Computational Mechanisms of Addiction: Recent Evidence and Its Relevance to Addiction Medicine

AbstractPurpose of ReviewIn this article, we provide a brief review of recent computational modelling studies of substance use disorders (SUDs), with a focus on work published within the last 5  years. While reinforcement learning (RL) approaches are most prominent in recent studies, we also review work from other perspectives that focus on Bayesian (active) inference and perceptual processing.Recent FindingsRecent work in RL shows evidence that goal-directed (model-based) planning processes are impaired in SUDs, leading to impulsive, habitual decision processes focused on short-term reward despite long-term negative consequences. Bayesian approaches offer a complementary perspective, suggesting that drug-induced overconfidence in prior expectations prevents substance users from appropriately updating their beliefs in the face of negative outcomes. Recent neurocomputational studies have shown promise in differentiating those who will and will not relapse.SummaryComputational modelling has made progress in identifying specific prospective decision-making processes that are impaired in SUDs, but further research is necessary before these approaches can directly inform medical practice on an individualized level.

Source: Current Addiction Reports - October 1, 2021 Category: Addiction Source Type: research