Where Does Your Intelligence Live?

First position

A thought can appear before it has a purpose

Much writing about thinking-with-tools assumes a task: a brief, a problem statement, a question already well enough formed to be typed. That assumption is convenient and often wrong. A good deal of what later turns into work does not start as a task at all. It starts as an interruption — an oddness noticed in passing, a phrase that will not settle, a picture with no argument attached, a suspicion that two unrelated things are somehow the same shape.

• “Wait —” a pause before a reason for it
• A question with no field attached yet
• A hunch confidence arriving before the argument
• An image a shape that precedes the sentence
• An association two things that should not meet
• An objection something is off, and it is unnamed

None of these is yet an intention, and none is yet an output. Calling them “thoughts” here means nothing mystical: they are ordinary mental events, unremarkable in themselves, and interesting only because of what may or may not happen to them next. The claim in this essay is modest and empirical in spirit — that a thought of this kind has a subsequent history, and that the medium it enters shapes that history.

What the research supports
Established

Thinking already happens partly outside the head

External resources can take part in the thinking, not only record it

Andy Clark and David Chalmers argued in 1998 that if an external resource plays the role an internal belief would play — reliably available, readily consulted, generally trusted — there is no principled reason to draw the boundary of the cognitive system at the skull. Their example is Otto and his notebook. This is a philosophical argument, not an experiment, and it remains contested. What it grants is licence to treat the page, the sketch and the conversation as parts of a thinking process rather than as containers filled after the thinking is done. (Clark & Chalmers, 1998)

Some actions are performed to think, not to accomplish

David Kirsh and Paul Maglio, studying expert Tetris play, distinguished pragmatic actions — those that advance one towards a goal — from epistemic actions, performed to make computation easier, faster or more reliable: rotating a piece on screen rather than mentally, because looking is cheaper than imagining. The study is narrow, and its evidence comes from one task with a particular structure. But the distinction it draws is the one that matters here: turning a thought over externally can be part of the thinking itself, not a report on it. (Kirsh & Maglio, 1994)

Offloading is easy, useful, and changes what stays with the person

Evan Risko and Sam Gilbert's 2016 review describes cognitive offloading as the use of physical action to reduce information-processing demands. People offload readily, and their decisions to do so track their beliefs about their own memory, which are not always accurate. The review is careful about the trade-off: offloading can improve performance on the task at hand while altering what is retained internally. It is a review of behaviour, and it makes no recommendation about how external thinking should be arranged. (Risko & Gilbert, 2016)

Taken together, these three lines of work make a background claim that is now fairly safe: external representation participates in thinking, and the properties of the external medium are therefore not neutral.

What changed

The medium can now answer back

A notebook does not object. A map does not propose an alternative route unprompted. The external media of thought have historically been patient and silent: they held a thought steady so it could be looked at again. A conversational model is a different kind of external medium, because a partial thought put into it comes back with something attached — an example, a name for the thing, a counter-case, an adjacent literature, five directions of which one is interesting.

The most defensible way to describe this is in terms of reach. A person searching alone is limited by what they already know to look for; the vocabulary they have determines the questions they can pose. A responsive medium widens the accessible search space around a thought — it puts within reach material that the person would plausibly not have retrieved, and framings they had no term for. That is a real change, and it should not be overstated. It is not omniscience, not correctness, and not judgment. Widened reach means more of the space is available; it does not mean the useful part of the space has been found, and it does not settle which of the offered directions deserves the person's attention.

• An example from a field the person had no reason to know existed.
• A name for a pattern that had been noticed but not labelled.
• A counter-case that makes the hunch weaker and more interesting.
• A framing that is wrong, and useful precisely because it is wrong.
• Five plausible continuations, delivered before the person has decided what they were doing.

The last item is not a complaint about a technology. It is a description of a condition. The same responsiveness that widens reach also arrives with a gradient: it is easier to accept a proposed direction than to stay with an unformed one.

