NOTE-1508: Jennings — Attention as the Source of a Free Will

Contribution

It gives an account of attention that keeps the traditional view, that a subject directs attention, while answering the two standard objections: the homunculus and the unspecified “resource”. The subject is distributed: a collection of interests, with no central controller (§6). It then argues that this subject has a non-reducible causal power, through contextual emergence, and that this power is the source of a free will (§§5–6). Its one move not found in its sources is a condition on contextual emergence, a boundary between the emergent phenomenon and its context (pp. 18–20; fn. 15 says this is what the paper adds to Jennings 2020).

Key insight

“Some things only emerge in the right context” (p. 17). If the context is neither a mere background condition nor just more parts, because a boundary separates it from what emerges and that boundary is part of why it emerges, then the emergent whole can constrain its parts without downward or circular causation (fn. 14). For a brain the context is the skull and meninges, which separate its electrical activity from the surround, and the brain stem, which carries energy and signal in, like the container and the heating element of Bénard cells (p. 29). The subject that emerges is the whole collection of interests, and what it does is prioritize among them.

Assumptions

• Interests are the subject’s parts, taken as a “ground truth” for living beings (fn. 16, p. 22), restricted to “organism-level” interests that direct sensory and motor systems (p. 22).
• Three kinds of interest (pp. 23–26):
  • fixed, hereditary and structural (red, bright, moving, loud things);
  • formed by personal history, through memory and reward systems;
  • fluid, “neither inherited nor formed by experience”, such as new interests and the case of equally weighted options in conflict.
• A taxonomy of emergence by the parts’ necessity and sufficiency (p. 17): weak emergence, parts necessary and sufficient; strong, neither; supervenience, sufficient but not necessary; contextual, necessary but not sufficient. Attributed to Bishop and Atmanspacher 2006 (LIT-1987) for the distinctness of contextual emergence; the necessity–sufficiency gloss is the paper’s.
• Biological systems are not deterministic “according to the best science” (p. 2, citing Potter, Ellis and Mitchell 2026), and the paper “does not address global determinism” (p. 3).
• Agency and free will are treated as “roughly synonymous” (fn. 2).

Key results

The paper is argument and review. It reports no new data.

• Against the filter view (§3, pp. 9–11). If attention can be a mere filter, nothing separates it from natural selection or Müller-cell filtering, so it should be tied to the subject (p. 11). Against the traditional view: the homunculus (pp. 11–12) and unspecified resources (pp. 12–13). Biased-competition models (Desimone and Duncan 1995) avoid both but use only weak emergence (pp. 14–15).
• The boundary condition (pp. 18–20). Without a functional boundary, contextual emergence “would collapse into either weak emergence or supervenience”: the context’s parts could be added to the base (p. 19). With one, “a macro-level difference between the parts of the emergent phenomenon and its context leads to the emergent phenomenon”, and since “the thermal and material properties at stake here are macro-level ones, … contextual emergence is a better fit” (p. 20).
• The mechanism (pp. 25–26). Following Voigt 2022, when two options are equally valued, dlPFC and precuneus raise attention on one, which both breaks the stalemate and updates its value in memory, “so attention does not only help to break the stalemate but allows a fluid interest to become a formed interest over time” (p. 26).
• Kane (p. 26). Self-forming actions fit this role, but “what is missing from both Voigt and Kane is an explanation of how exactly the person manages to resolve the stalemate”. If it is noise or indeterminacy, “the outcome arguably wouldn’t be ‘up to us’”. The paper’s answer: “attention by an emergent subject can resolve the conflict”. Fn. 19 notes Kane’s later mixed agent- and event-causal view.
• The signature (pp. 27–29). 1/f scaling and phase-amplitude coupling, “evidence that low frequency waves constrain high frequency waves”, associated with attention (Szczepanski et al. 2014, LIT-1982). The analogy: a quiet flute must wait for a loud bassoon to fall silent to be heard (p. 28).
• Where fluid interests enter (pp. 30–32). Formed interests act through striatal oscillations and dopamine, combined with fixed ones through the thalamus; fluid interests need prefrontal feedback. Hence the criterion: “One way of determining this is through the presence or absence of feedback from dorsolateral prefrontal cortex” (p. 32).
• Free will (pp. 32–33). “The subject of attention has a power that goes beyond their component parts, and can exert control over those parts, showing how they can be such a source.” The account “does not assume global indeterminism” but denies determinism “as a workable theory of what is happening at the local level for living beings” (p. 32). “Constraint, in a noisy environment, is what gives us power over ourselves” (p. 33).
• Objections (§7). Time scale (Klein): slow waves last seconds, subjects far longer. Reply: the subject is the collection, not a wave, and formed interests use memory (pp. 33–34). Weak emergence: reply by ephaptic coupling, which “demonstrates that brain activity cannot be described simply through local interactions” (p. 35; Anastassiou et al. 2011, LIT-1953). No-self views (Metzinger; Lindahl and Britton): loss of the sense of self is not loss of the self (pp. 36–38).

