Welcome to the Research and Strategy Services at in today's fast-paced.

For people who regularly use AI, an interesting contradiction may be emerging. Tasks that once required hours of research, writing, or problem-solving can now be completed in minutes. At the same time, AI can make it possible to explore unfamiliar scientific questions, develop complex projects, and engage with ideas that previously seemed beyond reach.
These experiences can produce two quite different impressions. On one hand, AI appears to be expanding what people can accomplish intellectually. On the other, it can raise a quieter question: if a machine is doing more of the thinking, what happens to our own cognitive abilities?
Much of the public discussion has framed this as a choice between two possibilities. Either AI makes us smarter by extending our capabilities, or it makes us less capable by encouraging mental shortcuts. Research findings are often presented as evidence for one side or the other, even when the studies are examining very different aspects of cognitive performance.
But there is a more useful way to understand what is happening. AI changes the distribution of cognitive work between the human and the machine. Whether that change is beneficial depends not simply on how much thinking is saved, but on which mental processes are being replaced, which remain active, and what new forms of thinking become possible.
Human cognition has always relied on external tools. Writing allows us to store information without remembering everything. Calculators reduce the need for mental arithmetic. Navigation systems help us reach unfamiliar destinations without constructing a detailed route ourselves.
These technologies reduce certain cognitive demands, but we do not ordinarily interpret every reduction in mental effort as a loss of intelligence. The important question is what the saved effort allows us to do, and whether the delegated skill remains important to our goals.
AI introduces a more complex version of this relationship. Unlike a calculator, a large language model can generate explanations, organize arguments, compare evidence, propose solutions, and evaluate competing ideas. It can therefore take over activities that previously required substantial reasoning, not merely memory or computation.
Consider someone preparing a business proposal. AI might assemble background information, organize the document, and produce a polished first draft. The person may spend far less time on the task, yet still exercise considerable judgment in deciding whether the recommendations are appropriate.
Alternatively, they might accept the output with little examination, delegating both the analysis and much of the judgment. The visible result could look similar, but the underlying cognitive activity would be quite different.
Less mental effort does not necessarily mean less meaningful thinking. Equally, producing sophisticated work does not necessarily mean the person has developed a sophisticated understanding of it.
There are legitimate reasons to investigate whether extensive AI assistance changes cognitive engagement.
A 2025 study from Microsoft Research and Carnegie Mellon University examined 319 knowledge workers and 936 examples of generative AI use. Researchers found that greater confidence in AI was associated with less self-reported critical-thinking effort, while greater confidence in one's own abilities was associated with more critical engagement.
Interestingly, the researchers also observed a shift in the nature of critical thinking. Rather than focusing primarily on producing information, workers increasingly described checking AI outputs, integrating responses, and overseeing the overall task.
This suggests that AI may not simply eliminate thinking. It can change where thinking takes place.
Another widely discussed study from MIT Media Lab examined essay writing under different conditions, including writing unaided and writing with ChatGPT. Researchers reported differences in EEG connectivity patterns, recall, and participants' sense of ownership over their work.
These findings raise meaningful questions about what happens when people delegate substantial parts of a learning-intensive task. However, differences in brain activity during an experiment are not equivalent to evidence of declining intelligence or permanent cognitive damage.
The distinction matters because a reduction in cognitive activity during a particular task could reflect several possibilities, including reduced engagement, altered strategies, or greater reliance on external assistance. Determining what these changes mean for lasting cognitive capability requires additional evidence.
The existing research therefore gives us reasons to examine how AI influences thinking, but not a sound basis for declaring that ordinary AI use is making people generally smarter or less intelligent.

One useful way to explore these differences is to consider three broad patterns of AI interaction. These are explanatory categories rather than a scientifically validated classification, and most real-world interactions involve some combination of them.
The first is cognitive substitution. Here, AI performs much of the reasoning-intensive work and the person primarily receives or applies the result. Asking an LLM to summarize a long document, produce a routine report, or complete an administrative task can be an efficient and entirely appropriate use of the technology.
The potential concern appears when the delegated activity is also something the person needs to learn. A student who repeatedly receives completed explanations without constructing their own understanding may miss opportunities to practice retrieval, reasoning, and integrating new knowledge.
The second pattern is cognitive collaboration. Instead of simply receiving an answer, the person works with AI to develop ideas, compare alternatives, identify gaps, and refine a solution. The technology handles some of the information processing, while the human remains actively involved in deciding what matters.
For example, someone researching an unfamiliar scientific topic might use AI to explain terminology, compare theories, locate weaknesses in an argument, and connect concepts from different disciplines. The interaction can become a process of exploration rather than merely answer retrieval.
The third pattern is cognitive challenge. Here, AI is deliberately used to scrutinize the person's reasoning. Someone develops an initial interpretation, asks the system to identify weaknesses, evaluates competing explanations, and revises their position.
This can create substantial demands on evaluation, cognitive flexibility, and metacognitive awareness—the ability to examine and regulate one's own thinking. It may also expose assumptions that would otherwise remain unchallenged.
None of these approaches is universally superior. Delegating a routine task may be sensible, while independent reasoning may be essential when learning a new skill. The important distinction is whether the pattern of interaction matches the cognitive outcome being sought.

