...
Colored Trails is a research test-bed for investigating decision-making in groups comrising people, computer agents, and a mix of these tootwo.
Motivation
Widespread use of the Internet has fundamentally changed the ways in which we use computers, not only as individuals, but also within organizations. Settings in which many people and many computer systems work together, despite being distributed both geographically and in time, are increasingly the norm. Examples of group activities in which computer systems participate, whether as autonomous agents or as proxies for individual people, include online auctions, hospital care-delivery systems, emergency response systems, military systems, and systems administration groups. These increasingly prevalent, heterogeneous group activities of computer systems and people-- whether competitive, cooperative, or collaborative--frequently require decision-making on the part of autonomous-agent systems or the support of decision-making by people.
...
Grosz and Kraus developed the game Colored Trails (CT) as a testbed for investigating the decision-making that arises in task settings, where the key interactions are among goals (individual and group), tasks required to accomplish those goals, and resources needed to perform the tasks. CT allows the modeling of all of these phenomena and exploration of their causes and ramifications. It provides the basis for development of a testbed that supports investigations of human decision-making and comparisons among for computational strategies is needed, one that would enable human decision-making to be studied both in homogeneous groups comprising only people and in heterogeneous groups consisting of people and computer system and for computational strategies to be studied both in settings where computational agents interact only with other such agents and in heterogeneous settingssystems.
The proposed project would convert the existing code-base to make it robust and available for distribution to other researchers, extend the range of decision-making scenarios which could be modeled and tested using CT, and perform machine-learning research and testing of computerbased decision-making strategies in increasingly complex situations. The structure of CT provides the right basis for development of a decision-making testbed. The rules of CT are simple, abstracting from particular task domains, and thus enabling investigators to focus on understanding, modeling, and testing decision-making strategies rather than specifying and reasoning with domain knowledge. It is simpler to implement computer agents to play CT than games like Diplomacy. Nonetheless, CT has proven interesting for people to play. In its abstracting from "realworld" domains, CT is similar to the games developed in behavioral economics. However, CT abstracts less than typical economic games do. In particular, unlike behavioral economics games, it provides a clear analogue to task settings, and it provides situational contexts and interaction histories in which to make decisions. Thus, it takes into account the problems of framing effects (for example, the difference between describing a prisoner's dilemma game as "the Wall Street Game" and "the Community Game"), but does not reduce all decisions to choices between explicitlygiven utility values. As a result, some of the social factors that are known to influence human decision-making (e.g., inequality aversion, reciprocity) may be directly studied in CT, providing a better basis for the design of computational agents.
...