The thought is spread across the table
A person is trying to solve a difficult problem.
They begin with a blank page. Then a note is moved to the left. A number is written in the margin. A browser tab stays open because one fact must remain visible. A colleague says a sentence that changes the frame. The person stands up, walks to a whiteboard, draws three boxes, crosses one out, photographs the board, and later asks an AI system to compare two possibilities.
At the end, the solution feels like a thought.
It is difficult to say where the thought was.
Some of it was remembered.
Some of it was written down because it could not be remembered.
Some of it appeared only because the notes were arranged in space.
Some of it belonged to another person.
Some of it was produced through the constraints of the tools.
The environment is not only where thinking happens.
It is often part of how thinking happens.
That is the doorway into cognitive ecology.
Definition
Cognitive ecology, in the human-systems sense used here, refers to the network of internal and external resources through which cognition is organized in real environments.
The term has another established use in behavioral ecology, where researchers study relations among ecological conditions, cognition, behavior, and evolution across species. That literature should not be silently folded into the present meaning.
For the Institute, the closest foundations are distributed cognition, external cognition, cognitive offloading, embodied cognition, and research on human-computer interaction.
Edwin Hutchins, James Hollan, and David Kirsh argued that cognition can be studied at the level of whole systems in which people and artifacts coordinate information over time.
David Kirsh showed how people intelligently arrange space to reduce cognitive demand and make tasks easier to perform.
Evan Risko and Sam Gilbert define cognitive offloading as using physical action to alter a task's information-processing requirements and reduce cognitive demand.
Clark and Chalmers pushed the philosophical argument further with the extended-mind thesis, asking when external resources may count not merely as aids to cognition but as parts of a cognitive process.
These positions are not identical.
They share one important correction to an overly individual picture of intelligence: human thinking is deeply scaffolded.
The desk is a cognitive instrument
A desk can hold what working memory cannot.
A checklist can preserve a sequence under stress.
A diagram can make a relationship visible that prose hides.
A calendar can carry future intentions.
A library can extend memory across generations.
A colleague can hold expertise that no individual possesses.
A well-designed cockpit distributes attention across instruments, procedures, and crew.
The cognitive environment is therefore not decorative.
It changes what a person can do.
This is why poorly designed environments can make capable people appear less capable.
A confusing interface consumes attention.
An inaccessible filing system turns memory into search burden.
A meeting without a shared visual model forces participants to hold too much in mind.
A noisy notification environment repeatedly breaks task state.
Cognitive performance is partly a property of the system in which the person is embedded.
Offloading is not automatically decline
The language of offloading often becomes moralistic.
People once memorized phone numbers. Now they do not. People once navigated from maps. Now they use GPS. People once drafted without generative systems. Now many do not.
From this, it is easy to infer that external support weakens the mind.
Sometimes it may.
But offloading is also one of the ways humans become more capable.
Writing is cognitive offloading.
Notation is cognitive offloading.
Calendars, calculators, diagrams, indexes, and libraries all reorganize what must be carried internally.
Risko and Gilbert's review is useful precisely because it treats offloading as a strategy with benefits, triggers, and consequences rather than as evidence of laziness.
The more precise question is what is being offloaded and what remains necessary.
A calculator can free attention from arithmetic so a person can reason about a larger model.
A summary can free time from repetitive reading.
But if the offloaded process was also the process through which understanding formed, the trade-off changes.
The same assistance can be liberating in one task and developmentally costly in another.
AI changes the scale of the cognitive environment
Artificial intelligence is not simply another external memory aid.
It can participate in interpretation.
It can draft.
It can rank.
It can propose categories.
It can generate alternatives.
It can remember prior interaction.
It can translate between forms.
It can become conversational enough that the external cognitive resource begins to feel socially responsive.
This expands the cognitive ecology dramatically.
The person no longer only stores information outside themselves. They can externalize parts of problem formulation, synthesis, comparison, and expression.
This may increase capability.
It may also change metacognition.