Claims

id	claim	strength	support
C1	Attention, unlike filtering, is selection that answers to the whole collection of a subject’s interests	moderate as a definition	conceptual argument from the filter view’s lack of a principled boundary (p. 11)
C2	Feedback from dlPFC marks the involvement of fluid interests, and so of attention proper	weak	inferred from reviews that prefrontal feedback is needed for maintained goals (pp. 31–32); no study is cited that uses it as a criterion
C3	Attention breaks stalemates between equally weighted interests and forms new interests	moderate	Voigt 2022’s model, an Opinion article (p. 25)
C4	Contextual emergence needs a boundary between the emergent phenomenon and its context, without which it collapses into weak emergence or supervenience	moderate as a philosophical proposal	argued on Bénard cells (pp. 18–20); stated by the paper as a stipulation (fn. 15)
C5	The subject is contextually emergent, with a power over its parts that weak emergence cannot give	weak	analogy with Bénard cells; phase-amplitude coupling and ephaptic coupling (pp. 27–29, 35), both consistent with weak emergence
C6	Attention so understood is the source of a free will, without indeterminism	weak	follows only if C5 holds and a non-reducible power suffices for sourcehood, which is asserted (p. 32)
C7	Loss of the sense of self in altered states is not evidence against a self	moderate	distinction between self and sense of self, with the reviews it cites (pp. 36–38)

Concepts

• Interest — “that which drives” a living being; for organisms like us, “tendencies to seek out and respond to stimuli in a certain way” (p. 22).
• Subject — “a collection of interests with the power to constrain those interests, and that power derives from the collection of interests being in a specific context” (p. 21).
• Attention — “prioritization by an emergent subject”, contrasted with pre-attentive filtering and selective mechanisms (pp. 21–22, 32).
• Fixed, formed and fluid interests — as under Assumptions.
• Contextual emergence — parts necessary but not sufficient, with a functional boundary between phenomenon and context (pp. 17–19).
• Machretic determination — Gillett’s term for the constraint an emergent whole places on which parts it can use (p. 20).

Connections

• Bishop and Atmanspacher 2006 (LIT-1987) and Bishop, Silberstein and Pexton 2022: the source of contextual emergence. Bishop’s own later statement, the SEP entry of 2024 (LIT-1961, NOTE-1535), presents contextual emergence as real constraint by larger-scale stability conditions, with convection as the case, and sets it against causal powers. So “the whole constrains its parts” is Bishop’s; what the paper adds is the boundary condition and the step to causal power and free will. (Corrected on 2026-11-03; this line first glossed Bishop and Atmanspacher as about stability conditions for a well-defined description only. The 2006 paper is Unfetched.) Bishop et al. 2022 is LIT-1993, registered unread.
• Kane (LIT-1985 and Kane 1999, J. Phil. 96, cited; the 1999 paper is not held): self-forming actions as the case the account serves.
• Voigt 2022 (LIT-1976, registered unread): the model of stalemate-breaking.
• Desimone and Duncan 1995: biased competition, the weakly emergent rival. Not held.
• Moreno and Mossio; Thompson (fn. 14): circular or downward causation as the account of biological agency the paper avoids. The record holds readings close to them (LIT-1027).
• Potter, Ellis and Mitchell 2026, “Reframing the free will debate: the universe is not deterministic” (Synthese 207: 71). Not held.
• Frankfurt is not cited, and Wegner 2017 appears only as the sceptic named in the abstract.

