The Algorithmic Agent Perspective and Computational Neuropsychiatry: From Etiology to Advanced Therapy in Major Depressive Disorder

Ruffini, Castaldo, Lopez-Solà, Sanchez-Todo, Vohryzek — Entropy 26(11), 953, 2024.

Why I cared. Depression is described mostly by its symptoms, and it is wildly heterogeneous. I wanted a mechanistic account instead — to ask what is actually broken in an agent that stays persistently low.

What we did. Using the Kolmogorov Theory of consciousness, we built a foundational model in which an algorithmic agent interacts with the world to maximise an objective function that scores affective valence. Depression, here, is a state of persistently low valence — and we mapped the agent’s parts onto brain circuits and functional networks.

What we found. Low valence can come from very different failures: an inaccurate world model (cognitive bias), a dysfunctional objective function (anhedonia, anxiety), deficient planning (executive deficits), or simply a bad environment. These routes line up with depression biotypes — and suggest that stimulation, psychotherapy, and plasticity-enhancing compounds like psychedelics each repair a different part.

What it opened. If there are distinct routes into low valence, can we read which one a given person is on — and match the therapy, and the personalised model, to the route rather than to the label?