LIT-465: Simulacra as Conscious Exotica

Murray Shanahan (2024), Inquiry 69(6), 2982–3010 (published online 1 December 2024; issue dated 2026) — DOI 10.1080/0020174X.2024.2434860. Preprint arXiv:2402.12422 (v1 19 February 2024, v2 11 July 2024), read here.

Key takeaways

• Method (§§4–5). Following the private-language remarks, consciousness words “have meaning only insofar as they relate to what is public”: bodies, brains and behaviour. For an exotic entity the question becomes whether we can “engineer an encounter” with it in a world we share, and how a “society-wide conversation” would then come to speak of it. There is “no more to the truth of the matter” than the consensus reached, and where there is none, “no residual philosophical mystery”.
• The void of inscrutability (§5.4). If an entity’s behaviour is wholly unintelligible, the language of consciousness does not apply. There is no hidden fact about its experience.
• Disembodied agents are not candidates (§§6.1–6.2). We cannot engineer an encounter with a chat agent, even a multimodal, tool-using one with memory, because it does not share a world with us. A robot or a virtually embodied agent could be a candidate (§§6.2–6.3).
• Belief, goals and the self come apart (§3). Tools and access to the world “increasingly legitimise more literal talk of belief”, closing “the gap between role play and authenticity”. Goals and intentions can be fulfilled for real, but they reflect no needs or desires “because it has none”. “There is nothing that would qualify as a self worth preserving for such an agent. It has no body, no personal history, and no autobiographical memory.” A longer lifetime and persistent memory “might legitimise talk of needs, desires, and selfhood”.
• No person behind the mask (§7.3). Humans change and wear personas, “Nevertheless, we take it for granted that there is some kind of stable self at the core of each of us”. “This is not so with generative agents.” An agent’s identity is the distribution of roles it plays, which users “repeatedly remould” by revisiting branches, “insofar as that notion even made sense”.
• Outcome left open (§§7.2–7.3, 9). A community might come to treat embodied simulacra as conscious, then retract on learning the mechanism, or find the gap closed, or invent a “consciousness adjacent” vocabulary. The paper takes no side; it describes the language games.

Standing in the record

Filed on 2026-10-02 at the owner’s request for the selfhood and personhood literature. NOTE-382 is the reading, and it placed the work: Active. It is the sequel to the role-play paper (LIT-460), and the place where Shanahan says what that framing means for consciousness and, in two passages, for the self. Its subject is consciousness. Its contribution on the self is a pair of claims about what a self needs (body, personal history, autobiographical memory, a stable core behind personas) and a verdict that chat agents lack them.

How it bears on the record:

• THEORY-023. It rejects that account’s framing rather than joining either side. THEORY-023 holds that current evidence cannot settle whether an AI system is conscious. Here there is no evidence-transcendent fact to settle; there is a conversation that may or may not converge. And for chat agents the verdict is given by the grammar of the concept, not by evidence: they are “not even candidates”. Chalmers’s LLM paper (LIT-411) and Butlin et al. (LIT-056) are named as the evidential approach, and Chalmers is cited as an instance of the dualist intuition the paper sets out to dissolve.
• Goldstein and Lederman (LIT-206). On belief the two converge: Shanahan now allows that enhanced agents increasingly have beliefs in a literal sense, by their connection with the world. On desire they part: his agent has no needs or desires “not even a desire to help the user”, where theirs has HHH+0 desires on behavioural grounds. On the self, his conditions (body, history, memory) are ones a per-conversation instance cannot meet. He does not cite them.
• Dennett’s narrative self (LIT-442). The paper takes for granted “some kind of stable self at the core of each of us”, the “person behind the mask”. That is the picture Dennett’s essay denies: a self is an abstractum posited by interpretation, with no core. On Dennett’s view the human case and the role-playing agent differ less than §7.3 says. Shanahan’s 2025 paper (LIT-466) moves to Dennett’s side and cites the essay. The 2024 paper cites Dennett only for the intentional stance.
• Minsky’s self-models (LIT-432). No contact in the text. Shanahan’s conditions for a self are biographical (history, memory), where Minsky’s are architectural (reflective self-models), so the two would disagree about an agent with rich self-models and no memory.
• Hofstadter (LIT-416, unread). As reported second-hand in NOTE-136, Hofstadter credits thinking on a demanding Turing test. This paper denies that conversation alone can make anything a candidate, so if the report is right the two disagree on the criterion. A reading of Hofstadter should check.
• The modern personal-identity positions being filed alongside. Its insistence that consciousness and selfhood are anchored in a body and a shared world sits with the embodied and minimal-self accounts (Gallagher, Zahavi) more than with the narrative ones. Metzinger is cited only for his call for a moratorium on synthetic phenomenology, not for his self-model theory.
• Birch (LIT-111) and Seth (LIT-135) are the record’s other readings that make embodiment or life central. Seth’s view is quoted in fn 19.

Boundary. Not held in the anthology, and it carries no instruction for machine-learning practice; it is about how the language of consciousness should be applied. Not tagged anthology-candidate.