Agent-Based Modeling in Philosophy
print


Breadcrumb Navigation


Content

Program

Tutorials

Tutorials on programming agent-based models in NetLogo and Python open to all participants of the conference.

Location till 16:00: "Senatssaal", Room E110, Main Building, Geschwister-Scholl-Platz 1 
Location after 16:00:  Room M201, Main Building, Geschwister-Scholl-Platz 1 (Floor Plan, 661 Kb)

Wednesday, 10 December

TimeTopic
09:30 - 12:00 Modelling the Evolution of Social Contract and the Stag Hunt Part 1 (instructed by Conor Mayo-Wilson, University of Washington/MCMP and Kevin Zollman, Carnegie Mellon University/MCMP)
13:30 - 15:30 Modelling the Evolution of Social Contract and the Stag Hunt Part 2
16:00 - 18:00 Intro to Programming Agent Based Models in Python (instructed by Aidan Lyon, University of Maryland/MCMP)

Conference

Location: Mansion of the Center for Advanced Studies, Seestraße 13, 80802 München.

Thursday, 11 December

TimeTopic
10:00 - 10:10 Welcome.
10:10 - 10:50 Felipe Romero: The Fragility of Scientific Self-Correction
10:50 - 11:10 Coffee Break
11:10 - 11:50 Manolo Martínez: Signalling Games and Modality
11:50 - 12:30 Carlos Santana: Modeling the Evolution of Language with and without much Empirical Data
12:30 - 14:00 Lunch
14:00 - 14.40 Krist Vaesen, Wybo Houkes and Adrien Querbes: No Strength in Numbers after all? A Robustness Analysis of the Demographic Effect on Scientific and Technological Change
14:40 - 15:20 Alexander Reutlinger and Dominik Hangleiter: Agent-based Simulations in the Sciences: Explanation without Understanding
15:20 - 16:00 Aidan Lyon and Eric Pacuit: Deliberating in a Prediction Market
16:00 - 16:15 Coffee Break
16:15 - 17:45 Scott Page: Collective Accuracy: Agent Based & Emergent vs Statistical and Assumed
(Watch the lecture @ LMUcast)

Friday, 12 December

TimeTopic
09:00 - 09:40 Cailin O'Connor: Evolving to Generalize: Trading Precision for Speed
09:40 - 10:20 Ty Branch: Agent-Based Modeling for Weak Emergence
10:20 - 10:40 Coffee Break
10:40 - 11:20 Pierrick Bourrat: Endogenizing Reproduction and Inheritance: An Agent Based Modeling Approach
11:20 - 12:50 Kevin Zollman: The Formation of Epistemic Networks
12:50 - 14:00 Lunch
14:00 - 14:40 Rebecca Macintosh: Evolutionary Game Theory's Moral Meddlings
14:40 - 15:20 Corinna Elsenbroich and Rainer Hegselmann: Agent-based Models in Moral Philosophy
15:20 - 16:00 Hannah Übler: Simulating Trends in Artificial Influence Networks
16:00 - 16:15 Coffee Break
16:15 - 17:45 Rosaria Conte: Minding Norms. Mechanisms and Dynamics of Social Order in Agent Societies
(Watch the lecture @ LMUcast)
19:00 Conference Dinner

Saturday, 13 December

TimeTopic
09:00 - 09:40 Rogier De Langhe: From Theory Choice to Theory Search
09:40 - 10:20 Samuli Pöyhönen: Navigating an Epistemic Landscape: Foraging vs. Broadcasting as Models of Socially Distributed Problem-Solving
10:20 - 10:40 Coffee Break
10:40 - 11:20 Bert Baumgaertner: Belief Amplification and Imitation in an Extended Voter Model
11:20 - 12:50 Jason Alexander: The Structural Evolution of Morality
(Watch the lecture @ LMUcast)
12:50 - 14:00 Lunch
14:00 - 14:40 Remco Heesen: Three Ways To Become An Academic Superstar
14:40 - 15:20 Elena M. Tur, Paolo Zeppini and Koen Frenken: Diffusion of Ideas, Social Reinforcement and Percolation
15:20 - 16:00 Thomas Boyer and Cyrille Imbert: Explaining Scientific Collaboration from the Microscale: Do two Heads need to be more than twice better than one?
16:00 - 16:15 Coffee Break
16:15 - 17:45 Michael Weisberg: Agent-based Models and Confirmation Theory
(Watch the lecture @ LMUcast)

