An MIT study from 2025 confirmed what teachers have long feared: relying on AI to do your thinking creates "cognitive debt," missing an opportunity to build your brain's neural connections when you work alongside it. But the solution isn't to ban AI—it's to change how you struggle with it.
First up, the TL;DR:
If you only have two minutes, read this.
Here's what happened: Researchers at MIT monitored the brain activity (EEG) of essay writers and found that those who used ChatGPT had significantly weaker neural connectivity in the Alpha, Beta, and Theta bands compared to those who wrote manually.
Essentially, the AI users outsourced the cognitive heavy lifting. The result? Memory atrophy. A shocking 83% of the AI group couldn't quote a single sentence from the essay they had just "written." Hard to remember something you never wrote, and perhaps never even read in the first place.
Even worse, when these students were forced to write without AI later, they performed worse than people who had never used AI at all. This is "cognitive debt"—borrowing intelligence today at the cost of capability tomorrow.
THE EVIDENCE: A massive study from Wharton on math students backs this up with hard numbers:
- The Cheat Code: Students using standard ChatGPT (giving direct answers) performed 17% worse on exams than those who used no tech.
- The Sparring Partner: Students using an "AI Tutor" (programmed to give hints, not answers) saw a 127% improvement in practice scores without hurting exam performance.
WHY IT MATTERS / WHAT TO DO: The future belongs to the "Judgment Economy," where knowledge is cheap but "Taste" and "Agency" are priceless. You cannot build judgment if you outsource the struggle.
So, adopt this "Struggle-First" Principle:
- Wrestle for 20 mins: Before prompting, try to solve the problem (code, draft, math) yourself.
- Think it through, and document your thinking as you go (this will become your prompt).
- Spar, Don't Copy: Use AI as a "Socratic Tutor" to challenge your logic, and ask as many questions as possible as you wrestle with the material.
- The Quote Test: If you can't quote the insight back from memory after closing the chat, you haven't learned it. So try again.
- Follow Learning Science Principles: Use Spaced Repetition, Topic Interleaving, and "Desirable Difficulties" (see Uncommon Sense Teaching) to actually improve your learning.
The "Cognitive Debt" Crisis: Why AI is weakening your brain (and how to fix it)
If you feel like your brain is getting "smoother" the more you use ChatGPT, you aren't imagining things.
We finally have neurological evidence that outsourcing your thinking to AI doesn't just make you lazy—it physically alters how your brain forms connections (or doesn't).
A recent study from MIT titled "Your Brain on ChatGPT" used electroencephalography (EEG) to scan participants while they wrote essays. The results were stark. The group allowed to use LLMs showed significantly weaker neural connectivity across the board (Alpha, Beta, Delta, and Theta bands) compared to those who wrote with just their brains or a search engine.
The most damning stat? 83% of the AI users couldn't remember a single sentence of the essay they had just completed.
The researchers call this "Cognitive Debt" (as described by Ethan Mollick in One Useful Thing). Much like technical debt in code, cognitive debt accumulates when you take shortcuts. When the AI users were later forced to work without the tool, they didn't just revert to average; they performed worse than the control group. They had lost the ability to struggle through a problem.
The "Desirable Difficulty" Paradox
This isn't just about essay writing. A massive study from Wharton involving nearly 1,000 students found the exact same trend in math:
- GPT Base (The Answer Machine): Students given a standard chatbot that provided answers performed 17% worse on exams than kids who never touched AI.
- GPT Tutor (The Coach): Students given a bot designed to offer hints (but refuse answers) saw a 127% boost in practice performance with zero negative impact on learning.
The takeaway is clear: AI doesn't rot your brain. Using it as a crutch does.
🧠 The Science of Learning (and How to Hack it With AI)
Most of us were taught to study by re-reading notes, highlighting textbooks, and cramming the night before. The science says all of that is garbage.
According to the book Make It Stick (and research from the Bjork Learning Lab at UCLA), these passive methods create an "illusion of competence." You recognize the material, so you think you know it. But you don't.
Real learning happens through "Desirable Difficulties"—struggles (as discussed by Simon Sinek on The Diary of a CEO and polyglot Steve Kaufmann) that trigger the brain to physically encode new neural pathways.
Here are the 4 "Horsemen" of sticky learning, and exactly how to use AI to trigger them instead of bypassing them.