A wrong turn worth naming

This is not, at bottom, a storage problem

There is an obvious reading of everything above in which the difficulty is retrieval: thoughts get lost, so keep better records. That reading is not false, and it is not sufficient. Notes, archives and memory features address findability — whether a thought can be located again. They do not address whether a thought that has not yet decided what it is can go on developing.

Those are different problems. A perfectly retrievable record of a half-formed idea can still be a dead record, because the part that mattered — the pull of it, the reason it seemed worth noticing, the direction it was leaning — was never the text. The question this essay is after is not where did it go but what was it still becoming, and whether the conditions it met allowed it to keep becoming anything at all.

What the research supports
With caveats

A thought can go quiet without being dead

One of the more durable findings about creative work is that setting a problem aside and returning to it can help. Benjamin Baird and colleagues, in a 2012 experiment, gave participants an incubation interval filled with either a demanding task, an undemanding task, rest, or no break, and found improved performance on creative-uses problems that had been encountered before the break — but only for the undemanding-task condition, and not for problems seen for the first time after it. Self-reported mind wandering was higher in that condition. It is a single experiment with a specific paradigm; the effect it reports is real within that design and should not be inflated into a general claim that distraction makes people creative. (Baird et al., 2012)

Simone Ritter and Ap Dijksterhuis reviewed the incubation literature in 2014 and are explicit that the mechanism is unsettled: candidate explanations include unconscious processing, the weakening of a fixating first approach, and chance re-encounters that let a stalled problem be re-cued. Their own account favours unconscious contribution, and it is an argued position within a contested field rather than a settled result. (Ritter & Dijksterhuis, 2014)

More recent work continues in the same direction without closing it. A 2025 preregistered study by McDaniel and colleagues found that self-reported mind wandering during an incubation period predicted subsequent increases in creative performance on a writing task. It is a within-study predictive relationship, based on self-report probes, in one task and one session — evidence that the phenomenon is worth taking seriously, not that the mechanism has been identified. (McDaniel et al., 2025)

This is the point at which the responsive medium becomes interesting rather than merely convenient. Systems are generally optimised to converge: to take an underspecified input and return something finished. Convergence is what makes them useful. It is also, structurally, the opposite of incubation.

Two-sided

Encountering other people's ideas helps and constrains

The romantic account of influence — that seeing what others have done opens the mind — is half right, and the experimental literature has documented the other half for decades.

Examples constrain

David Jansson and Steven Smith's 1991 studies of design fixation found that participants shown an example solution reproduced features of that example in their own designs, including features that had been explicitly identified as flawed. The studies are on conceptual design tasks with student and professional participants; the effect is robust enough that “design fixation” became a standing term. (Jansson & Smith, 1991)

Examples also produce

Joel Chan, Steven Dow and Christian Schunn analysed ideas and their cited inspiration sources in a large online ideation community, and found the relationship between the conceptual distance of a source and the creativity of the resulting idea to be considerably more complicated than the standard “far analogies are best” account predicts — with near sources performing better than that account allows. This is field data and correlational; it constrains a theory rather than establishing a rule. (Chan, Dow & Schunn, 2015)

Kosa Goucher-Lambert, Jonathan Moss and Jonathan Cagan, combining behavioural measures with neuroimaging in design ideation, likewise report that inspirational stimuli affect what people generate, and that the usefulness of “near” versus “far” stimuli depends on the problem rather than following a simple rule. (Goucher-Lambert, Moss & Cagan, 2019)

The honest summary is that exposure to other ideas is neither inspiration nor contamination in general. It is both, and which one dominates depends on timing, on the form of the exposure, and on what the person is being asked to do with it. Any medium that supplies examples abundantly is therefore operating on both sides of that line at once.

The same tension, amplified

What experiments with generative systems show

Recent work reproduces the inspiration–fixation tension in generative settings, and adds a detail worth holding onto: the results depend heavily on how the encounter is arranged.

