164 27%32 GradualDecline 3 Shaped 5 600 730 370 7%33 GradualDecline 3 Shaped 5 1000 1110 626 10%34 GradualDecline 3 Shaped 10 200 340 130 4%35 GradualDecline 3 Shaped 10 600 730 374 6%36 GradualDecline 3 Shaped 10 1000 1110 634 12%46 GradualDecline 5 Shaped 2 200 340 128 3%47 GradualDecline 5 Shaped 2 600 730 397 27%48 GradualDecline 5 Shaped 2 1000 1110 501 27%49 GradualDecline 5 Shaped 5 200 340 141 21%50 GradualDecline 5 Shaped 5 600 730 364 3%51 GradualDecline 5 Shaped 5 1000 1110 521 24%52 GradualDecline 5 Shaped 10 200 340 139 2%53 GradualDecline 5 Shaped 10 600 730 334 26%54 GradualDecline 5 Shaped 10 1000 1110 527 10%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentCases	for	'GradualDecline'	EnergySupply	&	'Shaped'	PDFs	Correlation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%A@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.110  133    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 -0.09525	Ticks 0.330Energy Adjustment -0.334	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks55 Crash 1 Flat 2 200 340 116 25%56 Crash 1 Flat 2 600 730 359 7%57 Crash 1 Flat 2 1000 1110 471 10%58 Crash 1 Flat 5 200 340 123 25%59 Crash 1 Flat 5 600 730 301 8%60 Crash 1 Flat 5 1000 1110 583 8%61 Crash 1 Flat 10 200 340 122 16%62 Crash 1 Flat 10 600 730 298 12%63 Crash 1 Flat 10 1000 1110 551 24%73 Crash 3 Flat 2 200 340 116 14%74 Crash 3 Flat 2 600 730 356 8%75 Crash 3 Flat 2 1000 1110 575 6%76 Crash 3 Flat 5 200 340 117 13%77 Crash 3 Flat 5 600 730 329 25%78 Crash 3 Flat 5 1000 1110 479 14%79 Crash 3 Flat 10 200 340 123 14%80 Crash 3 Flat 10 600 730 291 15%81 Crash 3 Flat 10 1000 1110 432 15%91 Crash 5 Flat 2 200 340 143 8%92 Crash 5 Flat 2 600 730 325 8%93 Crash 5 Flat 2 1000 1110 431 6%94 Crash 5 Flat 5 200 340 127 0%95 Crash 5 Flat 5 600 730 336 11%96 Crash 5 Flat 5 1000 1110 610 7%97 Crash 5 Flat 10 200 340 110 14%98 Crash 5 Flat 10 600 730 331 6%99 Crash 5 Flat 10 1000 1110 507 25%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentCases	for	'Crash'	EnergySupply	&	'Flat'	PDFsCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%A@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.230  134    EnergyWorld	Simulations	-	SensitivityExplorationPearson	Coeefficient SumSquaresCode	Version	1 0.04225	Ticks 0.112Energy Adjustment -0.374	Supply Acceleration Message Suggested Starting Population Population	%Run	# Shape Factor PDFs Intensity Population Population After	20	Ticks After	20	Ticks64 Crash 1 Shaped 2 200 340 126 9%65 Crash 1 Shaped 2 600 730 281 2%66 Crash 1 Shaped 2 1000 1110 429 21%67 Crash 1 Shaped 5 200 340 127 5%68 Crash 1 Shaped 5 600 730 304 23%69 Crash 1 Shaped 5 1000 1110 465 22%70 Crash 1 Shaped 10 200 340 118 7%71 Crash 1 Shaped 10 600 730 291 20%72 Crash 1 Shaped 10 1000 1110 436 16%82 Crash 3 Shaped 2 200 340 117 14%83 Crash 3 Shaped 2 600 730 304 3%84 Crash 3 Shaped 2 1000 1110 535 4%85 Crash 3 Shaped 5 200 340 125 5%86 Crash 3 Shaped 5 600 730 294 2%87 Crash 3 Shaped 5 1000 1110 462 3%88 Crash 3 Shaped 10 200 340 119 23%89 Crash 3 Shaped 10 600 730 293 5%90 Crash 3 Shaped 10 1000 1110 401 2%100 Crash 5 Shaped 2 200 340 139 6%101 Crash 5 Shaped 2 600 730 354 4%102 Crash 5 Shaped 2 1000 1110 475 3%103 Crash 5 Shaped 5 200 340 136 3%104 Crash 5 Shaped 5 600 730 308 24%105 Crash 5 Shaped 5 1000 1110 426 7%106 Crash 5 Shaped 10 200 340 132 4%107 Crash 5 Shaped 10 600 730 313 6%108 Crash 5 Shaped 10 1000 1110 512 5%Notes:'Shaped'	=	Best	guess	for	differential	PDFs	for	each	age/gender'Message	Intensity'	=	NumMessages	PerTickPerAgentCases	for	'Crash'	EnergySupply	&	'Shaped'	PDFsCorrelation	Calculation	PairsPopulation%@20Ticks	vs	StartingPopulationPopulation%@20Ticks	vs	MessageDensityPopulation%A@20Ticks	vs	AdjustmentAcclerationFactor'GradualDecline'	=	0	at	Tick	0,	down	0.1	every	4th	Tick	after	that'Crash'	=	1.0	at	Tick	0	,	0.5	at	Tick	1	thru	Tick	25'Flat'	=	equal	probability	of	assigning	value	between	Minimum	and	Maximum0.154  135 Appendix F  EnergyWorld Scenario Exploration Results The volume of data produced in our EnergyWorld simulations is too large to include here. Instead, we have uploaded all our research materials to a freely-available Google Drive folder; see Appendix A Research Materials Database.     136 Appendix G  Additional Resources  We list here additional ABM initiatives that may interest the reader: • Air traffic control: Agent-based model of air traffic control to analyze control policies and performance of an air traffic management facility (Conway 2006). • Anthropology: Agent-based model of prehistoric settlement patterns and political considerations in the Lake Titicaca basin of Peru and Bolivia (Griffin and Stanish 2007). • Biomedical research: The Basic Immune Simulator, an agent-based model to study the interactions between innate and adaptive immunity (Folcik, An and Orasz 2007). • Chemistry: An agent-based approach to modeling molecular self-assembly (Troisi, Wong and Ratner 2005). • Crime analysis: Agent-based model that uses a realistic virtual urban environment, populated with virtual burglar agents (Malleson 2009). • Ecology: Agent-based model of predator-prey relationships of transient killer whales and other marine mammals (Mock and Testa 2007). • Energy analysis: Agent-based model for scenario development of offshore wind energy (Mast et al 2007). • Epidemic modeling: BioWar, a scalable multi-agent model that simulates individuals embedded in social, health and professional networks, and tracks the incidence of background and maliciously introduced diseases (Carley et al 2006). • Market analysis: Agent-based simulation that enables the possibilities for a future market in sub-orbital space tourism (Charinia et al 2006).   137 • Organizational Decision Making: Agent-based modeling approach to allow negotiations in order to achieve a global objective, specifically for planning the location of intermodal freight groups (van Dam et al 2007). 	

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0228887/manifest

Comment

Related Items