著者

守 真太郎 久門 正人 高橋 泰城

出版者

横断型基幹科学技術研究団体連合(横幹連合)

雑誌

横幹連合コンファレンス予稿集 第8回横幹連合コンファレンス

巻号頁・発行日

pp.B-1-1, 2017 (Released:2018-02-18)

---

Collective intelligence is a mechanism to aggregate information held by many agents and generate useful information. In order for the generated information to be useful, it is a sufficient condition that the agents should independently generate information to be aggregated. On the other hand, in the case of humans, as they give priority to the return of each individual, not to the accuracy of the collective intelligence, they often use social learning and imitate the choices and actions of others. Independence of information to be aggregated is not preserved and the collective intelligence mechanism does not function. We conducted an information cascade experiment using two-choice general knowledge quizzes. When quizzes are easy, collective intelligence functions and the correct answer rate improves. However, when the quizzes are difficult, it becomes random whether the correct answer rate improves or deteriorates This change is a phase transition due to a change in the number of stable states of nonlinear Polya urn process. The essence of the phase transition is whether the in uence of the rst subject's choice (Domino effect) percolates or not in the innitely long subjects' sequence.