The Cognitive Economy of Effort
AI saves mental effort, but some of that effort is how independent judgment grows.
Sahir Maharaj11 min read
Every generation inherits a set of fears about the technology of its moment, and those fears are often more revealing about human anxieties than about the technology itself. We feared that calculators would make people unable to do arithmetic. We feared that search engines would make people unable to remember anything. We feared that GPS would make people unable to navigate. In each case, the fear was technically accurate in a narrow sense and missing something important in a wider one: the cognitive capacity that the tool replaced was not the cognitive capacity that was actually worth worrying about. The person who cannot do long division but can model complex systems has not been diminished by the calculator. The person who cannot remember a phone number but can synthesise information across domains has not been diminished by contacts lists. The question about AI and thinking is whether it follows this same reassuring pattern, or whether something genuinely different is happening.
I want to argue that something genuinely different is happening, and that the difference is worth being precise about. The tools that replaced specific cognitive functions in previous eras replaced functions that were largely instrumental: the calculation, the memorisation, the route-finding. The thinking that those functions supported, the reasoning, the learning, the judgment that came from genuinely engaging with ideas, remained the work of the human mind. What AI threatens to replace is not the instrumental function but the thinking itself: the process of working through a problem, forming a view, constructing an argument, weighing evidence, and arriving at a conclusion through genuine cognitive engagement. When AI provides the conclusion directly, and when the conclusion is good enough that examining the reasoning behind it seems unnecessary, the cognitive work that the conclusion was supposed to reflect does not happen. And that work, unlike arithmetic, is not replaceable by any tool.
The specific mechanism through which this works is worth understanding clearly, because it is not primarily about laziness or intellectual sloth, though those tendencies exist and AI accommodates them. It is about the cognitive economy of effort: the natural human tendency to allocate cognitive resources efficiently by not thinking harder than the situation requires. When AI provides a good-enough answer, the situation does not require the harder thinking that a better answer, or a genuine engagement with the question, would involve. Over time, the habit of doing the easier thing, of taking the AI answer, atrophies the capacity for the harder thing, not through any dramatic failure but through the simple mechanism of disuse.
Thinking is not only the production of conclusions. It is a process that does several things simultaneously, and the conclusion is often the least important of them. When you genuinely work through a problem, you are building a mental model of the problem space that allows you to navigate related problems more effectively. You are identifying the assumptions that the problem rests on, which allows you to notice when those assumptions fail. You are developing the capacity to recognise when a conclusion does not follow from its premises, which is the foundational cognitive skill for detecting error and manipulation. And you are building the specific kind of understanding that allows you to know when you do not understand something, which is the prerequisite for seeking better understanding rather than accepting what is provided.
The person who consistently takes AI conclusions without engaging with the reasoning behind them is not simply saving time. They are systematically not building the mental infrastructure that genuine thinking produces, and they are not developing the evaluative capacities that would allow them to assess whether the AI conclusions are correct. This is the specific and serious concern that distinguishes the AI case from the calculator case: when you use a calculator, you still need to know whether the answer is plausible, which requires the mathematical intuition that genuine arithmetic engagement develops. When you use an AI to form a view on a complex question, you need to know whether the view is sound, which requires exactly the kind of critical engagement with the question that using the AI has displaced.
The social dimension of this matters as much as the individual one. Democratic societies depend on citizens who can think independently about complex questions, who can evaluate the arguments they are presented with, and who are resistant to manipulation that relies on the inability to assess the reasoning behind confident assertions. The gradual erosion of independent thinking capacity across a population produces a society that is more susceptible to manipulation, less capable of collective deliberation, and more dependent on whoever controls the AI systems that are doing the thinking for it. This is not a hypothetical concern. It is the direction in which the dynamics of AI adoption are pointing, and the speed at which it is developing makes the timeline shorter than comfortable.
The distinction I want to draw is not between using AI and not using AI, which would be an unhelpful and unrealistic position. It is between using AI as a tool that serves thinking and using AI as a substitute for it. The person who uses AI to gather relevant information quickly, and then does the work of evaluating, synthesising, and forming a judgment about that information, is using AI in a way that serves their thinking. The person who asks AI what to think and accepts the answer without genuine engagement is delegating the thinking itself. The difference is not always visible from the outside, and it is not always easy to maintain from the inside, because the AI answer is usually good enough that the engagement feels optional.
The practical disciplines that maintain the distinction are not mysterious, but they require deliberateness in an environment that consistently rewards taking the shortcut. They include: forming a preliminary view before consulting AI, which gives you something to measure the AI output against rather than simply receiving it. Engaging critically with the reasoning behind AI conclusions rather than only evaluating the conclusions themselves. Actively identifying the assumptions in an AI argument and testing whether they hold. And maintaining the habit of occasionally working through questions entirely without AI, not because the result will be better, but because the exercise maintains the capacity that the shortcuts would otherwise atrophy.
The deepest version of the concern about AI and human thinking is not about any specific individual's cognitive habits. It is about what kind of culture we are building. A culture in which the capacity for independent thought is cultivated, practiced, and valued is a different kind of culture from one in which it has been progressively delegated to AI systems and the skill has quietly rusted. The AI era will produce more of both cultures, and which one dominates in any given institution, community, or society will depend on choices that are being made now, mostly without the explicit recognition that a choice of this significance is being made.
The aspiration worth building toward is not a world without AI assistance but a world in which AI assistance is consistently used in ways that strengthen rather than substitute for human cognitive capacity. This requires changes at multiple levels. At the individual level, it requires the deliberate practices I described: the preliminary view, the critical engagement, the maintained habits of independent thought. At the educational level, it requires rethinking what we are trying to develop in students and ensuring that the development of genuine cognitive capacity is not sacrificed in the name of AI fluency. At the institutional level, it requires creating environments in which independent thinking is valued and practiced rather than simply rewarded for its outputs.
The educational challenge is the most urgent because it is the most time-sensitive. The students who are going through their formative cognitive development right now are doing so in an environment saturated with AI assistance, and the habits and capacities they develop or fail to develop during this period will shape the cognitive culture of the next generation. The educators who are navigating this challenge need more institutional support, more pedagogical clarity, and more cultural validation for the choice to maintain genuine cognitive challenge in the face of AI tools that make that challenge avoidable.
The real danger is not that AI thinks. AI thinking, applied thoughtfully, is one of the most powerful tools available for addressing the genuine challenges that human society faces. The real danger is that the convenience and quality of AI thinking gradually makes human thinking feel unnecessary, and that the atrophy that follows is not noticed until the capacity it produces is needed and found to be absent. Maintaining human thinking in a world of AI assistance is not a nostalgic commitment to doing things the hard way. It is the recognition that the capacity to think independently is the foundation on which everything else we value about human agency and human society depends.
Topics in this article
You might also like
View all
The Most Dangerous AI Skill Is Blind Confidence
AI sounds certain even when it is wrong. Knowing when to doubt is the real skill.
AI Will Reward People Who Think Clearly
The bottleneck moved from production to cognition. That changes who wins.
The Hardest AI Safety Feature Is Saying No
Useful AI completes tasks. Safe AI also needs to know when to refuse.