When a system produces fluent material instantly, people need to judge not only whether an answer is correct, but whether they themselves understand the question well enough to evaluate the answer.
The ecology becomes more powerful while the burden of calibration becomes more subtle.
Cognitive ecology and agency
A well-designed cognitive environment can expand agency.
Assistive technologies can make communication possible.
External reminders can support people with memory difficulties.
Shared tools can make collective work more reliable.
Translation can open participation across language.
At the same time, cognitive environments can shape agency by deciding which options are easy to see.
A recommender is a cognitive aid and a choice architecture.
A dashboard is a memory aid and a theory of what matters.
A generative assistant is a drafting aid and a source of frames.
The Institute therefore treats cognitive ecology as closely related to technological mediation.
The environment helps think with us.
It can also think us into a narrower field if its assumptions remain invisible.
Institutions are cognitive ecologies
An organization is not only a collection of minds.
It is a system for distributing cognition.
Who knows what?
Where is knowledge stored?
How is uncertainty communicated?
Which facts reach decision-makers?
Which people are interrupted?
Which records persist?
Who is allowed to challenge a model?
Which forms of tacit knowledge never enter the database?
These are cognitive-ecology questions.
A resilient institution does not need every individual to know everything.
It needs the right information, memory, expertise, and judgment to become available at the right time without erasing responsibility.
AI can improve this ecology by making knowledge easier to retrieve and compare.
It can damage it if generated summaries replace source contact, if central systems make local knowledge invisible, or if everyone becomes dependent on the same opaque intermediary.
Cognitive monoculture
One risk deserves particular attention.
When many people use the same models, rankings, templates, and recommendation systems, cognitive support can become cognitive convergence.
People may begin from similar summaries.
Teams may receive similar alternatives.
Writers may inherit similar rhythms.
Decision-makers may rely on the same categories.
This does not mean AI necessarily homogenizes thought. Models can also generate diversity and expose users to unfamiliar perspectives.
But a cognitive ecology needs variation.
Redundancy, independent judgment, local knowledge, dissent, and different forms of expertise can protect a system against shared error.
The question is not only whether a cognitive tool makes one person more efficient.
It is what happens when the same tool becomes infrastructure for everyone.
Cognitive reserve in a tool-rich world
A healthy cognitive ecology does not ask every human being to perform every function unaided. That would confuse independence with capacity.
It asks a different question: what needs to remain available when the support is absent, wrong, compromised, or inappropriate?
A pilot uses instruments because instruments extend perception. A clinician uses records because memory should not carry an entire history alone. A researcher uses databases because scholarship depends on collective storage. The presence of external support is not evidence of diminished cognition. Often it is evidence of mature cognition, because skilled people know what should be externalized.
But mature systems also preserve reserve.
The pilot must still recognize when an instrument is implausible. The clinician must notice when the record and the person do not match. The researcher must be able to read a source closely enough to detect when a summary has removed the qualifying sentence. Reserve is the capacity that remains able to challenge the scaffold.
Generative AI makes this distinction unusually important because the external support no longer looks inert. It proposes, explains, ranks, completes, and anticipates. The ecology begins to answer back. That can increase capability dramatically, but it can also make the boundary between support and substitution difficult to perceive.
The research question is therefore not simply whether people perform better with a tool. It is whether the ecology leaves them able to notice tool failure, reconstruct enough of the task when necessary, and understand which parts of the result they have not independently examined.
Learning environments are special cognitive ecologies
The same tool can have a different developmental meaning in a workplace and a classroom.
In one setting, offloading a calculation may be sensible because the capacity being evaluated is higher-order judgment. In another, the calculation itself may be the practice through which the learner builds a representation that later judgment depends on.
This is why educational debates about AI become confused when they ask whether using a tool is simply good or bad. The relevant unit is the learning ecology: what the learner is supposed to become able to do, what practice is being bypassed, what feedback remains available, and whether assistance disappears gradually enough for independent capacity to emerge.
A finished answer can improve immediate performance while weakening the very struggle through which a novice learns what experts later see almost automatically. Conversely, well-designed support can reveal structure, reduce irrelevant load, and make difficult practice accessible to more people.