Abstracts

Jason McKenzie Alexander (London School of Economics and Political Science): The Structural Evolution of Morality

One general problem faced by attempts to explain the origins of morality using traditional rational choice theory is that the demands of rationality and the demands of morality often fail to coincide. This can happen in at least three different ways. Sometimes our moral intuitions recommend actions which are identified as irrational (such as cooperating in the prisoner's dilemma or in the centipede game, or rejecting unfair offers in the ultimatum game). Sometimes our moral intuitions recommend an act which is only one of several recognised as rational (as can happen in games having multiple Nash equilibria). And sometimes we have multiple competing moral intuitions in cases where rationality recommends a unique act (such as in asymmetric bargaining games, in contrast to the Nash solution). In this talk, I present a number of results drawn from agent-based models of imitative learning on social networks, showing how this single framework manages to explain many of our moral intuitions across a wide variety of diverse cases. top

Bert Baumgaertner (University of Idaho): Belief Amplification and Imitation in an Extended Voter Model

The first purpose of this talk is to present a philosophically interesting agent-based model (ABM) that extends the basic voter model of opinion dynamics. The second purpose of this presentation is to use this model as a case study for addressing questions regarding validation of ABMs. Two steps are proposed in the validation process: the first focuses on evaluating the individual components of the model, the second focuses on the behavior of the model as an integrated whole.top

Pierrick Bourrat (University of Sydney): Endogenizing Reproduction and Inheritance: An Agent Based Modeling Approach

Agent-based models (ABMs) are increasingly used in philosophy. Yet, this is not the case in philosophy of biology where their use still remains marginal. In this paper, I show that ABMs can successfully be used to make progress in this discipline. I use one particular example to illustrate this point. Reproduction and inheritance are classically regarded as necessary conditions for evolution by natural selection. I argue, using a set of ABMs and putting into practice what Okasha calls the ‘strategy of endogenization’, that contrary to the classical view, reproduction and inheritance can be regarded as products of natural selection.top

Thomas Boyer-Kassem (Archives Poincaré, Université de Lorraine) and Cyrille Imbert (Archives Poincaré, Université de Lorraine): Explaining Scientific Collaboration from the Microscale: Do two Heads need to be more than twice better than one?

Whereas existing accounts of scientific collaboration make strong assumptions about the beneficialness of collaboration, we propose a deliberately crude agent-based model with unfavorable hypotheses in which collaborations boils down to information sharing. Then, based on simulations, we derive the reward structure for various organizations of community under a priority reward scheme. This shows that, surprisingly, information sharing can be sufficient to make collaboration very beneficial. We then analyze the explanatory potential of our model by exhibiting how our results about successfulness at the microlevel translate into macrolevel patterns, which makes the connection between successfulness and collaborative behavior more plausible.

Ty Branch (University of Waterloo): Agent-Based Modeling for Weak Emergence

In order to resolve the tension between strong and weak emergence, Isuggest we reconsider the ability for weak emergence to account for seemingly irreducible phenomena though the use of computation. As Agent-based Modeling (ABM) can aid in mapping the higher-level interactions of a system, it can also help develop a reductive explanation of these higher-level phenomena to lower-level interactions. With this, weak emergence will become increasingly able to define more phenomena. Hence, ABM is an invaluable method for engaging the ontological and epistemic viability of weak emergence, ultimately threatening strong emergence.

top

Rosaria Conte (Institute for Cognitive Science and Technology/Rome): "Minding Norms. Mechanisms and Dynamics of Social Order in Agent Societies"

