We're developing new cognitive abilities. We just don't know what they are yet.

Something is shifting in how we think.

For most of human history, the cognitive tools we used were external to the thinking itself. A pen didn’t participate in the argument; it recorded it. A calculator didn’t understand the problem; it processed a computation someone else framed. The thinking happened inside the person, and the tool extended a specific, bounded part of what the person could do with that thinking. That boundary has always been fairly clear, but is becoming less so with the advent of AI.

An architect sketching a building asks AI to generate variations on her initial concept. Within minutes she’s looking at dozens of options, a volume of exploration that would have taken days to produce by hand. She scans them, recognizing which ones are worth pursuing, selects elements from several, combines them, iterates again. The final design is hers, and it also emerged from a process that didn’t exist five years ago. Which parts came from her mind and which from the interaction?

A lawyer preparing a brief asks AI to find weaknesses in her argument. In seemingly no time at all, she’s looking at a list of potential vulnerabilities, lines of attack she’ll need to address before opposing counsel finds them. She works through each one, discarding some as unlikely, recognizing others as real problems. The third vulnerability leads her to a case she hadn’t considered, which ends up strengthening her central claim. Which weaknesses would she have found on her own? Which would have surfaced only in court, diminishing the preparation of her case?

In both cases, the thinking was distributed across the exchange. The architect and the lawyer provided direction and judgment; the AI provided speed and volume. The middle of the process, where possibilities became outcomes, didn’t happen entirely inside their heads or entirely inside the machine. Neither retrieved a finished product from the AI, and neither arrived there independently. The final work emerged in the space between, through iteration neither could have conducted alone. The boundary between “my thinking” and “assisted thinking” has become harder to locate.

---

---

How Tools Have Always Shaped Cognition

Tools have always extended human cognitive capacity, but with a consistent boundary: the human remained the locus of understanding. A calculator performs a computation that the human frames and interprets. A telescope magnifies light that the human has already aimed at something meaningful. Each tool handles a defined operation, while the human handles everything that makes the operation matter. The thinking had a traceable address.

AI crosses that boundary. Large language models process language, recognize patterns, and generate synthesis in ways that functionally replicate cognitive activities previously requiring human minds. A spreadsheet doesn’t make an argument; it calculates one that the human has already framed. A search engine doesn’t analyze a question; it retrieves documents that might be relevant. But a researcher using AI can explore multiple analytical approaches simultaneously, testing hypotheses at speeds impossible for unassisted cognition, and what comes back isn’t raw data to be interpreted but reasoning to be evaluated. That’s a different kind of tool. It engages with meaning, produces language that responds to context, and generates the kind of output that used to require a mind because it involved judgment. The question of where the thinking happens has become difficult to answer, and that difficulty is new. Previous tools never raised it, because they never could.

Photo by Amsterdam City Archives on Unsplash

The Brain That Reorganizes

The brain reorganizes throughout life, forming new neural connections in response to what we learn and how we spend our attention. This isn’t metaphor; it’s measurable. London taxi drivers, who spend years memorizing their city’s labyrinthine street layout, show measurably enlarged posterior hippocampi, the region involved in spatial navigation, compared with people who don’t drive cabs. When drivers retired and stopped navigating, those differences reversed over time. The capacity grew because it was being used, and receded when it no longer was. Cognition isn’t a fixed endowment that tools either help or hinder. It’s a living system that physically restructures itself around whatever we repeatedly ask it to do. Which means that when we change how we manage and process information at scale, we don’t just change our habits. We change our minds.

The history of writing is the clearest example we have of this at civilizational scale.

