Interactive Restless Multi-armed Bandit Game and Swarm Intelligence Effect
We obtain the conditions for the emergence of the swarm intelligence effect in an interactive game of restless multi-armed bandit (rMAB). A player competes with multiple agents. Each bandit has a payoff that changes with a probability pc per round. The agents and player choose one of three options: (1) Exploit (a good bandit), (2) Innovate (asocial learning for a good bandit among nI randomly chosen bandits), and (3) Observe (social learning for a good bandit). Each agent has two parameters (c, pobs) to specify the decision: (i) c, the threshold value for Exploit, and (ii) pobs, the probability for Observe in learning. The parameters (c, pobs) are uniformly distributed. We determine the optimal strategies for the player using complete knowledge about the rMAB. We show whether or not social or asocial learning is more optimal in the (pc, nI) space and define the swarm intelligence effect. We conduct a laboratory experiment (67 subjects) and observe the swarm intelligence effect only if (pc, nI) are chosen so that social learning is far more optimal than asocial learning.
KeywordsMulti-armed Bandit Swarm Intelligence Interactive Game Experiment Optimal Strategy
Unable to display preview. Download preview PDF.
- 10.Rendell, L. et al., “Why copy others? Insights from the social learning strategies tournament,” Science, 328, pp. 208–213, 2010.Google Scholar
- 11.Sutton, R. S. and Barto, A. G., eds., Reinforcement Learning: An Introduction, Cambridge, MIT Press, 1998.Google Scholar
- 12.Toyokawa, W., Kim, H. and Kameda, T., “Human collective intelligence under dual exploration-exploitation dilemma,” PLoS ONE, 9, p. e95789, 2014.Google Scholar
- 13.White, J. M., Bandit Algorithms for Website Optimization, O’Reilly Media, 2012.Google Scholar