Concurrent metacognitive evaluation is dissociable from perceptual evidence accumulation in the parietal cortex

Abstract

Adaptive behavior requires not only accumulating evidence toward a choice but also monitoring its reliability in real time. A central question is whether such concurrent uncertainty is read out directly from the circuit that accumulates perceptual evidence, or is computed by a separate process that runs alongside it. We trained two macaques on a perceptual discrimination task in which they could abandon deliberation at any moment and request assistance, so that uncertainty responses were in parallel with, rather than appended to, evidence accumulation. Behaviorally, a race between distinct perceptual and uncertainty accumulators outperformed models in which the opt-out decision was derived from the perceptual decision variable alone. To ask where the second accumulator resides, we recorded single units in the lateral intraparietal area (LIP), a canonical sensory-motor transformation area, and compared population dynamics with recurrent neural networks instantiating two architectures: one in which uncertainty is computed within the perceptual accumulator, and one in which it is computed outside and triggers the uncertainty response there. LIP exhibited all of these signatures for perceptual choices and none for uncertainty responses. Uncertainty-related activity rose late and stereotypically, lacking the coherence-scaled ramp of an accumulator, and separable from the perceptual accumulation. They are all predicted by the segregated architecture. Concurrent uncertainty is thus represented in LIP but is unlikely to be computed within LIP, indicating that metacognitive monitoring during deliberation is a dissociable operation rather than a derivative readout of the LIP perceptual circuit; its anatomical substrate is not localized here.