The difference cannot be read from the presence of technology alone. It has to be studied through transfer, retention, calibration, and what happens when the ecology changes.
How should cognitive ecology be studied?
There is no single cognitive-ecology score.
Researchers can examine task performance, error, memory, workload, attention, coordination, offloading behavior, trust, metacognitive confidence, transfer, skill retention, and recovery when tools fail.
Ethnographic methods are especially valuable because real cognitive systems often contain workarounds and tacit coordination that laboratory tasks miss.
Network and interaction analysis can show how information moves through teams and artifacts.
Longitudinal studies are needed when the question concerns development rather than immediate performance.
The unit of analysis must be chosen carefully.
Sometimes the person is the right unit.
Sometimes the person-plus-tool is.
Sometimes the team, workflow, or institution is the meaningful cognitive system.
What the evidence does not show
Distributed cognition does not prove that every tool is literally part of the mind.
The extended-mind thesis remains a philosophical position, not a settled empirical fact.
Cognitive offloading does not imply cognitive decline.
Better task performance with external support does not prove durable learning.
Worse unaided performance after tool use does not by itself prove that the tool is harmful if the person's real environment normally includes the tool.
And cognitive ecology does not remove individual responsibility by declaring the system the true thinker.
Responsibility can be distributed without disappearing.
What this changes for Inner Technology
Cognitive ecology changes the meaning of “inner” in Inner Technology.
It does not make the inner life disappear. It shows why capacities cannot be understood as private possessions that a person simply carries from room to room unchanged.
Attention is supported or consumed by environmental structure. Memory is redistributed through notes, devices, colleagues, and routines. Judgment is shaped by the sequence in which information becomes available. Agency depends partly on whether the surrounding ecology leaves room to inspect, override, and reframe.
This means practice architecture cannot end with the exercise. A person may learn a capacity in one environment and lose access to it in another whose rhythms, incentives, and tools continually train the opposite response.
For the Institute, the practical research question becomes: what ecology makes this capacity usable when it is actually needed?
That question also disciplines AI claims. If AI improves system performance, we still need to ask where the cognition moved, which human capacities became less necessary, which became more necessary, and how the system behaves when the tool is wrong or unavailable.
Designing an ecology worth thinking in
Return to the table.
The notes are still there.
The colleague's sentence remains in the margin.
The AI output is one object among others.
The person makes the final mark by hand.
Human intelligence has always been ecological.
The AI age makes the ecology more active, more responsive, and more difficult to see as environment.
The task is not to pull cognition back inside the skull.
It is to become more deliberate about the systems with which we think.
Research references
- Hutchins, E. (2010). Cognitive Ecology. Topics in Cognitive Science, 2(4), 705–715. https://doi.org/10.1111/j.1756-8765.2010.01089.x
- Hollan, J., Hutchins, E., & Kirsh, D. (2000). Distributed cognition: toward a new foundation for human-computer interaction research. ACM TOCHI, 7(2), 174–196. https://doi.org/10.1145/353485.353487
- Risko, E. F., & Gilbert, S. J. (2016). Cognitive Offloading. Trends in Cognitive Sciences, 20(9), 676–688. https://doi.org/10.1016/j.tics.2016.07.002
- Dunn, T. L., & Risko, E. F. (2016). Toward a Metacognitive Account of Cognitive Offloading. Cognitive Science, 40(5), 1080–1127. https://doi.org/10.1111/cogs.12273
- Kirsh, D. (1995). The Intelligent Use of Space. Artificial Intelligence, 73(1–2), 31–68. https://doi.org/10.1016/0004-3702(94)00017-U
- Clark, A., & Chalmers, D. (1998). The Extended Mind. Analysis, 58(1), 7–19. https://doi.org/10.1093/analys/58.1.7
- Mettke-Hofmann, C. (2014). Cognitive ecology: ecological factors, life-styles, and cognition. WIREs Cognitive Science. https://doi.org/10.1002/wcs.1289