AI Makes Everyone Sound Smart

The relationship between the quality of a person's written expression and the quality of their thinking has always been imperfect. Good thinkers who write poorly exist, and their thinking has historically been systematically undervalued relative to polished writers whose thinking was less sound. The correlation was real but not perfect, and it produced systematic distortions in how intellectual ability was perceived and rewarded. In this sense, the AI-driven disruption of the writing-quality signal is not entirely lamentable: there is a genuine equity argument for a world in which the quality of one's thinking is evaluated more directly and less through the proxy of writing quality, which itself correlates with education and privilege in ways that reflect the history of who had access to good education rather than the distribution of genuine intellectual ability.

But the signal, imperfect as it was, was doing something. The effort required to produce polished written communication was providing evidence about the level of investment behind the thought. A well-crafted argument was evidence, not proof, of careful thinking: you had to have thought carefully enough about what you were saying to say it well. The cost of producing polished text created a rough filter on what got polished: people invested in writing well when they were writing things they had genuinely thought through, and the quality of the writing reflected the quality of the investment. When the cost of polishing drops to near zero, the filter disappears, and the investment signal disappears with it.

There is also a reception effect that matters: audiences read well-written prose differently from poorly-written prose, and the difference in reading mode affects how critically they evaluate the argument. Well-written text is processed more smoothly, triggers fewer resistance signals, and is more likely to be accepted rather than interrogated. This is not a new phenomenon, but AI makes it more consequential: if anyone can produce well-written text on any position, the rhetorical advantage that good writing provides is detached from the merits of the argument and becomes available to any argument that is expressed through the tool. The most persuasive arguments in a world where everyone has access to AI writing assistance may be the most persuasive in a sense that has little to do with the quality of the underlying reasoning.

The disruption of the writing quality signal as a proxy for intelligence does not mean that intelligence becomes undetectable or that all signals of genuine intellectual quality disappear. It means that the signals that remain legible are the ones that cannot be easily generated by AI, and identifying and amplifying those signals is an important task for the people and institutions that need to evaluate intellectual quality.

The signals that are hardest for AI to fake are those that require genuine engagement with a specific situation in real time: the quality of questions asked in a conversation, the ability to navigate unexpected complexity, the responses to genuine challenge that reveal whether the apparent understanding is real or surface-level. Written work that requires demonstrating understanding rather than just presenting it, explanations that require working through a problem rather than summarizing its conclusion, responses to probing follow-up questions that are designed to find the edges of genuine understanding: these are evaluation approaches that are harder to AI-assist in real-time contexts and that therefore remain more informative.

The irony of the intelligence inflation problem is that the response to it tends to produce evaluation contexts that are more demanding rather than less, and more direct rather than more mediated. If polished text is an unreliable intelligence signal because anyone can produce it with AI, the alternative is evaluation approaches that produce signals AI cannot easily fake: genuine intellectual engagement in contexts that require it. This is, in some ways, a correction toward the thing that actual intelligence assessment was always supposed to be doing rather than a loss of something valuable.

The deepest concern about intelligence inflation is not just evaluative, not just about how institutions assess intellectual quality. It is cultural: about what happens to the norms of intellectual life when the appearance of intelligence becomes so easily producible that the aspiration to genuine intellectual development is systematically devalued. If anyone can sound smart without being smart, and if sounding smart is often indistinguishable from being smart, and if being smart provides social and professional rewards, then the incentive to invest in the genuine development of intelligence and understanding is reduced relative to the incentive to invest in the performance of it. This is a cultural shift with long-term consequences that are difficult to predict but easy to worry about.

The cultural context in which intellectual development has historically been valued is not a naturally occurring equilibrium. It has required institutional support: educational cultures that reward genuine understanding over performance, professional communities that distinguish between serious and superficial engagement with a field, social contexts in which the difference between genuine and performed intelligence matters enough to be worth the investment. These institutional and cultural supports do not disappear overnight. But they are under pressure from an environment in which the performance has become dramatically easier and the tools for distinguishing performance from substance have become less reliable.

What I find worth holding onto, in the middle of this concern, is that the value of genuine intellectual development is not entirely or even primarily instrumental. The person who has genuinely developed their understanding of something, who has engaged with its complexity, who has built the judgment that comes from real intellectual work, has something that the person who has only performed intelligence does not have: the actual experience of genuine understanding, which is among the more deeply satisfying things a human mind can do. The intelligence inflation problem is a real problem with real consequences for how intellectual quality is evaluated and rewarded. It is not a reason to stop caring about the genuine thing.