Social learning by means of classifier systems and the provision of a public good

The conventional approach on public goods provision stresses the inability of voluntary mechanisms to induce cooperation in a social group. However experiments and reality contrast with this sharp prediction. In this paper we develop a setup of social cooperation toward the provision of a public good, in which individuals exhibit bounded rationality that arises as the outcome of the cognitive and informational limitations. We devise the collective decision-making process as an evolutionary game, in which artificially intelligent players search for an optimal action at every stage of the game. Each agent is modeled as a classifier system, which is a set of rules or actions with their associated strength or fitness for a given state of the system. Individuals' strategies evolve according to their relative fitness, although some experimentation is allowed. Computational experiments show that, for a wide range of discount rates, individuals tend to positive contributions which lead to the voluntary provision of the public good. Although the overall provision is suboptimal, it outperforms that of conventional approach, i.e. Nash equilibrium.