Abstract
In natural environments, associations that indicate true learnable regularities are intermixed with those that arise from random and ultimately unlearnable relationships between events. To efficiently allocate cognitive resources and avoid inferring spurious patterns, organisms must distinguish learnable from unlearnable associations, but the mechanisms underlying this ability are not understood. We recently showed that monkeys performing a transitive inference task, while discovering the true hidden order in learnable image sets, also behaved to varying degrees as if they inferred subjective order in objectively random (unlearnable) image sets. Here, we show that the ability to detect learnability is encoded by neurons in the dorsal anterior cingulate cortex (dACC, area 24c). dACC neurons responded strongly after a decision outcome as reported in previous studies and, additionally, signaled whether a trial was from a learnable vs unlearnable set before outcome delivery, and showed interactions whereby their selectivity for the outcome (reward vs lack of reward) was stronger for learnable versus unlearnable sets. Learnability and interaction responses were independent of sensory or reward cues (which were equated for learnable and unlearnable sets) but their strength correlated with the monkeys' ability to avoid inferring false order in unlearnable sets. The findings suggest that the dACC is part of a network that monitors learnability and enables animals to appropriately focus learning on true patterns while avoiding false inferences about spurious and random associations.