Active Inference Psychotherapy and Dyadic Inference

Status: CLOSEST EXISTING TOPOLOGY for the capacity-restoration function.

Chamberlin (2023) -- most directly relevant

"The Active Inference Model of Coherence Therapy" describes psychotherapy as dyadic therapist-guided active inference that renders the probable hidden causes of behavior conscious. The therapist inverts his evolving generative model of the client's hidden generative model to specify likely suboptimal priors. Explicitly references Arnsten on stress-induced PFC degradation.

Structural parallels to this architecture

• External agent performs inference about another system's generative model = observer/state-estimator function
• Goal is restoring client's capacity for self-directed inference = capacity-as-objective
• "Bayes Optimal Pathology" (suboptimal priors + intact inference) = the problem is state, not motivation
• Therapist modulates search intensity based on client capacity ("regulating the temperature") = state-conditioned gating
• Therapeutic outcome is "emergence of agency and control" = terminal objective

Broader family

• Krupnik: therapeutic alliance as decreasing free energy of participants' generative models
• McParlin et al.: therapeutic synchrony as coupled systems minimizing free energy
• Kim et al.: therapy as weakening suboptimal priors, redeploying attention, installing new generative models
• Davey & Badcock: psychotherapies as re-weighting precision, addressing higher-order self-models, scaffolding future confidence

What the psychotherapy literature does NOT provide

• AI understanding layer. External agent is a human therapist, not AI. Cannot operate at scale.
• Real-time physiological state estimation. Therapists use clinical observation, not HRV/EDA/pupillometry/AI-derived signals.
• Formal objective function. No paper specifies capacity for self-directed action as a formal control target with measurable state variables and control laws.
• Caring governance as system invariants. No paper translates Mayeroff into testable invariants (I1-I6) or designs removal tests.
• Cost/access constraint. Therapy is inherently expensive and access-limited.
• Anti-dependency/removal test. No paper specifies outcomes-maintained-after-withdrawal as success criterion.

Positioning statement

The active inference psychotherapy literature independently articulated the core topology: an external agent restoring another agent's capacity for self-directed inference. What it provides is therapeutic validation that the topology works. What it does not provide is the engineering specification for doing this with an AI system, at scale, with formal governance constraints. The contribution of this work is not the topology itself. It is the specific combination of AI understanding, real-time state estimation, capacity-as-objective, and caring governance.

Evidence status

ESTABLISHED (peer-reviewed, empirically grounded psychotherapy literature).

Sources

• Chamberlin 2023 (Front. Hum. Neurosci. 16:955558)
• Krupnik, McParlin et al., Kim et al., Davey & Badcock (specific citations to verify)