A study on the nature of humanfs decision making strategy in a dilemma situation

350803102 Akihiro KUNIMATSU

Abstract

Many studies have been conducted on decision making, and many researches have used the Prisonerfs Dilemma (PD) game, even though the game requires a simple choice of hCooperateh and hDefectionh. Axelrod(1980a) investigated an effective strategy in the Repeated PD game, and found TFT (Tit For Tat) as the most effective strategy in this game. Our goal was to understand the nature of humanfs decision making in the Repeated PD game, and examine the relationship between human strategies and normative strategies, such as TFT, that were examined in the previous PD researches. Especially, we focus on whether humans take action like normative strategies like TFT or not. We used three approaches to investigate this problem. First, we built a decision making model (H model) based on the data of a human experiment conducted by Hirose (2008). The 2nd-Order Markov chain model was chosen as the H model based on the analysis of concordance rate of the human experimental data and the computer simulation. This result indicates that humans were very sensitive due to the reaction of the otherfs. Second, we classified 15 normative strategies and examined which model fits the best to the H model. As a result, we found that the H model belonged to the same category as TFT and JOSS. This result implies that humanfs decision making strategy has a close pattern to TFT and JOSS. Finally, we investigated TFT and JOSS based on the reproduction of the experimental data. We used the concordance rate and total defection count as main indexes. The results of the concordance rate indicated that both TFT and JOSS were better than the baseline performed by the H model. However, the results of the total defection count indicated that JOSSfs performance was close to the results of humans rather than TFT. JOSS is a strategy that takes an action extremely similar to TFT. The difference between JOSS and TFT is that JOSS eases Nice Rule. Nice Rule means not to choose hDefectionh before the other chooses defect. In conclusion, we suggest that humanfs decision making is close to TFT, but Nice Rule is liberalized.