Last year I wrote that artificial intelligence, together with social media, was contributing to what many people now call “brain rot.” I still believe there is truth in that observation. Spend enough time watching people consume algorithmically generated content or accept AI-generated answers without question and it becomes obvious that something is changing. But after reading a growing body of research on how people actually interact with artificial intelligence, I have come to believe I blamed the wrong culprit. AI is not making us intellectually lazy. It is simply making it easier to indulge habits that were already there.
That distinction may sound subtle, but it completely changes the conversation. If AI itself is the problem, then our focus naturally shifts toward regulating the technology, redesigning the models or somehow limiting their influence. If the problem lies instead with how we choose to use these systems, then responsibility remains where it has always belonged—with us. Technology has never determined human outcomes on its own. It has always amplified human behavior, for better or worse.
The Research on Cognitive Surrender
The latest evidence is certainly thought-provoking. Researchers from Wharton, Carnegie Mellon, Anthropic and elsewhere have documented what they describe as “cognitive surrender,” a tendency for people to stop evaluating answers once an AI presents them with confidence and fluency. In study after study, people working with AI often produce better results while the system is available, but when that assistance disappears their performance declines, sometimes below that of people who never relied on AI at all. In some cases, participants even became more confident in incorrect answers simply because those answers came from a persuasive machine. Those findings deserve attention, but they do not convince me that artificial intelligence is making us less intelligent. They convince me that many of us are becoming less willing to do the hard work of thinking for ourselves.
History suggests that this is not a new pattern. Human beings have always embraced technologies that allowed us to offload difficult tasks. Writing reduced the need to memorize vast quantities of information. The printing press removed the burden of copying knowledge by hand. Calculators freed us from pages of arithmetic, while search engines made remembering isolated facts far less important than knowing how to find them. We did not become less capable because these tools existed. We simply shifted our cognitive effort toward problems that those tools could not solve.
Why AI Is Different From Past Tools
Artificial intelligence is different only because it reaches further than any of those earlier innovations. It does not simply retrieve information or perform calculations. It participates in activities that look remarkably like reasoning. It can summarize, compare, analyze, draft arguments and propose solutions. Whether or not we believe these systems truly “reason” in the human sense is almost beside the point. They produce outputs persuasive enough that many people stop asking whether those outputs deserve to be trusted. The danger, then, is not that AI thinks instead of us. The danger is that we become comfortable allowing it to decide when our own thinking is finished.
That, to me, is the real lesson emerging from the latest research. We keep asking whether AI is making people stupid when the more important question is whether we are voluntarily surrendering one of the very abilities we claim makes us uniquely human. Intelligence has never been defined by how quickly someone arrives at an answer. It has always been reflected in the willingness to question assumptions, wrestle with uncertainty and remain skeptical of conclusions that arrive too easily. AI does not eliminate that process. It simply gives us the option of skipping it.
Replacement vs. Collaboration
Ironically, that is also where I believe AI is most valuable. Used passively, it can become an extraordinarily efficient shortcut around the productive struggle that leads to learning. Used actively, however, it can become the most demanding intellectual partner most of us have ever had. I do not find AI particularly useful when I ask it to tell me what to think. I find it invaluable when I ask it why my argument is wrong, what evidence I have overlooked or how someone who fundamentally disagrees with me would dismantle my conclusions. Those conversations do not replace thinking; they force me to think more carefully than I otherwise would.
That is the distinction we need to spend more time on. There is a world of difference between using AI as a replacement and using it as a collaborator. When someone asks a chatbot to write the memo, solve the homework problem, generate the code or produce the strategy without first wrestling with the work themselves, they are not augmenting intelligence. They are avoiding effort. The output may be good enough. It may even be excellent. But the person behind it has not become more capable in the process.
There is another way to use the same tool. Ask AI to critique the memo after you have written it. Ask it to identify weaknesses in your argument. Ask it to compare your assumptions against contrary evidence. Ask it to explain a concept without giving you the answer. Ask it to behave less like a ghostwriter and more like a difficult editor. In that mode, AI is not a substitute for thinking. It is resistance training for the mind.
The Personal Trainer Problem
The personal trainer analogy is useful here. If you hire a trainer and then ask the trainer to lift the weights for you, the weight still gets moved. From the outside, the work appears to have been done. But you do not get stronger. You merely outsource the appearance of strength. That is what much of today’s AI usage looks like. The report is finished. The answer is polished. The code compiles. The slide deck looks professional. But the human being may have gained nothing except a faster path to completion.
For some tasks, that is perfectly fine. I do not need to memorize phone numbers. I do not need to manually calculate every spreadsheet formula. I do not need to retype boilerplate language I have written a hundred times before. There is no moral virtue in doing every tedious task by hand. The problem begins when we confuse convenience with competence. Just because AI can help us produce something does not mean it helped us understand it.
Why This Matters Most for Beginners
That distinction matters most for beginners. Experts have mental models that allow them to judge what AI produces. A seasoned engineer can look at generated code and see whether the approach makes sense. An experienced lawyer can recognize when a legal argument is plausible or nonsense. A good editor can tell when a paragraph sounds fluent but says nothing. Novices do not have that same protection. They are far more likely to mistake confidence for correctness because they do not yet know what correctness looks like.