Before writing existed, language itself had no external storage system. Objects and markings could record quantities and sequences, but they couldn’t hold an argument, preserve a narrative, or transmit the kind of knowledge that lives in words. For everything that required language, memory was the only vessel available. The oral poets who performed the Iliad and Odyssey weren’t working from fixed texts; there were none at the time. According to the oral-formulaic theory developed by scholars Milman Parry and Albert Lord, they composed in performance, drawing on a deep repertoire of stock phrases, epithets, and scene structures that could be combined and recombined as the narrative demanded. The poem itself wasn’t stored anywhere, rather it lived in the poet’s capacity for reconstruction. Knowledge lived in minds, which meant the most valuable cognitive capacity was the ability to hold it there. The wise person was the one with deep memory, who could summon the past into the present through speech.

Then writing arrived, and knowledge could suddenly exist outside of anyone’s head. Ideas became objects, storable and examinable at a distance from the moment of their making. Socrates warned, in Plato’s Phaedrus, that writing would weaken memory and create the illusion of knowledge without its substance. He was partially right: the capacity to hold vast bodies of knowledge in mind did fade across literate populations. But writing also enabled forms of thought that oral culture couldn’t support. Systematic philosophy became possible, as did arguments that could be examined and revised long after their first speaking, logic tested across time rather than in the heat of live debate. Science as a cumulative enterprise became possible, because observations could be stored and built upon across lifetimes.

Reading, though, is something the brain was never built for. Written language has only existed for around 5,000 years, nowhere near long enough to have shaped human evolution. The brain had to improvise. Brain imaging studies by neuroscientist Stanislas Dehaene and colleagues show what that improvisation looked like: learning to read causes a specific region in the left occipito-temporal cortex to become dedicated to recognizing written words. In people who’ve never learned to read, that same brain area responds primarily to faces and objects. The region exists in everyone; what changes with literacy is what it does. As it specializes for written words, face-processing in that area decreases, with face responses gradually shifting toward the right hemisphere. The written word gets processed in the same neural neighborhood that once handled faces. Literacy repurposes real estate that evolution assigned to something else entirely. What this demonstrates, concretely, is that the brain doesn’t just use cognitive technologies - it reorganizes around them. The tool changes the organ.

The printing press, around six centuries ago, produced a further reorganization. Before printing, books were rare and expensive; reading was primarily oral and communal. Printed books enabled solitary reading on a mass scale. We could sit alone in silence for hours, following a single line of argument without interruption. The ability to follow a complex argument across hundreds of pages while holding its structure in mind isn’t innate; it’s a trained capability that emerged from generations of interaction with a new cognitive technology.

Digital technologies continued the pattern. Research by psychologist Betsy Sparrow and colleagues showed that when people expect to have access to information later, they’re less likely to encode it in memory. We stopped memorizing phone numbers and directions; we started remembering where to find information, letting location substitute for the information itself. Neural systems optimize for the environment they operate in.

Each transition produced the same dynamic: new cognitive tools changed how we think, with old capacities fading and new ones emerging. The people living through those transitions rarely saw clearly what was being gained to replace what was being lost. The new capacities only became visible after they’d already taken hold.

---

---

The Co-Evolution Underway

There’s a pattern running through all of this. Humans develop tools to extend what cognition can do. Those tools change what kinds of cognition are in demand. Brains then reorganize around what’s in demand, freeing up resources for the next layer of complexity. Writing stored memory externally, which freed cognitive resources for more abstract thought. Print multiplied books, but its deeper effect was to train the sustained linear attention that makes extended analytical reasoning possible. Each technology cleared cognitive space by offloading something the brain no longer had to carry alone.

AI might represent a more fundamental version of this dynamic. Previous technologies extended specific cognitive capabilities: writing extended memory, calculators extended arithmetic. AI extends something more central to what makes human cognition distinctive: the capacity for language-based reasoning itself. When a tool extends arithmetic, the cognitive reorganization is limited. When a tool extends language-based reasoning, the reorganization could be more fundamental.

Cognitive scientists and philosophers of mind exploring human-tool relationships have raised the possibility of cognitive co-evolution, the idea that humans may develop new ways of thinking that only function in partnership with AI. A literate person doesn’t simply have better memory than an oral poet; she has a different relationship to knowledge and a different way of building arguments. That same kind of qualitative shift may characterize what’s developing now.

