Counter intuitive learning: An exploratory study

Nobuyuki Hanaki (1) , Alan Kirman (2) , Paul Pezanis-Christou (3)

1 GREDEG - Groupe de Recherche en Droit, Economie et Gestion
2 CAMS - Centre d'Analyse et de Mathématique sociales
3 University of Adelaide

The literature on learning in unknown environments emphasises reinforcing on actions which produce positive results. But, in some cases, success requires shifting from a currently successful actions to others. We examine, experimentally and theoretically in a very simple framework, how individuals initially learn by exploiting information from the pay-offs of actions taken but also from exploring new actions. We analyse if and how they learn that pay-offs are inter-temporally dependent. We then ran the same experiments but where individuals could observe the actions taken or the pay-offs obtained by others or both. Such observations improved pay-offs if one of the pair had learned to obtain the maximum pay-off.

Mots clés: multi-armed bandit; reinforcement learning; eureka moment; pay-off patterns; observational learning

Domaines: Economies et finances; Psychologie

Nobuyuki Hanaki, Alan Kirman, Paul Pezanis-Christou. Counter intuitive learning: An exploratory study. 2016. ⟨hal-01358716⟩