The utility of a latent-cause framework for understanding addiction phenomena

Abstract

Computational models of addiction often rely on a model-free reinforcement learning (RL) formulation, owing to the close associations between model-free RL, habitual behavior and the dopaminergic system. However, such formulations typically do not capture key recurrent features of addiction phenomena such as craving and relapse. Moreover, they cannot account for goal-directed aspects of addiction that necessitate contrasting, model-based formulations. Here we synthesize a growing body of evidence and propose that a latent-cause framework can help unify our understanding of several recurrent phenomena in addiction, by viewing them as the inferred return of previous, persistent "latent causes". We demonstrate that applying this framework to Pavlovian and instrumental settings can help account for defining features of craving and relapse such as outcome-specificity, generalization, and cyclical dynamics. Finally, we argue that this framework can bridge model-free and model-based formulations, and account for individual variability in phenomenology by accommodating the memories, beliefs, and goals of those living with addiction, motivating a centering of the individual, subjective experience of addiction and recovery.

Keywords: addiction; craving; latent-cause inference; relapse.

Figure 1.. The latent-cause framework and its utility in understanding craving and relapse.

a-b. Differences between model-free RL (left column) and the latent-cause framework (right column). a. Outcome specificity of associations and craving states: Two outcome identities, O₁ and O₂, are paired with stimuli S₁ and S₂, respectively (middle). Left: Model-free RL abstracts away the outcome identities in favor of scalar values V (grey bars). Right: latent causes labelled C₁ and C₂ also represent abstractions, however, these are of both stimuli (cues) and outcomes, and retain outcome-specific expectations (colored bars) despite being “defocused” and generalizable, much like craving states. b. Asymmetry of learning and unlearning and vulnerability to relapse: In extinction, O₁ that was initially paired with S₁, no longer appears (middle). Left: In model-free RL, learning and unlearning are symmetric, such that the value of S₁ first increases (when paired with O₁) and then decreases in extinction (the first trial of extinction is marked by a dotted line). Right: In the latent-cause framework, the large change between acquisition and extinction can entail the formation of a new latent cause (pink) that is associated only with S₁ and not with O₁, leaving previously learnt knowledge (blue) intact, and therefore prone to relapse. c. Dynamics of latent-cause inference, craving and relapse: Simulation of the timecourse of latent-cause inference in reversal learning, showing potential connections to addiction phenomena. Early experience with an addictive substance leads to robust learning of drug associations and a strong policy of choosing the consumption action A₁ (square), bound to a latent cause with strong expectations of reinforcing outcomes (blue). Following reversal (first dashed line), the act of consumption no longer gives rise to the same level of reinforcement, either due to tolerance (reduction in the hedonic value of the drug outcome) or other negative consequences of drug use, leading to abstinence and a drop in outcome expectations. This sudden change leads to the creation of a new latent cause (pink) that is not as strongly bound to the action policy or its outcome, and this new cause persists for some time. As time passes, exhaustion of this persistence or exposure to drug-related cues or contextual triggers increase the probability of inferring the return of the original drug-associated cause (blue). This leads to an increase in outcome expectations, resembling incubation of craving. Eventually, outcome expectations and/or craving may grow so large as to trigger a relapse event (second dashed line), resetting the cycle.