Conference on Cognitive Computational Neuroscience

Working memory (WM) and decision-making (DM) are two higher cognitive functions for which detailed neural circuit models have been developed. Prominent classes of these models share a reliance on tunable attractor dynamics generated by the same circuit configurations, leading to the suggestion that both maintenance (for WM) and integration (for deliberative DM) of information over time are implemented within shared neural circuits. We present new lines of convergent evidence supporting this idea. Human participants and trained recurrent neural networks showed signatures of the same shared-circuit solution to WM and DM tasks, reflected in generalization of persistent neural codes across task contexts to represent both memoranda during WM and evolving decision variables during DM. While participants adapted across contexts, their behavior in both was also subject to noise and biases that likely arose within the shared circuit. Trial-related changes in pupil size suggested a mechanism – modulation by ascending arousal systems – by which the shared circuit is tuned to produce stable memories vs labile decision states. These findings promote an integrative view of WM and DM that holds promise for understanding brain disorders characterized by deficits in both functions.