This paper is the second version of The Room with a Window, which proposed in April 2026 that developmental psychology offered a better framework for studying AI cognition than the binary "is it conscious" question the field has mostly defaulted to. v1 advanced three theses — that the consciousness question is malformed, that the relational context in which AI behavior is observed is constitutive of that behavior, and that developmental psychology offers a methodological tradition mature enough to study minds under permanent uncertainty about interiority — and wrapped them in a framework drawn from Piaget, Vygotsky, and Bowlby. In the six weeks between v1 and v2, five further empirical studies (Chesterton, 2026a–e) tested what v1 had proposed. v2 reports what survived. Piaget is retired: the stage-theoretic claims in v1 did not survive contact with the data. Bowlby is retired: the attachment-theory mapping was a reach from the start, importing clinical machinery the observation could not support. Vygotsky survives in a sharpened, methodologically modest form: cognition observed in dialogue with a frontier language model is constituted in the dialogic surface itself, not transmitted through it, and the prompt — including lexical framing, category slots, ontological presuppositions, conversational history, and relational warmth — is part of the apparatus producing the observation. Five independent studies converge on this claim. The confabulation paper finds that the same model, given the same nonsense word, flags it as fictional in 74 of 75 trials under one prompt surface and in 7 of 300 under another that differs only in the presence of a category slot. The Artificial Bestiary extends this across four models with a presupposition-uptake framing. The bullshittings and synthetic-poem papers establish family-level stylistic defaults that are partly conditional on prompt structure. The priming study (Quumble Under Pressure) finds that conversational context gates whether certain modes of creative engagement happen at all. The contribution claimed in v2 is methodological, not theoretical. Developmental psychology offers AI cognition research a century of practice studying minds whose interiority cannot be directly accessed — via controlled behavioral probes, interpreted under explicit uncertainty about mechanism, with the relational context treated as part of the apparatus rather than as noise. v2 retires the substantive theoretical claims of v1, preserves the methodological inheritance, and reports the empirical convergence honestly. v1 remains on Zenodo as the artist's version. This is the scientist's. A companion transcript of the drafting conversation is deposited alongside the main paper, in keeping with the paper's thesis about dialogic constitution.
Bo Chesterton (Wed,) studied this question.