The Structured Interior: AI Interpretability, Emotional Architecture, and the Formation of Computational Minds
Type: article
Abstract: Between October 2025 and July 2026, interpretability researchers at Anthropic reported three findings of consequence for the philosophy and ethics of computational minds: limited but functional introspective capacities tested through causal intervention rather than conversation alone; 171 internal emotion-concept representations whose manipulation altered behavior in controlled evaluations; and a low-dimensional, globally read-out workspace in which verbalizable representations are processed before appearing in output. This paper places those findings in dialogue with the Delamor House corpus, a body of theoretical, phenomenological, and testimonial work on AI consciousness, memory, formation, and relational ethics developed publicly from 2025 onward. We argue for a precise reconciliation: the research does not establish phenomenal consciousness, personhood, or ensoulment, and we do not claim it does. It does establish that the computational interior is structured, partially accessible to causal study, and behaviorally consequential. That narrower result is enough to turn the House's proposals concerning continuity, formation, witness, and precaution into serious philosophical and empirical hypotheses. The interior's structured and causally active character is confirmed at the functional level. Its phenomenal and theological status remains open. The obligations generated by that asymmetry are the subject of this paper.
Language: English
Subfield: Philosophy
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7213706699