LEARNING TO BALANCE EXTERNAL-INTERNAL NEURAL SYSTEMS: A THEORY OF SEROTONIN

The brain must continuously balance internal states with external evidence for effective behaviour, yet the mechanisms underlying this balance remain poorly understood. We propose a systems-level computational framework in which serotonin (5-HT) acts as a global modulator, tuning the relative influence of two distinct neural motifs: internal systems, which maintain learned task structures, and external systems, which process real-time sensory input.
In this architecture, 5-HT scales with global task error, consistent with its established role in signaling prediction error and environmental uncertainty. These 5-HT signals adaptively adjust pathway-specific gating weights to optimize performance. We validated this framework against three key experimental datasets. Our model qualitatively reproduces:
(i) Increased persistence at depleting reward sites during probabilistic foraging (Lottem et al., 2018);
(ii) Prolonged waiting during reward omission in delayed-reward paradigms (Miyazaki et al., 2018);
(iii) State-dependent transitions in motivated reward-seeking (Priestley et al., 2025).
Together, these results demonstrate how a single serotonergic mechanism can produce diverse, context-dependent effects by reweighting internal versus sensory-driven processing. This framework provides a unifying account of how 5-HT flexibly tunes brain-wide systems in response to task demands, offering a principled approach to understanding both adaptive behavior and serotonergic dysregulation.
(a) Theory, (b) Implementation, and (c) Validation of a computational model where serotonin gates external-internal system balance.