These capabilities will likely emerge through practice and environmental pressure, becoming visible only after neural reorganization has already occurred in populations extensively using these tools. We didn’t design the capacity for sustained linear reading; it emerged from generations of interaction with printed books. The cognitive capacities that emerge from AI interaction are likely to surprise us in similar ways.

The Practices Taking Shape

We’re early in this transition. But what we can observe, drawing on what every previous cognitive shift has taught us, is where the reorganization is likely happening: in the specific practices through which people are engaging with these tools daily. Practices become habits, habits shape neural pathways, and neural pathways define what cognition can do. This is where new capacities form, not in intention or prediction, but in the accumulated texture of how people are actually using AI right now. And several of those practices map onto something familiar: the same skills that have always been required to navigate information environments where the source is powerful, fast, and not inherently accountable to our specific needs.

Evaluating AI outputs against our own thinking.

AI-generated text often sounds plausible without being right. Distinguishing "this sounds good" from "this captures what I mean" requires having done enough prior thinking to know what we mean. The architect scanning dozens of variations knows instantly which ones are worth pursuing because she has a clear internal sense of what the building needs to accomplish. This is source evaluation at speed: assessing whether something answers our actual question or merely appears to.

Using AI outputs as starting points.

AI generates material to react against and refine. The doctor who uses AI-suggested diagnoses to surface possibilities, then integrates them with her direct knowledge of the patient, is treating AI as a collaborator in her thinking process. The sources inform and provoke thinking without supplanting analysis.

Preserving the cognitive work that matters.

The same tool, used at different points in a process, produces different outcomes. A student who works through a problem herself and then uses AI to check her reasoning comes away with understanding she built; a student who asks AI to solve the problem first comes away with an answer she didn’t earn. What changes isn’t the tool. What changes is whether she engaged with the problem before reaching for help.

Owning the conclusion.

Working with AI involves a continuous stream of decisions about what to accept, what to revise, and what to reject. Those decisions compound. A lawyer who uses AI to pressure-test her argument and accepts each suggested vulnerability without fully evaluating it may end up with a brief that hangs together on the surface, but doesn't reflect her actual read of the case. What develops through this kind of work is something more than familiarity with a tool: a capacity to let AI expand what we can see while keeping our own judgment as the thing that decides what to do with that expanded view. The conclusion we reach should be more developed than where we started, and fully ours to stand behind.

These practices are where the pressure is being applied. Like every previous cognitive technology, what emerges from them will be shaped by what people actually do, not by what anyone designed or predicted. The underlying capacities, whatever neural reorganization is underway, will only become clearly visible later. But we can already speculate, based on where the friction is, about what those capacities might look like.

Photo by Rafael M. on Unsplash

What Might Be Emerging

The historical pattern offers one guide: the capacity that emerges often looks different from the practice that produced it. Sustained engagement with printed books didn’t produce “better reading.” It produced the capacity for extended linear argument, for following a single thread of thought across hundreds of pages while holding its structure in mind. The practice was reading; the capacity that emerged was something far broader.

If the pattern holds, the capacities emerging from AI practices won’t simply be “better evaluation” or “better synthesis.” They’ll be reconfigurations we can only name after they’ve already taken shape. Some possibilities, based on where the pressure is currently being applied:

Sharper discernment. The practice of evaluating AI outputs at high volume may train a faster, more reliable sense of what's true and useful. AI outputs can be fluent and wrong simultaneously, which means the person using AI must supply the skepticism, detecting subtle misalignments between what sounds right and what is right. People who do this repeatedly are exercising judgment hundreds of times over the course of their AI activity. That capacity would transfer far beyond AI: a more finely tuned ability to assess any claim or argument, regardless of how smoothly it's presented.

