Schedule of Events | Symposia

Behavioral and Neural Correlates of Learning Predictable Rules Intermixed with Random Reinforcement

Poster Session B - Sunday, March 30, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Leo Yuhao Jin¹ (yj2525@columbia.edu), Greg Jensen², Jacqueline Gottlieb¹, Vincent Ferrera¹; ¹Columbia University, ²Reed College

Animals are constantly learning novel predictable relationships to better adapt to their environment. However, such “learnable” associations are often intermixed with coincidental stimulus-outcome pairings that are “unlearnable”. It is advantageous for organisms to differentiate these scenarios so that cognitive effort is not wasted on unlearnable relationships. Here, we ask whether animals are capable of making this distinction and how neural activity in dorsal anterior cingulate (dACC) supports this behavior, We exposed monkeys (N=2) to two pictorial sets: a “learnable” set in which the stimuli were implicitly ordered and the rule was always to choose the higher-rank stimulus, and an “unlearnable” set in which stimuli were unordered and feedback was random regardless of the choice. Monkeys learned the ordered list. However, their responses to the unlearnable set were clearly divided into two categories: they either showed a consistent, non-random subjective preference ordering or they responded randomly. To investigate the neural basis of this behavior, we recorded neural activity from dorsal anterior cingulate (N=921) using multi-channel electrodes. GLM models and population decoding showed that dACC activity was strongly modulated by learnability, reward, and their interactions, as well as behavioral choices. Furthermore, the encoding strength correlated with behavior such that when monkeys differentiated learnable and unlearnable lists, the learnability-related modulation of the reward response increased and the modulation by subjective ordering of the unlearnable list decreased. Our results demonstrate that dACC activity is involved with learnability detection and monitoring, which deepens understanding of multi-rule learning and the formation of persistent superstitious biases.

Topic Area: THINKING: Reasoning