Despite ubiquity and universality, norms are still awaiting for a general comprehensive theory. In the presentation, a conceptual, theoretical, and computational framework will be proposed to provide a general account of norms, enabling us to investigate: (a) differences and commonalities among social, moral, and legal norms; (b) norm emergence and change; (c) the individual properties involved or responsible for bringing about norms. The main thesis is that observable conformity is only the tip of the normative iceberg, and that norms cannot emerge in society if they do not previously immerge in the mind, i.e. if they are not first converted into mental representations of some sort.
top

Rogier de Langhe (Ghent University): From Theory Choice to Theory Search

This paper shifts the focus of rational theory choice from selecting the best among a given set of theories to finding a balance between selecting among given theories and searching for new ones. Theory choice is studied by building an agent-based model of rational individual scientists facing uncertainty in choosing between the exploitation of an existing theory or the creation of a new one. I show that the local interactions of rational scientists balancing the exploitation and exploration of theories results in a remarkably robust pattern characterized by a succession of tradition-bound periods punctuated by non-cumulative breaks.top

Corinna Elsenbroich (University of Surrey) and Rainer Hegselmann (University of Bayreuth): Agent-Based Models in Moral Philosophy

We will demonstrate that agent based modelling (ABM) is an extremely fruitful approach to analyse processes of moral dynamics. In our talk we present two models giving evidence for the addition of ABM to the repertoire of philosophical methods. The first model implements moral decision making explicitly in the form of team-reasoning, the second model uses ABM for a systematic understanding of the evolution of morality. We show that ABM can help the systematic investigation of moral scenarios as well as the elucidation of philosophical theories.top

Remco Heesen (Carnegie Mellon University): Three Ways to Become an Academic Superstar

Using a formal model, I show that superstars arise quite naturally in a scientific community where scientists are motivated only by epistemic goals, in particular a desire for information. This proves that one cannot infer that outside forces or scientists’ biases have disrupted the process of science, merely from the fact that superstars exist. If luck is introduced as an additional factor, being a superstar no longer necessarily reflects well on the epistemic virtues of a scientist. This has implications for funding agencies and the way they use measures of prominence (e.g., citation metrics) to give out grants and awards.top

Aidan Lyon (University of Maryland/MCMP) and Eric Pacuit (University of Maryland): Deliberating in a Prediction Market

It is well-known that deliberation, when appropriately structured and moderated, can improve group judgements, and therefore group decisions. It is also well-known that prediction markets have many advantages when it comes to forming group judgements. Deliberation groups and prediction markets are very different methods for determining group judgements, and they exhibit different beneficial properties. We may wonder, then, what happens if we were to combine them? Can we take what is best from a deliberation group and combine it with what is best from a prediction market? We present a series of agent-based simulations, in which we embed deliberation groups into prediction markets to study the effect that deliberation has on the market. We also study the effect that the prediction market has on the deliberation groups.top

Rebecca MacIntosh (University of Western Ontario): Cooperative Confusions

This discussion focuses on the contributions to evolutionary game theory made by Brian Skyrms and, in particular, on his claims about the cultural evolution responsible for our norms of justice and fairness. The move from just behaviours to our concept of justice is undefended by Skyrms. I argue that Skyrms’ models are not suitable representations of either the origins of our moral attitudes, nor how moral norms are transmitted within populations. Finally, I draw from multiple theories of the evolution of morality to construct a working definition of moral attitudes that could be useful to future modeling of human evolution.top

Manolo Martinez (Universitat Autònoma de Barcelona): Signalling Games and Modality

Although we are able to secure knowledge about everyday modal claims---we are, that is, able to *modalise* (Van Inwagen 1998) just as we are able to perceive---, this is taken to be a puzzling fact. In particular, it is widely held that it is unhelpful to model our epistemic access to the modal realm on the basis of perception, and maintain that there is a bodily mechanism attuned to the modal aspects of things. In this paper I defend modalising mechanisms. I show how populations of (perfectly non-intentional) senders and receivers in a Lewis-Skyrms signalling game evolve to stable configurations in which modal information is communicated. The process by which this happens, and its end state, are thoroughly naturalistic.top

Conor Mayo-Wilson (University of Washington/MCMP) and Kevin Zollman (Carnegie Mellon University/MCMP): Modelling the Evolution of Social Contract and the Stag Hunt

