Reinforcement learning relieves the vaccination dilemma

The main goal of this paper is to study how a decision-making rule for vaccination can affect epidemic spreading by exploiting the Bush-Mosteller (BM) model, one of the methodologies in reinforcement learning in artificial intelligence (AI), which can realize the systematic process of learning in humans, on complex networks. We consider the BM model with two stages-vaccination and epidemiological processes-and address two independent rules about fixed loss consideration and average payoff of neighbors to update agent's vaccination behavior for various stimuli, such as loss of payoffs and environments during the vaccination process. Higher sensitivity not only favors higher vaccination coverage rates but also delays the transition point in relative vaccination costs when transitioning from full vaccination (inoculation level 1) to incomplete vaccination (inoculation level less than 1). Extensive numerical simulations demonstrate that the vaccination dilemma can be overcome to some extent, and the distribution of the intended vaccination probabilities in both independent rules is either normal or skewed when different parameters are considered. Since AI is contributing to many fields, we expect that our BM-empowered learning can ultimately resolve the vaccination dilemma.