Oversampled and undersolved: Depressive rumination from an active inference perspective

Rumination is a widely recognized cognitive deviation in depression. However, researchers have struggled to explain why patients cannot disengage from excessive rumination, although it clearly depresses their mood and fails to lead to effective problem-solving. In this integrative review, we connect insights from cognitive science, neuroscience and computational psychiatry to rethink rumination as repetitive but unsuccessful mental problem-solving attempts. In particular, appealing to an active inference account, we first suggest that adaptive problem-solving is based on the generation, evaluation, and performance of possible behaviors (candidate policies) that increase epistemic value and minimize prediction errors. Next, we discuss how this problem-solving algorithm is distorted in depression. Specifically, we propose that depressive rumination can be understood as engaging in excessive yet unsuccessful Bayesian sampling of candidate behaviors that do not resolve uncertainty. Further, as policy candidates are sampled from policies that were selected in emotional states resembling the current one, similar candidates are re-sampled repetitively, resulting in a time-consuming feedback-loop, a ruminative „halting problem”. This leads to high opportunity costs, because potentially better policy candidates are not selected, as well as to learned helplessness as problem-solving attempts remain unsuccessful. Besides reviewing the evidence for the conceptual paths of this model of depressive rumination, we discuss the evidence for its neurophysiological correlates, particularly with respect to distorted connectivity patterns of different brain networks, such as an overactive default mode network. We conclude by suggesting future research directions into behavioral and biological features of our model and highlight clinical implications.