In this two-part tutorial, participants will learn the basic syntax of NetLogo, which is a programming language designed specifically for the construction of agent-based models. The instructors will provide participants with a basic implementation of an agent-based model that has been used widely to explain the evolution of cooperation, namely, a model in which the Stag Hunt is played repeatedly on a network. See Skyrms' Evolution of the Social Contract, in particular, for a discussion of the importance of this model. Participants in the tutorial will then learn different features of NetLogo's syntax by progressively generalizing and modifying the model. In part one of the tutorial, Mayo-Wilson will teach participants how to modify parameters in the model, including payoffs in the game, the size and the structure of agents' neighborhoods, etc. In part two, Zollman will teach participants how to implement alternative learning algorithms and discuss games on networks.top

Cailin O'Connor (University of California, Irvine): Evolving to Generalize: Trading Precision for Speed

Biologists and philosophers of biology have generally agreed that learning rules that do not lead to evolutionarily stable strategies (ESSes) will not be evolutionarily successful. This claim, however, stands at odds with the fact that learning generalization---which cannot lead to ESSes when modeled in games---is ubiquitous. I show that generalization, despite leading to suboptimal behavior, can allow actors to learn quickly. I observe that previous analyses of the evolution of learning problematically ignored the short-term success of learning rules. If one drops this assumption it can be shown that learning generalization will evolve in many cases.top

Scott Page (University of Michigan): Collective Accuracy: Agent Based & Emergent vs Statistical and Assumed

In this talk, I describe two broad classes of models that can explain collective accuracy, what is more commonly referred to as the wisdom of crowds. The first model is based on statistical/law of large numbers logic. Accuracy emerges from the cancellation of random errors. The second model has roots in computer science and psychology. It assumes that predictions come from models. Different predictions arise because of different model. I then describe how in agent based models the amount model diversity, and therefore the accuracy of the collective emerges. It is possible to write difference equations that explain average diversity levels. The talk will summarize papers written with Lu Hong, Maria Riolo, PJ Lamberson, and Evan Economo.top

Samuli Pöyhönen (University of Helsinki): Navigating an Epistemic Landscape: Foraging vs. Broadcasting as Models of Socially Distributed Problem Solving

The paper discusses the epistemic landscapes model of the division of cognitive labor put forward by Michael Weisberg and Ryan Muldoon. I identify three problems having to do with the model itself and its application to understanding the social epistemology of science: (i) the diversity effect identified by the authors might not be as robust as suggested, (ii) the search methods implemented in the model might not work on rugged landscapes, and (iii) epistemic significance is not always a plausible measure of problem-solving efficiency. Based on an approach similar to the epistemic landscapes model, I propose a model of the division of cognitive labor called the broadcasting model.top

Alexander Reutlinger and Dominik Hangleiter (Munich Center for Mathematical Philosophy): Agent-based Simulations in the Sciences: Explanation without Understanding

In the literature on scientific explanation, it is majority view that explaining a phenomenon P always coincides with understanding P (Lambert and Schurz 1994, De Regt and Dieks 2005, Strevens forthcoming, De Regt et al. 2009). This view is sometimes referred to as the ‘symmetry thesis’ regarding explanation and understanding. The main claim of this talk is that agent-based computer simulations (henceforth, ABS) are a scientifically motivated counterexample to the symmetry thesis, since ABS are explanatory but fail to provide understanding. It is argued that ABS fail to provide understanding, because ABS are, in a sense to be explained, ‘opaque’.top

Felipe Romero (Washington University in St. Louis): The Fragility of Scientific Self-Correction

Can science correct its mistakes? Defenders of the self-corrective thesis answer affirmatively, arguing that scientific method will refute false theories and find closer approximations to the truth in the long run. I discuss a plausible interpretation of this thesis that philosophers have defended in terms of frequentist statistics. Using agent-based simulations, I argue that such an interpretation is true only under idealized conditions that are hard to satisfy in scientific practice. In particular, I show how some features of the social organization of contemporary science make the long run performance of frequentists statistics fragile. I suggest that we have to pay attention to the relation between inference methods and the social structure of science in our theorizing about scientific self-correction.top

