From Santa Fe Institute Events Wiki
- Randy Casstevens, Computational Social Science, George Mason University (firstname.lastname@example.org)
- Vessela Daskalova, Economics, U. of London (email@example.com)
- Ozge Kalkan Dilaver, Economics and Management Science, Lancaster University (firstname.lastname@example.org)
- David Hendry, Political Science, U. of Illinois Urbana-Champaign (email@example.com)
- Coco Krumme, Media Lab, MIT (firstname.lastname@example.org)
- Davide Marchiori, Economics and Management, U. of Trento (email@example.com)
- Miriam Rehm, Economics, New School for Social Research (firstname.lastname@example.org)
- Jennifer Trueblood, Cognitive Science, Indiana U. (email@example.com)
- Olivia Woolley, Engineering and Applied Mathematics, Northwestern University (firstname.lastname@example.org)
- Matt Zimmerman, Ecology, U. of California Davis (email@example.com)
- Willemien Kets, Economics, SFI.
- John H. Miller (co- director), Economics, Carnegie Mellon University and SFI.
- Scott E. Page (co-director), Economics, U. of Michigan and SFI.
- Jon Wilkins, Theoretical Biology, SFI.
Each student began a research project during the two-week workshop. Below are brief descriptions of these various projects. These projects will form the basis for dissertation chapters and/or journal articles.
Randy Casstevens, Computational Social Science, George Mason University (firstname.lastname@example.org)
Randy is exploring the process of innovation in software development. He compares patterns of human-problem solving, as measured by a public MATLAB Programming Contest, and those that arise via evolutionary computation. The MATLAB contest explored programs that could solve a non-trivial peg solitaire game, and it had over 3000 submissions from over 100 programmers. The evolutionary computation problem was the Artificial Ant Problem from the genetic programming community. Both systems display punctuated equilibria in terms of overall performance metrics, had new efforts exploiting recently introduced innovations, and have some interesting similarities concerning measures of program complexity. One important parameter linked to the growth of program complexity is the amount of parsimony pressure applied to the evolving programs.
Vessela Daskalova, Economics, U. of London (email@example.com)
Vessela is investigating the dynamics of discrimination. The two most common explanations for treating individuals differently because they belong to a particular group involve tastes and imperfect information. The model here provides a dynamic account of how discrimination can emerge via categorization. Initially, she considered how an individual learns to categorize using a genetic algorithm, and she found that after relatively short periods of evolution the individuals were able to categorize with relatively low errors (around 3% mis-categorization). She is now implementing a social interaction framework where a population of agents use categorization to decide how to play a Prisoner's Dilemma game against one another. Agents are able to share information (with those whom they have cooperated in the past) about their past experience with others. Extensions include endogenizing the number of categories in the model and incorporating more formal social networks.
Ozge Kalkan Dilaver, Economics and Management Science, Lancaster University (firstname.lastname@example.org)
Ozge is looking at patterns of technological adoption by consumers. She focuses on issues that surround technological diffusions in cases where there is the potential for decreasing returns to adoption, such as medical procedures (like cosmetic surgery), business practices (for example, 24/7 availability), and so on. In these cases, a social dilemma compels individuals to adopt the innovation even though this may eventually lead to a bad social outcome. The model incorporates a social network with preferential attachment along with institutions, such as norms, that can influence adoption via ties to various network points. For some products, like fashion, the adoption pattern follows a "W" curve with successive waves of adoption. As institutional behavior and network links are altered, systematic changes in the resulting adoption patterns can be observed.
David Hendry, Political Science, U. of Illinois Urbana-Champaign (email@example.com)
David is pursuing a model of how race and policy preferences can be altered through interpersonal contact. The core issue is whether such individual-level contact can aggregate sufficiently at the societal level to alter electoral outcomes. Agents of two races are situated in a social network, and have policy preferences (both ideal positions and saliences) tied to race. Agents continually update their salience based on a weighted-average of those to whom they are connected. His initial findings suggest that variations in network structure and sampling strategy can predict whether various preference enclaves can persist. For example, in networks characterized by preferential-attachment and elite sampling, numerous enclaves can arise and survive for long periods, subject to a long term decrease in overall salience levels for the racial issues.
