Environmental volatility amplifies Pavlovian control in decision-making

Abstract

Volatile environments require continuous updating of action values, making instrumental knowledge less informative. Under such conditions, Pavlovian control may offer a more stable alternative, since the value attributed to reward-predictive cues is not altered by continuously updated instrumental predictions. Using a modified Pavlovian-to-Instrumental Transfer paradigm in two preregistered experiments, we test the hypothesis that increased volatility promotes the recruitment of Pavlovian control, whereby reward-predictive cues bias choice to the predicted outcome, simplifying (and possibly overcoming) the high computational cost of the instrumental prediction. Crucially, using entropy-based metrics we demonstrate that Pavlovian control is not unitary, but can instead respectively increase or reduce entropy depending on whether the choice outcome is present or not. Together, these findings show that volatility redirects control in Pavlovian-instrumental arbitration toward the cues, rather than merely adding noise to behaviour.