SOCIAL
Social Game Theory
Evolutionary simulation environments such as the Cultaptation game have been developed in order to investigate the phenomenon of social learning (learning from observations of the behavior of others), which is an important capability in many animal species, especially humans.
We have developed an algorithm to compute near-optimal strategies for Cultaptation, and have shown that the strategies produced by our algorithm are near-optimal, both in their expected utility and their expected reproductive success. Our algorithm can be used to provide insight into evolutionary conditions under which learning is best done by copying others, versus the conditions under which learning is best done by trial-and-error.
In addition, several of our algorithmic techniques should also be useful in a variety of evolutionary games other than just the Cultaptation game. These include a way to compute near-optimal strategies with a finite-horizon search rather than searching all the way to the end of the game, and a way to compute strategies over sets of states rather than individual states, thereby reducing the computation time by an exponential factor without any loss in accuracy of the computed strategies.
The classical game-theoretic model of "rational" agents assumes that such agents will maximize their own expected utility, but there is much empirical evidence that human decision-making does not coincide with that assumption. This is due, at least in part, to influences such as evolutionary history, the physical environment in which one is acting, and one's social orientation and tendencies. We are looking at agent models that include various of these factors, with the objective of shedding some light on real world behavior and experiences.
For example, our recent results show that there are evolutionary environments in which agents with state-dependent risk preferences (i.e., agents that are sometimes risk-prone and sometimes risk-averse depending on the outcomes of their previous decisions) have an evolutionary advantage over agents that make decisions purely by expected-utility maximization.
Decision making is a core component of any intelligent system. Recently, social decision making is starting to emerge as an important research problem that involves decision making under the conditions of social factors such as trust, confidence, volatility (i.e., relationships that change over time), and social uncertainty. Although there have been made several impressive advances recently in AI decision making, these advances are severely limited by restrictive assumptions that exclude reasoning and inference for decisions over social factors and interactions, and thus, they have not received much practical utility in social decision-making problems.
In this project, we are investigating how social factors such as the ones above can be modeled and made operational in automated inference and planning algorithms. We are working in the context of Semantic Web; our applications range from real Web-based social networks to Semantic Web service composition.
Social interactions with other people is a core component in a person's operations in the world. Most of the actions a person takes depends on and/or affected by his/her interactions with others. The decision making process the person undertakes, what is best or not in the immediate or long-term future usually involves a very complex analysis of those interactions, the underlying meanings and purposes behind the interactions. Our objective is to investigate how people make decisions based on their social interactions in the environment.
Investigations into social interactions has been discussed in the context of virtual worlds, but not well studied in real-time strategy (RTS) games. In addition to their entertaintment value, RTS games have emerged to become virtual platforms that simulate real-world, real-time physics, scenarios, characters, and strategies. Particularly, multi-player online RTS games are providing a new model of a model of human interaction that is in line with decision theory, game theory, planning, learning, and other concepts from research fields such as Computer Science, Artificial Intelligence, Economics, and Behavioral Sciences.
Every day, people and organizations make plans about how to travel and transport commodoties among themselves. These take into consideration the many options the relations among people, constraints, and information from people about which options work best and why. To date, the dynamics of the physical world and trust between users have been largely neglected in systems that help to make plans in these environments.
This project will develop new representation models for decision making with physical resources, constraints, and social characteristics of the events that involve individuals or groups of individuals operating in the world. These models will allow us to develop algorithms that produce knowledge about how decisions based on resources, geospatial constraints, and objectives evolve in time. We will also build models of trust in social networks that change over time, reflecting the dynamics present in the physical space as well as the dynamics of the interpersonal relationships.
Finally, we will combine the two research agendas above into a unifying framework, called Behavior Network Diagrams (BNDs). We believe that BNDs will be the ground work for developing novel algorithms and will generate patterns of behavior of the individuals or groups of individual that explain the evolution of the world, socially, geospatially, and temporally. This approach will be evaluated theoretically, in simulation, and by direct comparison to decision-making strategies produced by users presented with the same scenarios.
Project lead: Dr. Dana Nau.
For additional information, please contact Dr. Dana Nau or Dr. Inon Zuckerman.
Last updated: November 2009 by John Dickerson.
Publication links contingent upon publication!
This page will be updated as our work enters print. For information about receiving draft publications, technical reports, and conference presentations, please do not hesitate to contact team members.
The following sections may include links to restricted access material. Please do not hesitate to contact a group member for instructions regarding how to obtain a username and password.
Presentation links contingent upon publication!
The following sections may include links to restricted access material. Please do not hesitate to contact a group member for instructions regarding how to obtain a username and password.
Software links contingent upon publication!