Jack McGuire, David De Cremer and Tim Van de Cruys, writing in Scientific Reports in 2024, found that the arrangement of creative collaboration matters — co-creating an output, and a person's belief in their own creative capability, were both consequential for the work and how it was judged. Jennifer Heyman and colleagues evaluated Supermind Ideator, which wraps a language model in structured prompts drawn from collective-intelligence research, and found that participants using the scaffolded tool produced ideas judged more creative than those produced without it. Both are experimental results on specific tasks over short horizons; both support the narrow claim that structure placed around a model changes what people get out of it. (McGuire et al., 2024); (Heyman et al., 2024)

Samangi Wadinambiarachchi and colleagues, at CHI 2024, ran a visual ideation experiment in which participants supported by AI image generation showed greater fixation to the examples they saw and produced fewer, less varied and less original ideas than a baseline condition. This is one controlled study of one modality, and the mechanism it implicates — fixation to salient examples — is the same one Jansson and Smith described in 1991. (Wadinambiarachchi et al., 2024)

At the aggregate level, a 2025 systematic review and meta-analysis by Nicolas Holzner, Sebastian Maier and Stefan Feuerriegel, covering 28 studies and 8,214 participants, reports a modest positive effect of generative AI on individual creative performance alongside a substantial negative effect on the diversity of ideas produced. It is a preprint at time of writing and should be read as such, but the shape of the finding is consistent with the experimental work above: better individual outputs, a narrower collective space of them. (Holzner, Maier & Feuerriegel, 2025)

The point is not that these systems help or harm creative thought. The evidence does not support either verdict as a general claim. The point is that the design of the encounter — when examples appear, how many, how finished they look, and whether the person is positioned as author or recipient — appears to determine which side of the inspiration–fixation line a given interaction falls on.

An inference
Not a finding

Is a finished form the end of a thought?

Here the essay moves past what the cited research establishes, and should be read as reasoning rather than evidence. A thought that reaches a complete form — an essay, a design, a decision, an answer — has certainly reached a state. Whether it has reached a terminal one is a different question, and it is not one the literature has tested directly.

What the literature does supply is circumstantial: incubation research shows that a set-aside problem can be improved by return; inspiration research shows that existing ideas function as sources for later ones; revision is the ordinary condition of most serious work. If those are typical rather than exceptional, then the more plausible reading is that a completed form is often a milestone in a longer process of revision, branching and recombination, and only sometimes an ending. This is an inference about how work behaves, not a claim that thoughts possess a life of their own.

Open

Ideas lead to other ideas

The inspiration literature supports a further claim that is easy to state and hard to handle: prior ideas act as sources for subsequent creative work. Chan and colleagues could analyse this precisely because their participants cited what had inspired them; Goucher-Lambert and colleagues manipulated stimuli directly. In both cases an idea's effect extended beyond the person who had it.

That is not the same as copying, and the fixation results are the reason to keep the distinction sharp. A source can open a direction the person would not have found, and a source can also close the space around a first plausible answer. The research does not offer a reliable way to tell these apart in advance, and it offers nothing at all about how such influence should be acknowledged when it occurs at the speed and volume that responsive media allow.

So the question of lineage stays open in two directions: what a thought owes to what preceded it, and what it might yet be the beginning of.

The gap
Our synthesis

What none of this settles

Placed side by side, these literatures describe a thinker whose thinking runs partly through external things; who offloads readily and imperfectly; whose problems can improve when set aside; who is helped and constrained by examples; and who, working with a responsive system, tends toward better individual outputs and narrower collective ones. Each study takes a task, an interval or a session as its unit.

None of them takes as its unit the thing this essay has been following: a thought that has not yet decided what it is, tracked across the returns, revisions, branches and long silences through which it might become something. There is no established account of what conditions allow such a thought to stay open, no measure of what is lost when it is resolved early, and no shared vocabulary for the lineage that connects one person's idea to what it later triggers.

