Neuromodulatory Agentic Learning (NAL): Four Criteria, a Binding Operator Φ, and Three Timescales

Abstract

Research note fixing the author's definition of Neuromodulatory Agentic Learning (NAL), an instantiation of the Agentic RL paradigm (Zhang et al., arXiv:2509.02547). It defines four criteria for a deployed learning agent (persistent memory; post-deployment parameter change; post-deployment reward shaping behavior; and a growing input space through a binding operator Φ), introduces learning maturity as the target property, and maps dopamine, serotonin, noradrenaline and acetylcholine onto three timescales of learning (after Doya, Neural Networks 2002). Includes a measurement and containment section. Claims no priority over the Agentic RL paradigm or over the continual-learning metrics it builds on. Date of record 2026-10-11 (JST); OpenTimestamps proof included.