Bearing on the record

• CLAIM-539: the owner’s “Frankfurt with a hat on”, now checked against the text. The architecture is two-level: a collection of interests decides which interest wins when they conflict (p. 31). The upper level is prioritization, a process, not a higher-order volition or an endorsement. The paper never mentions endorsement or identification. And it places itself “in between libertarianism and compatibilism” (p. 3), with a power it says “qualifies as a non-reductive causal power, supporting libertarian free will” (p. 3), and without indeterminism (p. 32). A14 reported both descriptions, and both are the paper’s. It is not an inconsistency in A14’s report.
• CLAIM-027. A mechanism-level candidate for its precedence relations: “When interests conflict, the collection will determine which of the interests wins. Without an emergent subject …, the winner would depend only on the existing strength of the interests in conflict and their local interactions” (p. 31). That is CLAIM-027’s non-reducibility to regime contents, plus a claim of non-reducibility to micro-dynamics that CLAIM-027 does not make. CLAIM-159’s cells place it as “emergent from the whole” acting on the parts.
• CLAIM-334. The paper denies supervenience explicitly and says it needs neither downward nor circular causation (fn. 14). Its argument that the boundary resists redescription is that the boundary’s properties “are macro-level ones” (p. 20). But a macro-level description of a boundary is still a description of micro constraints, the global supervenience the paper itself states and sets aside (p. 19). So the claim’s premise is denied rather than refuted. The paper offers no intervention on the macro level that holds its base fixed.
• CLAIM-540. The stalemate-breaking mechanism is a causal update: attention on the chosen item updates its stored value (p. 25). That is the structure CLAIM-540 says self-formation needs, a mixture that changes because of the choice. It is shown, in the paper, by citing Voigt’s model, not by an intervention.
• CLAIM-533. The paper’s reversal of the classic problem bears on the owner’s: “while the question was once how we can be undetermined in a deterministic universe, the task now is to explain how we manage to find stability in a stochastic one” (p. 3), and “constraint, in a noisy environment, is what gives us power over ourselves” (p. 33). That is CLAIM-478’s constraint read as freedom. It is also the opposite emphasis to Kane’s chaos: for Jennings what makes the outcome the agent’s is a stable constraint, not amplification.
• QUESTION-364. Fluid interests turning into formed ones are a worked case of a will revising itself. The paper answers whose revision it is by the subject’s being the whole collection. It says nothing about a collection whose interests were installed, so QUESTION-071 and THEORY-040’s manipulation case are untouched.
• QUESTION-033, QUESTION-034, CLAIM-156, CLAIM-158. “Who drives” has the answer “the collection, through its context”; QUESTION-034’s second option, a change in which interests may act, is the mechanism.
• THEORY-1296 and Metzinger. The paper takes Metzinger’s no-self view as an objection and answers it by separating the self from the sense of self (pp. 36–37). Metzinger’s M-autonomy, the record’s nearest neighbour, is not cited.
• LIT-388 (ephaptic coupling). The second appearance of ephaptic coupling as a basis of subjecthood. The paper offers it against weak emergence, but a field effect is another physical channel among neurons. It is non-local in wiring, not non-reducible.
• The exchange’s review (A5). Its page citations check: fn. 2 on p. 2, the bassoon on p. 28, ephaptic coupling on p. 35, no-self on pp. 36–38. Its objection to the bassoon analogy is half right. The analogy describes a fast signal timed to a slow one’s phase, which is phase-dependent gating; the flute’s amplitude does not change with the bassoon’s phase, so it is not phase-amplitude coupling as the paper defines it a sentence later. A5’s “Park and colleagues (2017)” is not in the paper.
• No instruction for machine-learning practice. The paper cites the transformer’s “attention mechanism” (p. 11) only as a bottom-up model.

Limitations

• The emergence argument rests on one analogy. Bénard cells are used to show that a boundary does work the parts cannot. But the paper’s own alternative description of them, global supervenience (p. 19), is not shown to fail; it is called a worse “explanation” (p. 19).
• The neural evidence does not discriminate. Phase-amplitude coupling and ephaptic coupling are predicted by models in which all causal work is micro-level. The paper says the account “would be further supported by evidence linking ephaptic coupling to attention” (p. 35); none is cited.
• The criterion for attention is a proposal. dlPFC feedback is inferred from the claim that maintained goals need prefrontal feedback.
• The free-will step is asserted. That a non-reducible power over one’s parts makes one an ultimate source (p. 32) is not argued against the manipulation or luck objections, and the paper does not address them.
• Fluid interests are defined negatively, “neither inherited nor formed by experience” (p. 25), which leaves open where they come from.
• Manuscript. The published version may differ.

Open questions

• Does intervening on dlPFC feedback abolish both the resolution of a stalemate and the later change in preference, as the account predicts and biased competition need not?
• Can a weakly emergent model with the same boundary conditions, encoded as micro constraints, fail to reproduce the macro constraint? That is what would separate contextual emergence from global supervenience (THEORY-1314).
• Would a subject whose interests were installed be free on this account?

Note of 2026-11-03 (later): the sources, read

• Szczepanski et al. (NOTE-1528) show correlation only: coupling strength predicted reaction time at 28 of 123 coupled electrodes, and nothing was manipulated. The paper’s “associated with” is accurate; its “improve” (p. 28) is not.
• Voigt 2022 (NOTE-1517) is an Opinion article. It proposes that attention updates value and does not show it, and its attention step is bottom-up: the precuneus picks the salient item and dlPFC holds it.
• Anastassiou et al. (NOTE-1516) bear out the quotation on p. 35 and not the inference: the authors explain field effects by circuit physics.
• Unverified. A search summary attributes a necessary-but-not-sufficient taxonomy of emergence to Bishop and Atmanspacher themselves; check when LIT-1987 is fetched.