Coco Krumme, Media Lab, MIT (firstname.lastname@example.org)
Coco is focused on how groups respond to rare events. Studies have shown that humans tend to fall prey to various probabilistic-thinking foibles, such as the gambler's fallacy. She develops a framework that has networked agents playing a simple game that has a low, safe payoff regardless of the state of the world, and a typically high payoff that may turn quite negative on rare occasions. She finds that risk taking can be linked to various learning rules, memory lengths, wealth distributions, and so on. These initial results suggest that carefully designed public policy may allow social systems to achieve appropriate levels of risk seeking.
Davide Marchiori, Economics and Management, U. of Trento (email@example.com)
Davide is using an adaptive model based on counter-factual reinforcement learning to explore the dynamics of equilibrium selection in games. Based on psychological observations about human behavior, he has developed a notion of equilibrium relying on the net reward that a player receives when she takes an action. Using an adaptive procedure linking a simple updating rule that incorporates the net reward to a counter-factual---defined by what would have happened in terms of the net reward if alternative actions were taken---he tests whether this dynamic system will converge on the proposed equilibrium. He finds, in a sample of games, that this adaptation does lead to a rapid convergence to the expected net reward equilibria, and that when any one of the proposed model elements are eliminated the system tends to wander away from the expected equilibria.
Miriam Rehm, Economics, New School for Social Research (firstname.lastname@example.org)
Miriam is refining our understanding of migration and remittances. There are a number of theories about what drives remittances, including self-interest, altrusim, co-insurance, and loans, and it has been difficult to discriminate among them. Each of these theories can be easily incorporated into an agent-based model. The next stage of the project explores the impact of these various theories on the behavior of the models. These explorations allow new perspectives on one's ability to discriminate among the theories and on the potential for new empirical methods to be used to understand more fully the behavior of this system. Ultimately the model will be calibrated against data from Ecuadorian migrants coming to New York.
Jennifer Trueblood, Cognitive Science, Indiana U. (email@example.com)
Jennifer wants to understand the origins of risk aversion as an adaptive strategy in uncertain environments. A well-known bias of human behavior is to be risk avoiding when facing gains, while risk seeking when confronting losses. In her model agents interact with an environment to gain some needed resource. At every time period agents must choose which of two habitats to explore. Agents use a two-state Markov chain to embody behavior over the two habitats. Using a genetic algorithm, she evolves the Markov chain under various environments. By altering the underlying parameters governing the uncertainty in the two habitats, she is developing a new understanding of how such uncertainty influences agent behavior.
Olivia Woolley, Engineering and Applied Mathematics, Northwestern University (firstname.lastname@example.org)
Olivia models the emergence of inequality in hierarchical structures. There are two dimensions of power in the model: economic and political. In her view hierarchy emerges when these two dimensions become coupled. The model consists of two, partially overlapping networks---an invariant social one and a varying economic one that imposes costs and benefits on whatever links are formed. Agents get value from economic connections and can bargain, using an ultimatum approach, over the potential surplus created by linking with neighbors. They adapt their bargaining strategies using a reinforcement mechanism. She finds that offer prices are much less stable than ask prices, and that feedback appears to be closely linked to the emergence of economic networks.
Matt Zimmerman, Ecology, U. of California Davis (email@example.com)
Matt is refining our understanding of culture and warfare. The core questions driving this work are how did warfare evolve and how does culture influences warfare decisions. In the model, individual-level interactions are tied to a Prisoner's Dilemma while group-level interactions embrace a Hawk-Dove game. Agents are spatially situated, with groups determined by neighborhoods. Agents have binary tags, and groups make warfare decisions depending on the Hamming distance between the tags linked with an evolving strategic bias. Early results indicate that parochialism evolves only when the tags are relatively stable. He is currently extending the model to incorporate more heterogenity across within group tags, migration, and biased transmission.