Modeling agents based on aspiration adaptation theory

Abstract

Creating agents that realistically simulate and interact with people is an important problem. In this paper we present strong empirical evidence that such agents should be based on bounded rationality, and specifically on key elements from Aspiration Adaptation Theory (AAT). First, we analyzed the strategies people described they would use to solve two relatively basic optimization problems involving one and two parameters. Second, we studied the agents a different group of people wrote to solve these same problems. We then studied two realistic negotiation problems involving five and six parameters. Again, first we studied the negotiation strategies people used when interacting with other people. Then we studied two state of the art automated negotiation agents and negotiation sessions between these agents and people. We found that in both the optimizing and negotiation problems the overwhelming majority of automated agents and people used key elements from AAT, even when optimal solutions, machine learning techniques for solving multiple parameters, or bounded techniques other than AAT could have been implemented. We discuss the implications of our findings including suggestions for designing more effective agents for game and simulation environments.We would like to acknowledge the many productive conversations and visits with Prof. Reinhard Selten, Dr. Martin Hohnisch, and Sabine Pittnauer of the University of Bonn. We would also like to thank the anonymous reviewers for their comments, as well as

Problem restructuring for better decision making in recurring decision situations

“…Furthermore, their solution has many non-intuitive properties. Nevertheless, as evidenced in the results section (and in prior literature [14,69]), both people and the agents they program fail to follow the optimal strategy when engaged in sequential economic search. Supplying the theoretic-optimal solution for the economic search to the searcher is not always feasible.…”

“…First, it can validate or invalidate the effectiveness of different heuristics, which have been proven to be useful in extensive experimentation with agents, when applied to people. The level of similarity between peoples' strategies and agents' strategies is not conclusive; some work suggests a relatively strong correlation between the behaviors of the two [14], while in other work the agents have been reported to act different than people to some extent [26,69]. Second, it is notable that while agents can be programmed only by people with programming skills, the experiments with people as decision makers enable testing the approach with the general population.…”

“…Chalamish et al [13] use Peer-designed agents (PDAs) for simulating parking lots, giving evidence for some similarity between the behaviors exhibited by people and the agents they program. Empirical investigation of the level of similarity observed between PDAs and people in general is not conclusive: some work suggests a relatively strong correlation between the behaviors of the two [14,52,51], while other work reports that PDAs behave differently [26,69].…”

“…Other work has shown that preference elicitation for such decisions is difficult [28]. A variety of work on autonomous agents suggests that even though they do not share people's cognitive limitations and may have sufficient computational capabilities to determine theoretically optimal strategies, they also make suboptimal decisions simplicter, whether as a matter of game theoretic issues [12] or systems design uninformed by theory [69] or both. The increasing presence of agents operating as proxies for people, not only for decisions simpliciter (e.g., bidding on eBay), but also for sequential decision making situations (e.g., algorithmic trading) makes the design of effective agent strategies for sequentially dependent decisions an important multi-agent systems challenge.…”

Toward Adapting Cars to Their Drivers

“…In contrast, in more complex environments a lack of data makes it unfeasible to elicit an accurate model with machine learning alone. In these types of problems we found that using attributes from cognitive models allowed for significantly more accurate models than models created from machine learning or the cognitive models alone (Rosenfeld and Kraus 2009;Rosenfeld and Kraus 2012;Zuckerman, Kraus, and Rosenschein 2011).…”