Colored Trails is a research testbed for analyzing decision-making strategies of individuals or of teams. It enables the development and testing of strategies for automated agents that operate in groups that include people as well as computer agents. The testbed is based on a conceptually simple but highly expressive game in which players, working individually or in teams, make decisions about how to deploy their resources to achieve their individual or team goals. The complexity of the setting may be increased along several dimensions by varying the system parameters. The game has direct analogues to real-world task settings, making it likely that results obtained using Colored Trails will transfer to other domains. We describe several studies carried out using the formalism, which investigated the effect of different social settings on the negotiation strategies of both people and computer agents. Using machine learning, results from some of these studies were used to train computer agents. These agents outperformed other computer agents that used traditional game theoretic reasoning to guide their behavior, showing that CT provides a better basis for the design of computer agents in these types of settings.
[1]
D. Rubin,et al.
Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper
,
1977
.
[2]
Colin Camerer.
Behavioral Game Theory: Experiments in Strategic Interaction
,
2003
.
[3]
Sarit Kraus,et al.
The influence of social dependencies on decision-making: initial investigations with a new game
,
2004,
Proceedings of the Third International Joint Conference on Autonomous Agents and Multiagent Systems, 2004. AAMAS 2004..
[4]
Ya'akov Gal,et al.
Social Preferences in Relational Contexts
,
2004
.
[5]
Ya'akov Gal,et al.
Learning Social Preferences in Games
,
2004,
AAAI.
[6]
Ya'akov Gal,et al.
Adapting to agents' personalities in negotiation
,
2005,
AAMAS '05.