The gap, then, is not a missing finding. It is a missing unit of analysis — and, downstream of that, a missing design question.

The narrowing

Where this leaves the question

Sources

References

1. Clark, A., & Chalmers, D. (1998). The Extended Mind. Analysis, 58(1), 7–19. Philosophical argument; contested. consc.net/papers/extended.html
2. Kirsh, D., & Maglio, P. (1994). On Distinguishing Epistemic from Pragmatic Action. Cognitive Science, 18(4), 513–549. Empirical study of a single task domain. doi.org/10.1207/s15516709cog1804_1
3. Risko, E. F., & Gilbert, S. J. (2016). Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676–688. Review of empirical literature. doi.org/10.1016/j.tics.2016.07.002
4. Baird, B., Smallwood, J., Mrazek, M. D., Kam, J. W. Y., Franklin, M. S., & Schooler, J. W. (2012). Inspired by Distraction: Mind Wandering Facilitates Creative Incubation. Psychological Science, 23(10), 1117–1122. Experimental; condition-specific effect. doi.org/10.1177/0956797612446024
5. Ritter, S. M., & Dijksterhuis, A. (2014). Creativity—the unconscious foundations of the incubation period. Frontiers in Human Neuroscience, 8, 215. Review and argued position; mechanism unsettled. doi.org/10.3389/fnhum.2014.00215
6. McDaniel, M. A., et al. (2025). Mind wandering during creative incubation predicts increases in creative performance in a writing task. Scientific Reports. Preregistered; self-report measures, single session, one task. doi.org/10.1038/s41598-025-09736-y
7. Jansson, D. G., & Smith, S. M. (1991). Design fixation. Design Studies, 12(1), 3–11. Experimental; conceptual design tasks. doi.org/10.1016/0142-694X(91)90003-F
8. Chan, J., Dow, S. P., & Schunn, C. D. (2015). Do the best design ideas (really) come from conceptually distant sources of inspiration? Design Studies, 36, 31–58. Field data; correlational. doi.org/10.1016/j.destud.2014.08.001
9. Goucher-Lambert, K., Moss, J., & Cagan, J. (2019). A neuroimaging investigation of design ideation with and without inspirational stimuli. Design Studies, 60, 1–38. Behavioural plus neuroimaging; task-specific. doi.org/10.1016/j.destud.2018.07.001
10. McGuire, J., De Cremer, D., & Van de Cruys, T. (2024). Establishing the importance of co-creation and self-efficacy in creative collaboration with artificial intelligence. Scientific Reports, 14, 18525. Experimental; task-specific. doi.org/10.1038/s41598-024-69423-2
11. Heyman, J. L., Rick, S. R., Giacomelli, G., Wen, H., Laubacher, R. J., Taubenslag, N., Knicker, M., Jeddi, Y., Ragupathy, P., Curhan, J., & Malone, T. W. (2024). Supermind Ideator: How scaffolding human–AI collaboration can increase creativity. Collective Intelligence. Evaluation of one tool. doi.org/10.1177/26339137241305117
12. Wadinambiarachchi, S., Kelly, R. M., Pareek, S., Zhou, Q., & Velloso, E. (2024). The Effects of Generative AI on Design Fixation and Divergent Thinking. CHI '24. Controlled experiment; visual ideation. doi.org/10.1145/3613904.3642919
13. Holzner, N., Maier, S., & Feuerriegel, S. (2025). Generative AI and Creativity: A Systematic Literature Review and Meta-Analysis. arXiv:2505.17241. Preprint meta-analysis; not peer reviewed at time of writing. arxiv.org/abs/2505.17241

Entries 1 and 5 are cited as argument, not as evidence; entry 13 is a preprint. Paraphrases throughout are the authors' summaries and are not presented as quotations. The synthesis, the inference in “Is a finished form the end of a thought?” and the gap identified in this essay are ours, and are not findings of the works cited.