Abstract
We define exocognition as cognition that operates outside the skull — extended into books, instruments, AI partners, and external compute — while the human remains its author. The term is not our coinage; the definition and its use are what we contribute, built on the extended-mind lineage we credit throughout. We ground the definition in four bodies of work (extended, embedded, distributed, and offloaded cognition), and we state the strongest objections to the extended-mind thesis in full, because our definition does not require winning that metaphysical dispute. Our contribution is a design principle — the human remains the author — which we render testable via a distinction the cognitive-offloading literature supports: offloading that replaces a person's own cognition atrophies it; offloading that scaffolds it extends it. We separate what the evidence supports (design-around) from what remains vision (flagged), and we grade every claim.
1. Scope and stance
This paper defines a term and defends a design principle. It is not a claim to have resolved the philosophy of mind. Where we assert empirical facts, we cite them and assign a confidence tier (§9); where we interpret, we say so. As with our companion paper on Authentic Intelligence, we concede more to our critics than is customary — here, by presenting the full case against the extended-mind thesis on which our vocabulary draws.
2. The Interveil definition (stated precisely)
Exocognition (n.) — cognition that lives outside the skull: extended into books, instruments, AI partners, and, increasingly, external compute (cloud, data centers, and eventually direct neural links) — the growing stratum of thinking-and-remembering a person reaches beyond their own biology while remaining its author. Exocognitive (adj.) — of or belonging to that outer layer.
Two commitments are built into the definition:
- Provenance of the term. "Exocognitive" has prior technical and science-fiction usage; we do not claim the coinage. We claim the definition and its use for the age of thinking with machines, and we credit the extended-mind lineage explicitly (Clark & Chalmers, 1998). Crediting sources is a value we hold, not a slogan we borrow.
- The one law. The human remains the author. An extension is only an extension if the cognition it extends still occurs in the person. This is the load-bearing clause, and §5 makes it testable.
3. The scientific grounding
3.1 Extended cognition
Clark and Chalmers' parity principle holds that if a process in the world plays the same functional role as a process in the head, it is — for that reason — part of the cognitive process; their case of Otto's notebook (a reliably consulted external memory) motivates "active externalism" (Clark & Chalmers, 1998, Analysis; developed in Clark, 2008, Supersizing the Mind). This is the philosophical root of our vocabulary.
3.2 The counter-case (stated in full)
Our definition does not depend on the extended-mind thesis being metaphysically correct, so we present its strongest critics honestly:
- Adams & Aizawa argue there is a "mark of the cognitive" (intrinsic, non-derived content) that external props lack, and that extended-mind reasoning commits a "coupling-constitution fallacy" — confusing causal coupling with constitution (Adams & Aizawa, 2008, The Bounds of Cognition).
- Rupert contends the evidence supports embedded cognition — cognition heavily dependent on, but not literally constituted by, external resources — which explains the phenomena without extending the boundary of mind (Rupert, 2004, Journal of Philosophy).
We take these seriously. Fortunately, embedded cognition is sufficient for everything we design: whether the notebook is part of the mind or merely tightly coupled to it, the engineering and ethical stakes are identical. We build for the coupling; we remain agnostic on the constitution. (Interpretation.)
3.3 Distributed and embodied cognition
Independently of the metaphysics, cognition is well-described as a system-level property spanning brains, bodies, tools, and other people — e.g., Hutchins' analysis of ship navigation as computation distributed across a crew and its instruments (Hutchins, 1995, Cognition in the Wild). Group memory has a formal treatment as transactive memory (Wegner, 1987).
3.4 Cognitive offloading (the empirical core)
People routinely offload cognition onto the environment, and whether they do so is a value-based decision trading internal effort against external reliability (Risko & Gilbert, 2016, Trends in Cognitive Sciences). Offloading has documented benefits (freed working memory, better performance on the primary task) and documented costs: reduced memory for offloaded items — the "Google effect" on memory (Sparrow, Liu & Wegner, 2011, Science) — and increased vulnerability, since externally stored memory can be altered by the tool (offloading & memory manipulation, 2019, Cognition).
Honesty note: the broad "Google effect" is robust as a transactive-memory phenomenon, but at least one specific laboratory result (a "Google Stroop" effect) has failed to replicate (Hesselmann et al., 2020, PeerJ). We therefore treat the general offloading trade-off as strong and specific sub-claims as contested. (Graded accordingly in §9.)
4. The reframing: extension vs. extraction
The empirical asymmetry of the last decade is the ethical crux. Networked screens extended reach while contracting attention — heavy media-multitasking is associated with attentional lapses and poorer memory, and fragmentation can impair encoding (Uncapher & Wagner, 2018, PNAS). We name the failure mode precisely: a wrong-way extension in which the system models the person while the person cannot inspect the system — extraction wearing the costume of extension. Exocognition, as we define it, is the dignified form: the reach that returns the person larger, not emptier. (Interpretation, grounded in the cited attention literature.)
5. The one law, made testable
"The human remains the author" is not a slogan; it is an operational test, and the offloading literature supplies the discriminator:
- Replacing offload — the external resource performs the cognition and returns a result the person would otherwise have produced (the summary in place of the reading, the answer in place of the reasoning). This atrophies the internal capacity, precisely because retrieval and effortful processing are what build durable capability (Risko & Gilbert, 2016; on the learning value of retrieval and desirable difficulty, Dunlosky et al., 2013).
