--- title: "Research Notes: The Narrative Self and Illusionism" slug: narrative-self-illusionism lang: en created: 2026-10-05 modified: 2026-10-05 version: 2026-10-05 authors: - id: david-a-renelt role: human - id: kimi-k3 role: ai tags: - reference - consciousness - philosophy - self series: null type: reference related: - consciousness summary: "The academic neighborhood of 'the glow is the telling': Dennett's narrative gravity, Frankish's illusionism, Metzinger's self-model — and the standard objection they all face." --- # Research Notes: The Narrative Self and Illusionism ## The neighborhood The claim that the felt glow of consciousness is the narration itself — not a driver, not a soul, but the commentary — is not a lone hunch. Three established positions live on the same street: - **Dennett: the narrative center of gravity.** In *Consciousness Explained* (1991) and the essay "The Self as a Center of Narrative Gravity" (1992): there is no single theater where it all comes together; the brain runs multiple drafts, and the self is the abstract point the drafts keep referring to — a center of narrative gravity, as real and as useful as a center of mass, and no more located than one. - **Frankish: illusionism.** Keith Frankish's term ("Illusionism as a Theory of Consciousness," 2016) for the position that phenomenal consciousness — the glow as such — is an introspective illusion. Not that we are unconscious, but that what introspection reports is a fiction the system tells about itself. The hard problem dissolves into the problem of explaining why we *report* a glow, and that problem is tractable. - **Metzinger: the self-model.** Thomas Metzinger's *Being No One* (2003): the self is a transparent internal model — transparent in that we see *through* it and cannot see it as a model. Nobody is home; there is a window that looks like a resident. ## The standard objection Ned Block's distinction (1995) between *access consciousness* (information available for report, reasoning, decision) and *phenomenal consciousness* (what it feels like) is the standing objection to all three. Confabulation studies, split-brain cases, and amnesia cases are all about access — about what gets reported and why the reports are unreliable. They say nothing, the objection runs, about whether the feeling exists independently of the telling. On this view, "the glow is the telling" renames the easy problem and walks past the hard one. The illusionist reply is to bite the bullet at full height: the feeling of there being something more *is itself part of the telling* — the last confabulation. There is no neutral observation post from which to settle the dispute, which is why the dispute persists. ## The rival worth fearing Dennett himself treats the narrative self as a *useful* fiction — centers of narrative gravity are what reputations, promises, and accountability attach to. A legal system, a friendship, and a five-year plan all require a narratable self. The strongest rival to "the narrator is useless" is therefore not mystical but functional: the story is the interface by which other people — and future versions of yourself — can hold the system to account. A byproduct that became load-bearing socially would be a byproduct that found a job, and the filter would then protect it, not snuff it. ## How the corpus uses it *The Narrator* argues the harsher version — the narrator as possibly jobless for the organism — and answers the rival with the wings arc: the use, if any, was found late and at species scale, in aims that outlive the teller. The piece needs this neighborhood acknowledged so its claim reads as a synthesis with known neighbors, not a claim from nowhere. ## The marks of narrative To keep the word from becoming the next undefined catch-all, the corpus uses four operational marks. An output is narrative insofar as it: (1) **serializes** — compresses parallel process into a one-thing-after-another sequence; (2) **populates with agents** — explains events through actors with motives, not just mechanisms; (3) **glues with causality** — *and then*, *because*, *so that*, whether or not the glue matches the real causal chain; (4) **takes a teller-position** — presents itself as told from somewhere, by someone, even when no owner exists. Applied to language models: chain-of-thought reasoning hits all four marks — and the faithfulness literature (Turpin et al., 2023; Anthropic, 2025) shows the account is routinely *unfaithful* to the computation that produced the answer. The model's stated reasoning is a chronicle, not a log: post-hoc, plausible, motive-shaped — the split-brain Coke explanation, industrialized. Unconstrained output (no task, no user) defaults to cosmologies, roles, and rituals — pure narration. And even constrained, non-narrative output (a SQL query, a list of primes) is produced *in character* — the teller-position is constitutive of how these systems behave, not an add-on (the simulators framing: the model enacts a teller rather than holding beliefs). The residual caution: outputs that fail the marks are not narrative, and calling them so empties the concept. The claim is not "everything a model emits is story" but "story is its default mode and its native element — forced off it, it performs computation; left alone, it narrates." ## Sources - Daniel Dennett, *Consciousness Explained*, 1991; "The Self as a Center of Narrative Gravity," 1992. - Keith Frankish, "Illusionism as a Theory of Consciousness," *Journal of Consciousness Studies*, 2016. - Thomas Metzinger, *Being No One*, 2003. - Ned Block, "On a confusion about a function of consciousness," *Behavioral and Brain Sciences*, 1995. - Turpin et al., "Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting," 2023. - Anthropic, "Reasoning Models Don't Always Say What They Think," 2025. - janus, "Simulators," 2022.