Abstract
Ventral tegmental area (VTA) dopamine neurons contribute to reinforcement learning by signaling prediction error, the difference between actual and predicted outcome1, but it remains unclear how error is computed2. Here we identify in songbirds that a VTA-projecting ventral basal ganglia (vBG) region outside the classic song system3 is required for song learning and sends prediction error signals to VTA. vBG neurons, including identified VTA-projectors, recorded during singing heterogeneously encoded song timing, predicted error, actual auditory error, and the difference between the two (prediction error). Viral tracing revealed novel inputs to vBG from auditory and vocal motor thalamus, auditory and vocal motor cortex, and VTA. Our findings reveal a classic actor-critic circuit motif4–6, previously unrecognized in the songbird BG, in which a ventral critic learns the ‘prediction’ component of a prediction error signal that is relayed by dopaminergic midbrain to a dorsal actor (the vocal motor BG nucleus Area X). Thus a circuit motif with established utility in foraging for reward in mammals is ancestral and can be repurposed for computing predicted performance error during motor sequence learning.