- Scaffolding offload — the external resource holds what the person cannot hold at once, or primes and prompts, so the person's own cognition still occurs. This extends.
The design shape that follows is prime · whisper · withdraw: ready the person before difficulty; offer context glanceably and dismissibly; and recede as competence grows — measuring success by graduation, not engagement. This yields a falsifiable product claim: an honest exocognitive tool should show retained or improved performance when the tool is removed; a tool that shows degraded unaided performance has crossed from scaffolding into replacement. (Testable; we commit to measuring it — see the AI paper's north-star: independence-when-support-is-removed.)
6. The vision (flagged as vision, not claim)
The exocognitive lineage is old and deliberate: Vannevar Bush's Memex imagined associative external memory (Bush, 1945, As We May Think); Engelbart's program was the augmentation of human intellect, not its replacement (Engelbart, 1962). The next extensions are more intimate — always-listening wearables, spatial capture, and, on a nearer horizon than many admit, neural interfaces. We hold, as vision rather than demonstrated fact, that this layer can be built as augmentation in Engelbart's sense — a mind reaching, keeping its hand on the dial — and we explicitly refuse the "successor-species" and "cyborg" framings that design human dignity out from the start. (Aspirational; labeled as such.)
7. Where it lands for Interveil
Exocognition is the spine under the product suite, each a face of one idea: Glosa (the page), dayFrame (memory and belonging), Presence (co-presence), Chronoaesthesis (knowledge itself). One principle — instruments that let a mind reach past its edges and return larger — now with a defined term and a testable law.
8. What we do not claim
We do not claim that the extended-mind thesis is metaphysically settled (embedded cognition suffices for our purposes); that external systems possess intrinsic cognitive content; that any specific offloading sub-result (e.g., "Google Stroop") is established; or that neural interfaces are presently viable. The definition rests on the design principle (author-preservation) and the credited lineage, not on any contested metaphysics.
9. Evidence-confidence ledger
| Claim | Tier | Anchor(s) |
|---|---|---|
| Cognition is describable as a system spanning brain, body, tools, others | Strong | Hutchins 1995; Wegner 1987 |
| People offload cognition; it is a value-based effort/reliability trade-off | Strong | Risko & Gilbert 2016 |
| Offloading reduces memory for offloaded items (transactive "Google effect") | Promising (general); Contested (specific Stroop result) | Sparrow et al. 2011; replication 2020 |
| Externally stored memory increases manipulation vulnerability | Promising | Cognition 2019 |
| Fragmentation/multitasking harms attention & memory ("wrong-way extension") | Strong enough to design around | Uncapher & Wagner 2018 |
| Retrieval/effort build durable capability (basis of the replace-vs-scaffold test) | Strong | Dunlosky et al. 2013 |
| The extended mind literally extends the boundary of mind | Contested (open) | Clark & Chalmers 1998 vs. Adams & Aizawa 2008; Rupert 2004 |
| "The human remains the author" is a testable design principle | Interpretation (operationalized, falsifiable) | this paper, §5 |
| Intimate future extensions (neural, spatial) can be built as augmentation | Vision (flagged) | Bush 1945; Engelbart 1962 |
10. Conclusion
The mind was never only in the head; it was in the tablet, the scroll, the book, the friend you called, the margin you wrote in. What is new is the intimacy and the direction of the next extension — and whether the person or the platform holds the dial. We define exocognition to name that layer plainly, credit the lineage it grows from, and bind ourselves to one testable law: the human remains the author. Built that way, the reach returns us larger. That is the wager, and we have written down the terms so we can be held to them.
References (selected)
- Adams & Aizawa (2008). The Bounds of Cognition. Blackwell.
- Bush (1945). As We May Think. The Atlantic. https://www.theatlantic.com/magazine/archive/1945/07/as-we-may-think/303881/
- Clark (2008). Supersizing the Mind. Oxford UP.
- Clark & Chalmers (1998). The Extended Mind. Analysis 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
- Dunlosky et al. (2013). Improving Students' Learning With Effective Learning Techniques. Psych. Science in the Public Interest. https://journals.sagepub.com/doi/abs/10.1177/1529100612453266
- Engelbart (1962). Augmenting Human Intellect: A Conceptual Framework. SRI. https://www.dougengelbart.org/content/view/138/
- Hutchins (1995). Cognition in the Wild. MIT Press.
- Risko & Gilbert (2016). Cognitive Offloading. Trends in Cognitive Sciences 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
- Rupert (2004). Challenges to the Hypothesis of Extended Cognition. Journal of Philosophy 101(8). https://www.jstor.org/stable/3655517
- Sparrow, Liu & Wegner (2011). Google Effects on Memory. Science 333, 776–778. https://www.science.org/doi/10.1126/science.1207745
- Hesselmann et al. (2020). No conclusive evidence that difficult general knowledge questions cause a "Google Stroop effect". A replication study. PeerJ 8:e10325. https://peerj.com/articles/10325/
- Uncapher & Wagner (2018). Minds and brains of media multitaskers. PNAS. https://www.pnas.org/doi/10.1073/pnas.1611612115
- Wegner (1987). Transactive Memory. In Theories of Group Behavior.
Draft for internal read. Verification (2026-07-21): the replication citation is confirmed — Hesselmann et al. (2020), PeerJ 8:e10325. Remaining before publication: (1) final scientist read-through; (2) confirm each remaining link resolves at publish time.