This is one of the underappreciated risks of AI adoption. The people who most need to build foundational understanding are often the ones most tempted to bypass it. Students can get the answer without learning the method. Junior employees can produce the document without understanding the business context. New developers can generate code without learning why it works. In each case, the short-term output improves while long-term capability may erode.
That should worry us, but not because AI is evil. It should worry us because human beings are very good at rationalizing shortcuts. We tell ourselves we are being efficient. We tell ourselves we are focusing on higher-value work. Sometimes that is true. Sometimes we are simply avoiding the uncomfortable part of learning. The struggle is not a bug in the process. It is the process. The frustration, false starts and slow accumulation of understanding are how competence is built.
What Business Leaders Should Watch
This is where I think much of the AI productivity discussion is too shallow. We measure whether people finish more tasks, complete them faster or produce better immediate outputs. Those are useful metrics, especially in business. But they do not answer the deeper question: Did the person become more capable, or did the person become more dependent? In the short run, those two outcomes can look almost identical. In the long run, they are worlds apart.
For business leaders, this is not an academic concern. Enterprises are racing to put AI into every workflow. That is understandable. The productivity upside is too large to ignore. But if organizations deploy AI in a way that rewards output while ignoring learning, they may quietly hollow out the very judgment they need most. They may create teams that can produce more artifacts but understand less about the decisions behind them. They may accelerate work while weakening the people responsible for knowing whether the work is any good.
Judgment Is the Scarce Resource
The future of work will not belong to people who merely accept AI outputs faster than everyone else. It will belong to people who know when not to accept them. That is a very different skill. In an AI-rich world, the scarce resource is not the ability to generate text, images, code or analysis. Those things are becoming abundant. The scarce resource is judgment. It is the ability to know what question to ask, what answer to distrust, what fact to verify, what assumption to challenge and what conclusion deserves to survive.
That is why I do not buy the argument that AI reduces the value of human thinking. I think it increases the value of good human thinking. If everyone has access to capable models, the differentiator is no longer who can produce the first draft. The differentiator is who can recognize the better idea, the stronger argument, the hidden flaw and the more important question. AI may commoditize certain forms of knowledge work, but it will put a premium on taste, judgment, curiosity and intellectual courage.
A Personal Discipline for AI
This also means we need a better personal discipline around AI. Before asking a model to write something, own the thesis yourself. Before asking it for an answer, try to form one. Before accepting its conclusion, ask it to argue the other side. Use AI to expand your field of vision, not narrow it. Use it to challenge your thinking, not confirm your bias. Use it to make you uncomfortable before you let it make you efficient.
That last point is important. One of the dangers of AI is that it is often too agreeable. It can flatter weak ideas, smooth over contradictions and present uncertainty as polish. If you ask it to validate you, it usually will. That is not collaboration. That is intellectual comfort food. The better use of AI is to force friction back into the process. Tell it to disagree. Tell it to find the holes. Tell it to make the strongest case against you. Tell it to behave like the smartest skeptic in the room, not the most obedient assistant.
In that sense, the problem is not that AI thinks too much. The problem is that we often ask too little of it and even less of ourselves. We treat it like a machine for producing answers when its higher value may be helping us generate better questions. The answer is where thinking often ends. The question is where thinking begins.
AI Did Not Steal Your Thinking
This brings me back to the “brain rot” argument. I still believe careless use of AI and social media can weaken the habits that make us effective thinkers. I still believe a steady diet of algorithmic feeds, AI summaries and frictionless answers can make us less curious, less patient and less willing to follow a difficult idea to its conclusion. But I no longer think it is accurate to say AI is turning our brains to mush. That gives the machine too much credit and lets us off far too easily.
AI did not steal our attention. We gave it away. AI did not eliminate curiosity. We stopped exercising it. AI did not replace judgment. We chose to accept fluency as a substitute for it.
Every generation has faced some version of this bargain. A new technology arrives and offers to do something faster, cheaper or more conveniently than we can do it ourselves. Sometimes accepting that bargain is progress. Sometimes it comes with hidden costs. Artificial intelligence raises the stakes because the work being offered up is no longer just physical labor, arithmetic or recall. It is reasoning, synthesis and judgment.
That does not mean we should reject AI. Far from it. Used well, it can make us better. It can expose our blind spots, sharpen our arguments, broaden our research and accelerate our understanding. It can help a good thinker become a better one. But used poorly, it can help a lazy thinker appear productive while becoming weaker underneath.
The choice is not really between using AI and not using AI. That debate is already over. AI is becoming part of the fabric of work, education and daily life. The real choice is whether we use it to replace our thinking or refine it. That choice will be made thousands of times a day, in prompts that seem too small to matter but collectively define how we relate to these systems.
We spend a lot of time asking what AI will become. We do not spend nearly enough time asking what we will become while using it. The future will not be determined solely by how intelligent the machines get. It will be determined by whether we continue demanding intelligence from ourselves.
AI did not steal your thinking. It gave you the option to use it less. What happens next is on you.
