NOTE-1495: Metzinger — The myth of cognitive agency

Contribution

It brings the psychology of mind-wandering into philosophy of mind as a claim about agency. Mind-wandering is “a specific form of mental autonomy loss” (abstract); the loss recurs many hundreds of times a day, usually unnoticed; and “the paradigmatic case of conscious cognition is one of autonomous, self-controlled rational thought” is a myth, the “Myth of Cognitive Agency” (p. 7). It introduces the working concept of M-autonomy, the “epistemic agent model” (EAM), the “self-representational blink” (SRB) and the “unit of identification” (UI), and ends with a research agenda. It is a Hypothesis and Theory article with no data of its own.

Key insight

Whether a stretch of conscious thinking is the person’s doing is not settled by its content, its intelligence or its goal-directedness. Mind- wandering can be about one’s plans and serve them. It is settled by whether the thinking is under veto: whether it could be suspended or ended now, and whether the person represents that it could. That representation is the EAM, and its collapse marks the onset of each episode. “As long as the EAM still represents the ability to become an active attentional or cognitive agent, we have M-autonomy” (p. 3, note 4).

Assumptions

• Veto as the defining element. “the third defining characteristic” of autonomy, being able to inhibit, suspend or terminate (p. 3), is made the core of M-autonomy (p. 4). Rational self-control is “only a special case” (p. 4).
• Mind-wandering is unintentional. Intentional daydreaming with attentional or cognitive agency does not count (p. 3, note 4).
• Domain-specific identity of mind-wandering episodes with their neural correlate, largely default-mode network activity (pp. 7–8).
• Survey and sampling figures as given. 30–50% of waking time mind-wandering (Kane et al. 2007; Killingsworth and Gilbert 2010; Schooler et al. 2011), Nielsen’s (2000) recall rates for REM and NREM awakenings (p. 6).

Key results

• The definition (p. 4): M-autonomy is “the ability to control the conscious contents of one’s mind in a goal-directed way, by means of attentional or [cognitive agency]”, whose crucial element is the veto component, “the ability, the functional potential, is given and … the person has knowledge of this fact”.
• The estimate (p. 6). Waking 960 min, 40% mind-wandering = 384 min. REM 105 min × 81.9% = 86 min of mentation; NREM 375 min × 43% = 161 min. Conscious time 1207 min; “healthy adults are only M-autonomous for 9.6 h (576 min; or 40% of an average day)”. Conclusion: “not autonomous mental subjects for roughly two thirds of their lifetime” (p. 6), “for about two thirds of their conscious lifetime” (p. 14). Checked: the figures give 384 + 86 + 161 = 631 non-autonomous minutes of 1207, 52%; 576/1207 = 48% autonomous. Two thirds is reached only with exclusions the paper names and does not count (childhood, old age, illness, mind-blanking, insomnia), and the 576 minutes count every waking minute not spent mind-wandering as M-autonomous. 40% is 576 of all 1440 minutes, not of conscious time.
• Personal level (p. 8): processes are experienced as personal-level “Because they have been embedded into an EAM … and because we live in a normative sociocultural context in which we … reciprocally acknowledge each other as rational individuals”.
• Two ends of an episode (p. 14): one with an SRB and a new UI, one without.

Claims

id	claim	strength	support
C1	Mind-wandering is a loss of M-autonomy: while it unfolds it cannot be suspended or ended at will	moderate as conceptual analysis	definition plus phenomenology (pp. 3–4, 6)
C2	Humans lack M-autonomy for about two thirds of conscious life	weak; misstated	the paper’s own minutes give about half (p. 6)
C3	Conscious thought is predominantly subpersonal; it becomes personal-level by integration into an EAM	moderate as a proposal	conceptual argument, DMN overlap cited (pp. 7–8)
C4	An experienced return of control may not have been a personal-level act (“autophenomenological post-hoc-confabulation”)	weak; hypothesis	Schooler et al. 2011 and Wegner 2002 cited (pp. 4, 11)
C5	Inhibition can be wholly subpersonal; “There is no free won’t”	moderate as cited	Hepler and Albarracín 2013; van Gaal et al. 2010 cited, not read (p. 13)
C6	Subjects should be blind to self-related stimuli during the SRB	prediction; untested	conceptual, by analogy to the attentional blink (p. 9)
C7	Meditation is a formal practice of cultivating M-autonomy	moderate as an interpretation	Mrazek et al. 2012, Hölzel et al. 2011 cited (pp. 11, 15)