One of the more interesting potential benefits of AI receives comparatively little attention in discussions of cognitive decline: it may expand the range of intellectual activities people are willing or able to undertake.
Many complex projects have traditionally required substantial preliminary expertise. Someone interested in computational biology, engineering, advanced statistics, or unfamiliar areas of philosophy might previously have encountered barriers that made meaningful exploration impractical.
AI can lower some of those barriers by explaining unfamiliar concepts, breaking problems into manageable components, translating technical language, and providing feedback during investigation.
This could allow people to engage with questions they would otherwise never attempt. The total amount of intellectual exploration in their lives might increase even as the effort required for individual subtasks decreases.
For example, a person might spend less time searching through technical documentation but considerably more time evaluating hypotheses, making conceptual connections, or testing alternative solutions.
There is an important limitation, however. Access to sophisticated analysis is not the same as possessing sophisticated understanding. AI can make complex material feel accessible while leaving gaps in the user's ability to explain, verify, or independently apply it.
The potential benefit lies not merely in reaching more advanced material, but in what the person actually learns through that access.
A particularly revealing experiment published at CHI 2026 investigated how the timing of AI assistance affected critical-thinking performance.
Researchers studied 393 participants completing a task that required evaluating information and developing a reasoned decision. They varied both when participants could access an LLM and how much time was available.
The results showed an important reversal. When time was limited, having AI available from the beginning improved performance. When participants had sufficient time, early or continuous AI access impaired performance relative to conditions in which participants began by reasoning independently.
The same technology therefore produced different effects depending on the conditions under which it was used.
This finding challenges the idea that AI assistance has a single, consistent cognitive consequence. Under tight time constraints, assistance may help people overcome processing limitations. When time permits deeper independent engagement, introducing AI too early may change how thoroughly the person explores the problem.
It also raises a broader point: the cognitive value of assistance depends partly on what the task is asking the person to accomplish. Completing a demanding decision quickly and developing independent reasoning ability are related objectives, but they are not identical.
Perhaps the most important unresolved question concerns what happens after the AI is no longer available.
A person can complete a task exceptionally well with assistance without necessarily becoming better at performing that task independently. Conversely, a difficult learning experience may produce a modest immediate result while developing understanding that becomes valuable later.
These are different outcomes, operating over different timescales.
Immediate performance tells us what the combined human–AI system can accomplish. Independent capability concerns what the person can still understand, remember, evaluate, or perform without that system.
This distinction is especially important in education and skill development. Retrieving information, constructing explanations, making mistakes, and correcting misunderstandings can all contribute to learning. If AI consistently removes these activities, it may also remove opportunities to develop the underlying abilities.
But it would be equally misleading to conclude that every delegated task represents lost cognitive development. Avoiding repetitive information gathering may leave more time for higher-level analysis, and receiving appropriate guidance can make difficult learning more accessible.
The relevant question is not how much effort was expended, but what that effort contributed to lasting understanding.

There is another subtle possibility worth considering.
AI-generated explanations often arrive in fluent, organized language. A complex subject can suddenly appear clear because the system has assembled the argument, defined the terminology, and connected the relevant concepts.
That clarity can be genuinely useful. But understanding an explanation while reading it is not always the same as being able to reconstruct the reasoning independently.
This creates a potential metacognitive problem. People may judge their understanding partly through how easily information is presented, rather than through whether they can explain or apply it themselves.
The same issue can arise when evaluating AI-generated conclusions. If the system produces both the answer and the justification for that answer, the person may have fewer independent reference points for identifying weaknesses.
This does not mean AI explanations are inherently misleading. It means that fluent assistance can change the signals people use to decide whether they understand something well enough.
As AI becomes more capable, that distinction between receiving a convincing explanation and possessing the ability to evaluate it may become increasingly important.
Research is beginning to identify situations in which AI assistance improves performance, changes cognitive strategies, or reduces independent engagement. What remains much less clear is how different patterns of repeated use affect cognitive abilities over months or years.
Many existing studies examine relatively short interventions, particular educational tasks, or immediate performance. Others rely on self-reported effort or indirect measures of cognitive activity.
These approaches can reveal meaningful effects, but they do not establish whether extensive LLM use causes broad, lasting changes in intelligence, memory, reasoning, or cognitive flexibility.
The longer-term outcome may also differ substantially between individuals. A person who uses AI primarily to avoid unfamiliar challenges may have a very different experience from someone who uses it to pursue increasingly ambitious intellectual work.
Even then, more demanding interaction should not automatically be equated with greater cognitive benefit. Effort can be productive, unnecessary, or misdirected depending on the circumstances.
Understanding these differences will require research that examines not just AI exposure, but the cognitive activities occurring during use and the capabilities retained afterward.
The question of whether AI makes us think less is understandable, but incomplete.
AI can reduce the effort required for certain tasks while increasing the complexity of others. It can substitute for reasoning, support collaboration, challenge assumptions, or open access to intellectual activities that previously seemed out of reach.
These effects can coexist within the same person, project, or working day.
What matters is the changing relationship between cognitive effort, human judgment, task performance, and learning. A person may accomplish more while practicing some skills less, or develop new forms of expertise while relying increasingly on assistance for familiar tasks.
The science has not yet established what these changes will mean for our long-term cognitive abilities. But it already provides a compelling reason to move beyond the simple assumption that less effort means cognitive decline, or that better AI-assisted results necessarily mean cognitive improvement.
The more revealing question may be not whether AI is doing more of our thinking, but how its involvement is changing the thinking that remains ours.

Welcome to the Research and Strategy Services at in today's fast-paced.
An evidence-based discussion of whether activities like crosswords and Sudoku meaningfully improve brain health, clarifying what they support, what they do not, and why benefits are often misunderstood.
Check out these excellent insights on the role of neuroscience in sports performance.
Learn about your brain's remarkable neuroplasticity.