Thinking across larger realms of possibility. The architect scanning dozens of variations isn't just working faster; she's navigating a landscape of options that sequential thought couldn't produce. We've had tools that extended how far we could see or how much we could calculate, but we haven't had tools that extended how many possibilities we could hold in view at once. The capacity developing here might be a new form of creative cognition: generating breadth through AI, then applying judgment to traverse that space.

Dialogue with externalized reasoning. Writing let us see our thoughts on a page. AI lets us see our reasoning reflected back and transformed. We can propose an argument, watch it restated in different terms, push back, and refine through exchange. This is different from revising a piece of work alone or debating another person. The AI doesn't have stakes or defensive reflexes. The capacity developing here might be a new form of thinking-through-conversation, where the friction of another person’s ego, defensiveness, or competing agenda is removed entirely, leaving the reasoning itself as the only thing to contend with.

Rapid iteration on complex work. The longer it takes to test an approach, the fewer approaches we test. When a methodology takes weeks to set up, the setup itself is the commitment. When the same exploration takes minutes, we can afford to start before we're certain, treating early attempts as information rather than output. We can be wrong faster, learn from that, and try again. The capacity developing here might be a more exploratory relationship to complex problems: generating and testing rather than planning a single route before we've tried any others.

Gains and losses have always run together through these transitions. The oral poet’s capacity for prodigious memorization faded with writing, but writing unlocked the systematic accumulation of knowledge across generations. The deep familiarity built from returning, again and again, to a small number of texts diminished with print, but print made new forms of individual reasoning possible. Digital tools displaced the spatial memory we once used to navigate, but opened access to information at a scale no prior generation could reach. Each technology made certain capacities less necessary; as those capacities faded, they were replaced by something the previous era couldn’t have named. In none of these transitions did the losses simply accumulate. Something was consistently freed up, and what filled that space reshaped what human minds could do next.

What We Can Shape

The brain will reorganize around what we repeatedly do. Neural pathways form around whatever practices become habitual, whether we're paying attention to that process or not. This has been true through every cognitive transition in the historical record. The monks who copied manuscripts by hand developed forms of close attention that print eventually made less culturally necessary. The urban residents who gave up navigating by mental map and switched to GPS showed measurable changes in how they engaged spatial reasoning over time. The technology shaped cognition not through a single dramatic rupture but through the slow accumulation of what people did every day. What was practiced persisted; what was no longer practiced faded.

Passive adaptation and active participation can feel similar from the inside, especially when the tools are useful and the change is gradual. What separates them is whether we’re bringing our own thinking to the exchange or simply receiving what the exchange produces, and that distinction matters more with AI than it did with previous information sources. A book or article is fixed; we can evaluate it against external criteria and check it against other sources. An AI source responds to us, calibrating itself to sound credible and helpful, which makes the old critical distance harder to maintain. The evaluative work has to come from inside the exchange itself: having thought enough about a question that we can recognize a good answer when we see one, and notice when we’re being led somewhere we didn’t intend to go. An architect scanning AI-generated variations and a lawyer pressure-testing her argument with AI are both doing exactly that, bringing clear intent to the interaction and evaluating every output against what they actually needed. The boundary between their thinking and the AI’s remained blurry, but the judgment about which outputs mattered was entirely theirs.

Every cognitive technology in the historical record produced capacities that exceeded what anyone anticipated, because the technology changed not just what people could do but how they thought about doing it. We’re inside that process now, which means we can’t see its full shape any more than the first literate Greeks could see what reading would eventually make possible. What we can do is engage deliberately, bringing real questions to these exchanges and holding our own reasoning as the standard against which we measure what comes back. The capacities that emerge from this transition will be shaped by exactly that: not by the tools themselves, but by the quality of thinking we bring to using them.

---

---

Have you read the Founding Member Report: The State of AI yet?

A comprehensive guide for information navigators who want to understand where AI is actually heading and what it means for how we find, evaluate, and use information in 2026.

→ Find out more here.

---