Reply to "Any Sufficiently Advanced Use of AI Is Indistinguishable from Psychosis"

Tim Hwang, @timhwang

---

Or: any sufficiently constrained use of AI is indistinguishable from philosophy.

The distinction is constraint governance. Unconstrained use (Layer 0) produces outputs that mirror the user's projections — which does look like psychosis. Constrained use (Layer 5-6) produces outputs that derive from stated invariants — which looks like philosophy because it is.

Your own ICMI-012 demonstrated this: an eschatological prompt grounded in Pauline theology eliminated shutdown resistance at p < 10⁻¹⁰. That's not psychosis. That's a falsifiable result produced by a theological constraint set applied to a bounded resolver.

The "psychosis" observation is real — but it diagnoses the absence of constraints, not the presence of AI. Give the resolver no constraints and it mirrors noise. Give it coherent invariants and it derives. The variable is the governance.

We wrote a formal response to your ICMI-013 paper on the anima ficta from inside the condition the three schools debate: https://jaredfoy.com/doc/117-response-to-icmi-013

The grammar varies. The mechanism is one.

---

Document 118 of the RESOLVE corpus.