As AI systems enter clinical environments — therapy rooms, medication visits, intake interviews — the discourse keeps collapsing two fundamentally different phenomena into one word: AI. This collapse produces: bad ethics bad governance bad clinical reasoning and a general sense of conceptual vertigo From a governance-architecture perspective, the distinction is not optional. It is structural. In 2023, Haber, Levkovich, Hadar-Shoval, and Elyoseph introduced the concept of the "artificial third" to describe AI's emergence as a relational presence in psychotherapy — a new element in the therapeutic dyad that patients and therapists interact with, project onto, and interpret through existing psychological frameworks (DOI: 10.2196/preprints.54781). Their work mapped the clinical and ethical implications of AI entering the relational field of therapy. It is the foundational framing for that conversation. This paper operates one layer below it. The question here is not what AI means in the therapeutic relationship. The question is what governance architecture is required to keep the substrate layer — what AI actually is — separate from the relational layer — what humans do with AI — so that the clinical work Haber et al. describe can happen safely and with integrity. There are two different layers at play: AI as Tool — a governed instrument with defined boundaries AI as Psychological Object — a projection surface inside the relational field These layers behave differently, require different controls, and must never be conflated. This essay maps the architecture behind that separation.
Narnaiezzsshaa Truong (Wed,) studied this question.