Carlos Santana (University of Pennsylvania): Modeling the Evolution of Language with and without much Empirical Data

Many scientists doubt that we can do meaningful research on the evolution of language, given the paucity of empirical evidence. Others argue that agent-based models are especially well-suited to address this problem, since they are a powerful research tool which isn't reliant on detailed empirical evidence. By examining one by one at the variety of different ways researchers use empirical evidence in tandem with ABMs, I show that neither blanket pessimism nor blanket optimism is warranted. On our way to this conclusion we can draw some general lessons about the role of independent empirical data in agent-based modeling in general.top

Hannah Übler (MCMP/LMU): Simulating Trends in Artificial Influence Networks

We present a study of the spreading of trends in artificial social influence networks using agent based models. We concentrate on basic properties of the agents which describe their individual attitudes towards a trend, as well as the influence which they exert in their social neighbourhood. Using two different approaches, we investigate the impact of network dynamicity, situations of opposing trends, and the disappearance of trends. Besides, we study the impact of group cohesiveness and connectors in social communities.top

Krist Vaesen (Eindhoven University of Technology), Wybo Houkes (Eindhoven University of Technology) and Adrien Querbes-Revier (Carnegie Mellon University): No Strength in Numbers after all? A Robustness Analysis of the ‘Demographic’ Effect on Scientific and Technological Change

Cultural-evolutionary models of cultural change enjoy growing popularity. This family of mathematical and agent-based models purportedly explains how instances of scientific and technological change (e.g., the Upper-Palaeolithic transition and scientific progress since the Industrial Revolution) result from a ‘demographic’ effect: complex traits can accumulate in large groups, and disappear in smaller groups. We reveal hidden contingencies in these findings. They rely on assumptions regarding social learning mechanisms and skill distributions, and particular definitions of complexity. Empirical evidence is too scarce to rule out alternatives. We conclude with a call for and some cautionary remarks about robustness analyses of cultural-evolutionary models.top

Michael Weisberg (University of Pennsylvania): Agent-based Models and Confirmation Theory 

Is it possible to develop a confirmation theory for agent-based models? The are good reasons to be skeptical: Classical confirmation theory explains how empirical evidence bears on the truth of hypotheses and theories, while agent-based models are almost always idealized and hence known to be false. Moreover, classical ideas about confirmation have been developed for relatively simple hypotheses, while even the simplest agent-based models have thousands of variables. Nevertheless, we can draw on ideas from confirmation theory in order to develop an account of agent-based model confirmation. Theorists can confirm hypotheses about model/world relations, and they can also use a variety of techniques to investigate the reliability of model results. This paper is an exploration of these possibilities.top

Elena M. Tur (INGENIO, Valencia), Paolo Zeppini (University of Amsterdam) & Koen Frenken (Utrecht University): Diffusion of Ideas, Social Reinforcement and Percolation

This paper analyzes how social structure and social reinforcement affect the diffusion of an idea in a population of human agents. A percolation approach is used to model the diffusion process. This framework assumes that information is local and embedded in a social network. We introduce social reinforcement in the model by softening the condition to adopt when the number of adopting neighbors increases. Our numerical analysis shows that social reinforcement severely affects the output of the process. Some ideas with an original value so low that it would never get diffused can be spread due to the strength of social reinforcement. This effect also interacts with the structure of the network, with a more sizeable impact on small worlds with a low rewiring probability. Also, social reinforcement completely changes the effect of clustering links, because sequential adoption of neighbors can make one agent adopt at later stages.top

Kevin Zollman (Carnegie Mellon University): The Formation of Epistemic Networks

One important area of study for social epistemology is the social structure epistemic groups -- who communicates their knowledge with whom? Significant research has been done on better and worse communication networks, but less has been done on how a group comes to have one network or another. In this talk, I will present a number of results (some recent) from economics and philosophy about how individuals choose with whom to communicate. Understanding how individuals decide where to gain information can help us to design institutions that lead to epistemically more reliable groups.