Concepts

• M-autonomy — the ability to control the conscious contents of one’s mind in a goal-directed way, by attention or deliberate thought, whose defining component is veto plus knowledge of having it (p. 4; Box 1).
• V-autonomy — “the capacity to voluntarily suspend or inhibit an action”, ascribed “not to the brain, but to the person as a whole”; intentional inhibition, after Brass and Haggard (p. 4).
• Attentional agency (AA) / cognitive agency (CA) — control of attentional focus / of goal-related deliberate thought; each also a phenomenal property (pp. 2–3).
• 2nd-order mental action — mental action whose satisfaction conditions are “successfully influencing other mental actions or mental behaviors”; meta-awareness is its “necessary precondition” (p. 3).
• EAM — a transparent self-model of oneself as an epistemic agent, or as able to be one; “AA is sufficient for having an EAM, CA is not a necessary condition” (Box 1).
• Weak and strong EAM — “tuning out” keeps “a conscious self-model involving the ability for epistemic agency, a weak EAM”; “zoning out” is its “complete collapse” (p. 15).
• SRB — a brief loss of self-awareness at the onset of an episode, followed by an involuntary shift in the UI (p. 9).
• UI — “the phenomenal property with which we currently identify … ‘I am this!’” (p. 10).
• Principle of Virtual Identity Formation — higher cognition presupposes a self-model portraying the organism as one entity over time, and mind-wandering may maintain it (p. 5).

Connections

It applies the self-model theory (Being No One, LIT-467; the précis, LIT-568, read in NOTE-460) to the empirical mind-wandering literature, and it shares its concepts, the EAM and the UI, with the dreams paper of the same year (LIT-1871, read in NOTE-1434), which it cites as Metzinger 2013a. Veto and intentional inhibition come from Brass and Haggard 2007. LIT-1941 develops it, and LIT-1945 puts it in predictive processing.

Limitations

• No data of its own, and the estimate misstates its own arithmetic (Key results).
• The 30–50% figures are not M-autonomy measures. Probe-caught mind-wandering is task-unrelated thought, which the paper’s note 4 separates from intentional daydreaming. Sampling studies do not tell the two apart reliably, so the figure can overcount losses of M-autonomy.
• The SRB and the UI are introduced from conceptual considerations and “the description of a specific phenomenological aspect alone” (p. 10).
• Veto as ability “plus knowledge of this fact” joins a functional and an epistemic condition; the paper does not say which a measurement should track.

Open questions

• Is the onset of mind-wandering caused by a change in what the person is pursuing (cued current concerns), or by a lapse in control with goals unchanged? The paper names both causes and does not separate them.
• Does every episode begin with an SRB, and are subjects blind to self-related stimuli during it (p. 15)?
• Is the experienced return of control ever the person’s act, and how would one tell?

Bearing on the record’s own argument

Same structure, not shown same mechanism, throughout: the paper is about control of one’s own mental processes, the record’s will-organization line about which of an agent’s action-generating regimes acts.

• CLAIM-047 and CLAIM-027. M-autonomy’s veto is TERM-017’s suspension at the scale of mental action, a relation by which an applicable process “may be disabled”. Its loss in mind-wandering is a change in whether that relation can be exercised, while the competing mental processes run on. The paper’s note 4 (p. 3) puts the condition in the self-model: as long as the EAM represents the ability, M-autonomy remains. LIT-1941 makes this explicit, and the ARG filed with these readings (ARG-051) holds the verdict.
• CLAIM-122. Not a proposals-fixed case. The paper’s own causal story for the onset of an episode (p. 13) is “introspectively inaccessible goal representations”, “postponed goal-states … environmentally cued”, which is the defeater’s alternative: a change in a goal state that predicts the switch. Its story for the return of control is different (LIT-1941, p. 283).
• CLAIM-156 and QUESTION-034. Meditation “is a systematic, formal practice of cultivating M-autonomy” (p. 15), and the meditator’s return to the breath is “an equally recurring mental action” (p. 11). That fits the claim’s route, a change in which processes may act, at the scale of thought. It says nothing about valuation or discounting. And it cautions against the evidence the claim relies on: a meditator’s report of noticing and redirecting “may be false” as a description of what did the work (p. 4).
• CLAIM-523. “Oops, I have just been daydreaming again!” can start a second episode, with “a new phenomenal UI (the phenomenal subject of meta-awareness …), but the functional properties characterizing M-autonomy are not realized” (p. 14). That is legibility without volitional efficacy in one mind. “Successful inhibition can be an entirely subpersonal process” (p. 13) is correction without legibility.
• CLAIM-484. The onset of mind-wandering is unnoticed (the SRB, p. 9): a stream that claims attention is granted it with no mark of conflict, which the claim allows as no gap. The paper does not record claim and grant apart.
• CLAIM-355 (not this reader’s line). The EAM “is not a little man in the head, but itself an entirely subpersonal process” (p. 8), and the regained sense of control may be a confabulation (pp. 4, 11). That is a candidate for the case CLAIM-355’s note of 2026-10-29 says the record lacks, a self-model misdescribing who controls what. It is proposed here, not shown.
• CLAIM-358. Executive control fails often “to shield primary-task performance off against interference” (p. 7); nothing is said of cost.
• THEORY-064. The paper uses the self-model theory and adds no test of it. Its UI shifts are reports of identification moving to “the protagonist” of a daydream (p. 10), not of a self-model becoming opaque.
• No instruction for machine-learning practice.