MIT professor Eric So has a warning for anyone who reaches for a chatbot at the first sign of effort: depending on AI can make your brain less nimble. His argument, as reported by Fox News, is that AI exploits two things the brain already does well, conserving mental effort and chasing instant gratification. The result is a feedback loop where people turn to chatbots for even the simplest tasks.
The useful part is what So says to do about it. By the end of this piece, you'll have a clear rule for when to use AI and a few concrete workflows that keep you doing the thinking.
What So Is Actually Arguing
Per the Fox News report, So says overreliance on AI can weaken memory, problem-solving, and foundational mastery. The mechanism is the removal of mental "friction," the effortful part of learning that makes knowledge stick.
He doesn't say AI is bad. He draws a line between two uses:
- AI as an answer machine: harmful, because it skips the struggle.
- AI as a tutor or cognitive trainer: valuable, because it can support the struggle instead of replacing it.
That distinction is the whole game, and the rest of this article is about putting it into practice.
Why the Loop Is So Easy to Fall Into
The pull is built into how we work. A chatbot gives you a polished answer in seconds, and your brain, which likes saving effort, rewards you for asking again. Each easy win makes the next unassisted attempt feel slower and more annoying.
My own read: the danger isn't the big, obvious uses like drafting a report. It's the small ones. When you ask a chatbot for something you could have worked out in a minute, you skip exactly the kind of tiny repetition that builds recall and reasoning. So's point about "even the simplest of tasks" is the one worth sitting with.
A Simple Rule: Attempt First, Ask Second
The easiest way to keep friction in your day is to order your steps. Before you open a chatbot, spend a few minutes on your own attempt, even a rough or wrong one.
- Write down what you think the answer is, or sketch your approach.
- Note where you're stuck, in a sentence.
- Only then bring AI in, and tell it what you've tried.
This turns the chatbot from a vending machine into a reviewer. You've already done the retrieval work, and the AI's response has something to land on.
Workflows That Treat AI Like a Tutor
These are my suggestions for applying So's tutor-versus-answer-machine distinction. They aren't from his remarks, but they follow the same logic.
For studying
Ask the AI to quiz you instead of summarizing the chapter. Answer from memory before it reveals anything. If you miss a question, ask for a hint rather than the solution, then try again.
For writing
Draft the first version yourself, even if it's ugly. Then ask the AI to point out unclear arguments or gaps in logic, and make the revisions yourself. You keep authorship of the thinking, and the AI acts as an editor.
For problem-solving and coding
Ask for the next step, not the finished solution. A prompt like "Give me a hint about what to check next, without writing the answer" keeps you in the driver's seat.
For brainstorming
List your own ideas first. Then ask AI for angles you missed. If you start with its list, you tend to anchor on it and stop generating your own.
Prompts That Preserve the Struggle
You can bake the tutor role into how you ask. A few phrasings to try:
- "Quiz me on this topic one question at a time and wait for my answer."
- "Don't give me the solution. Ask me questions that lead me toward it."
- "Here's my attempt. What's the weakest part, and why?"
- "Explain this concept, then ask me to explain it back in my own words."
Each one forces you to produce something, which is where learning happens.
Know When the Answer Machine Is Fine
Not every task is a learning task. Reformatting a spreadsheet or tidying a meeting summary doesn't build a skill you care about, and using AI there is reasonable. A practical test: ask yourself whether you'd want to be able to do this unaided in a year. If yes, treat AI as a tutor. If no, let it handle the chore.
The pressure to use AI constantly, especially for students, workers, and creators, makes this check more valuable. Using it often isn't the problem. Using it without deciding what you want to keep doing yourself is.
The Hidden Cost of Instant Answers
So's framing suggests that the cost of a quick answer is paid later. Memory, problem-solving, and foundational mastery are what you lose when you never have to wrestle with the material. You won't notice it on any single day. You'll notice it when the tool isn't available, or when a problem is new enough that no prompt quite fits.
Productive struggle feels worse in the moment than getting an immediate answer. That discomfort is not a sign you're doing it wrong. It's the signal that the learning is happening.
Your Takeaway
Tomorrow, pick one task you'd normally hand straight to a chatbot and make your own attempt first. Then bring the AI in to critique, quiz, or hint, never to simply finish the job. Do that consistently and you get the speed of AI without giving up the thinking that makes you good at your work.
