AI Could Make Intelligent People Less Smart

A mathematician opens up ChatGPT to check a calculation she could have easily solved in her head. An executive uses AI to draft an email he's written a hundred times before. A professor asks NoteBookLM to summarize a student paper she has the expertise to read herself.

Is there a difference between remembering information vs. remembering where to find it? Is there a difference in knowing how to do a task vs. knowing how to get it done for you?

Cognitive offloading, the act of delegating mental work to external tools, should concern anyone who values their own intelligence. When given the choice between thinking through a problem or letting technology handle it, will we choose the tool, even when we're perfectly capable of succeeding on our own?

The Metacognitive Trap

You can't build a skill you don't practice.

Sam Gilbert noticed in his experiments something he calls "reminder bias." This is the tendency to use external memory aids even when your own memory would serve you better. Participants chose to set digital reminders for tasks they could easily remember unaided. Even when offered money to encourage people to rely on their own memory, they still couldn't overcome it.

The root appears to be metacognitive. People underestimate their own cognitive abilities. They think they'll forget, so they set a reminder. They think they can't solve the problem, so they ask AI. Over time, this creates a self-fulfilling prophecy. By consistently choosing the external tool, they never exercise the internal capability.

Gilbert found that these biases become dispositional traits. Someone who habitually offloads cognitive work at age 25 will likely still be doing it at 45. The neural pathways that weren't exercised in early adulthood may never fully develop.

The Brain in Your Pocket

If the problem were limited to memory, it might not matter much. Afterall, what’s the problem with setting a reminder on your phone? But research suggests the cognitive costs extend far beyond remembering appointments.

Nathaniel Barr claimed that individuals who score lower on the Cognitive Reflection Test (a measure of analytical versus intuitive thinking) show greater smartphone dependence for information retrieval. These are people who, when faced with a problem, instinctively reach for the quickest solution.

AI hasn't created this tendency. Humans have often preferred mental shortcuts. But AI has removed nearly all friction. You don't have to go to a library or even type a search query. You just ask, and the answer appears, exactly as you need it.

My concern is that people may begin using AI in ways that prevent the development of analytical skills. For a college student who consistently uses AI to work through problems, I question whether they're building the neural infrastructure to handle similar problems independently later in life.

What Happens to Knowledge Without Memory

The traditional argument for cognitive offloading has always been straightforward: Why memorize facts you can look up instantly? Save your mental energy for higher-order thinking, right?

But I’m starting to think this framing misunderstands how the brain works. Knowledge isn't separate from thinking. It's the substrate, the foundation, that thinking requires.

Across every subject domain, expertise appears to depend on having vast amounts of information readily accessible in working memory; not theoretically accessible through a Google search, but immediately present in your mind. A doctor diagnosing a patient, a lawyer constructing an argument, a teacher responding to a question all rely on the instant availability of relevant content knowledge.

Artificial Intelligence Essential Reads

The Collateral Challenge of AI Sycophancy

Your Brain on AI: Cognitive Offloading, Debt, and Atrophy

If you only remember where information exists, but you don't actually know it, you have the illusion of knowledge without the substance. My concern is that AI may be making us feel knowledgeable because we can quickly retrieve information. But this doesn’t allow us to develop the internalized understanding that enables genuine expertise.

The Smart Person's Dilemma

If the research is correct, intelligent people face a paradox. They're the most capable of recognizing AI's power and integrating it into their workflows. But they may also be the most vulnerable to its cognitive costs. AI won’t make them less capable in absolute terms, but it may prevent them from developing the capacities they otherwise would have built.

A gifted student who uses AI to write every essay may still produce excellent work. But she never develops her personal voice that comes from struggling through dozens of drafts. A talented programmer who uses AI to generate code may ship products faster. But he never builds the deep structural understanding that comes from writing each function himself.

The output looks the same. The process is invisible. The cognitive development that didn't happen is impossible to measure until much later, when that person faces a problem AI can't solve.

The Question We're Avoiding

The difficulty with all of this research is that it arrives too late to be preventive. The generation that grew up with smartphones is already in the workforce. The students currently using ChatGPT for every assignment are developing cognitive habits that may prove difficult to reverse.

We don't yet know whether these effects are permanent or whether they can be mitigated through deliberate practice later in life. We don't know how much AI assistance is beneficial versus harmful. We don't even have good methods for measuring the cognitive skills that may be declining, because standardized tests can't distinguish between someone who thinks well and someone who searches well.

What we do have is a pattern in the research that's become increasingly difficult to ignore. Cognitive offloading appears to be habit-forming. It seems to be most appealing to people who already prefer mental shortcuts. And it may be preventing the development of the very capabilities it's meant to augment.

For smart people, it means that by outsourcing the cognitive work that made them smart in the first place, they'll quietly become less capable of the original thinking that AI still can't replicate. The decline won't announce itself. It will simply be the difference between the person you might have become and the person you are; a gap you'll never be able to measure because the alternative version never existed.

Are you willing to find out?