Modeling knowledge generalization capability in agent to replicate subject experiment result

Toward the agent model that can replicate human-like behaviors this paper aims at investigating a capability of our proposed agent model in terms of (1) knowledge size and (2) practical time for replicating them. For this purpose, we employ our agent that has the capability of (1) selecting which knowledge can be generalized among a lot of knowledge and (2) determining the timing when the selected knowledge should be generalized, and compare the result of our agents with these of agents employing heuristic methods with different knowledge generalization timing. Intensive simulations for comparisons in the sequential bargaining game have revealed the following implications: (1) both knowledge selection and knowledge generalization timing are critical for modeling agents; and (2) the proposed techniques enable agents to replicate the same subject experiment result, i.e., agents replicate it with a small number of knowledge (not sufficient numbers of knowledge) in practical times (the less iterations).