We first consider evidence for a common notion: the pleasures and pains that we feel act like the reward function of a reinforcement learning agent /2
What computations about reward, then, may emotions represent? The literature suggests three computations, concerning: (i) Expected reward. eg sciencedirect.com/science/articl… (ii) Evaluation of actions. eg psycnet.apa.org/record/2021-84… (iii) Uncertain prospects. eg sciencedirect.com/science/articl… /3
Examining research on the causes and consequences of different emotions suggests the three computations can directly map onto distinct classes of emotions /4
The cumulative results of each of these computations may be represented as moods, such as a depressed, irritable, or anxious mood. These then bias subsequent computations in the same direction (much like Bayesian priors). /5
We conclude that emotions and moods are integral to how biological agents learn and make decisions. They are involved in almost every facet of human learning and decision making, and they can and should be studied in animals as well. /6
Missing some tweets in this thread? Or failed to load images or videos? You can try to .