1. The Generation Effect
The Science: Your brain learns better when it attempts to solve a problem before it is shown the solution—even if you guess wrong. The struggle to "generate" the answer creates a cognitive hook that the correct answer can latch onto later.
- The AI Hack (The "20-Minute Rule"): Never open a prompt with a blank brain. Before asking AI for code, an essay structure, or a strategy, force yourself to draft a terrible version for 20 minutes. Then, prompt the AI: "I am trying to solve [Problem X]. Here is my first messy attempt. Do NOT fix it for me yet. First, point out the gap in my logic and give me a hint to solve it myself."
2. Retrieval Practice
The Science: Re-reading puts info in. Retrieval forces you to pull info out. Every time you retrieve a memory, you modify it and make the neural pathway stronger. The MIT study showed AI users failed because they never had to retrieve anything—it was all external.
- The AI Hack (The "Close the Tab" Protocol): After a session with AI, minimize the window. Open a blank document. Type out the key takeaways, code logic, or arguments from memory. If you can't, you haven't learned it.
3. Spaced Repetition
The Science: Cramming works for 24 hours. Spacing works for life. You need to let yourself slightly "forget" information before retrieving it again. This effort to recall faded knowledge signals to your brain that this info is vital for survival.
- The AI Hack (The Schedule Manager): AI is terrible at remembering context over long chats, but excellent at formatting calendars. Ask it: "We just covered [Topic]. Create a 'Spaced Repetition Schedule' for me. Give me 3 review questions to ask myself in 24 hours, 3 days, and 1 week. Format this as a CSV file I can import into my Google Calendar."
4. Interleaving
The Science: Traditional school teaches "AAA BBB CCC" (Block practice). Real life is "ABC BCA CAB." Interleaving mixes up different types of problems/subjects. It forces your brain to not just execute a solution, but to first identify which solution is required.
- The AI Hack (The Playlist Shuffle): If you are learning a skill (like coding), prompt the AI: "I am studying [Skill A], [Skill B], and [Skill C]. Generate a practice scenario that requires me to use ALL THREE skills to solve a single problem. Do not tell me which skill applies where."
🛠️ The 4-Step "Struggle" Workflow
We can't just ban AI (that's a career killer). Instead, we need to move from the "Knowledge Economy" (collecting facts) to the "Judgment Economy" (making decisions). Here is how to apply the science above in a specific workflow.
Step 1: Engage (The Blank Page)
Before you write a prompt, spend 20-30 minutes wrestling with the problem alone. Write the bad code. Draft the messy outline. This warms up your neural networks.
Step 2: Spar (The Dialogue)
Do not ask for the answer. Ask for a fight.
The "DeepMind" Protocol:
Want a prompt straight from a research scientist at DeepMind to help you learn through “Socratic tutoring”? Shared by Dwarkesh Patel (of the Dwarkesh pod), this prompt gets your AI to keep “asking you probing questions which reveal how superficial your understanding is.”
Copy/Paste this at the start of your next chat:
"I would benefit most from an explanation style in which you frequently pause to confirm, via asking me test questions, that I've understood your explanations so far. Particularly helpful are test questions related to simple, explicit examples. When you pause and ask me a test question, do not continue the explanation until I have answered the questions to your satisfaction. I.e. do not keep generating the explanation, actually wait for me to respond first. Thanks!"
Step 3: Synthesize (The Forge)
Once you have the solution, close the AI tab. Re-write the solution from scratch in a blank document.
- The Quote Test: Can you explain the concept to a 5-year-old right now without looking? If not, you haven't learned it.
Step 4: Architect (The System)
Use the Spaced Repetition hack mentioned above. Use AI to build the calendar, not to store the memory.
Expert Insights & Professional Advice
Dwarkesh Patel gave a great talk on the future of AI at HubSpot INBOUND, some of which might be familiar to us Neuron readers, and some that’s new (like his current timeline for a continually learning AGI 👀) but most importantly, he had this advice to share on what he uses AI for:
- Build a master doc with everything: Dwarkesh keeps one 20K-word Google Doc with all his work context—problem logs, meeting notes, email templates, common prompts. He pastes it at the start of relevant AI sessions. LLMs can instantly digest hundreds of pages of your company knowledge before answering. Humans can't do that.
- Use AI as a Socratic tutor, not a lecturer: Instead of “explain X,” try: “Act as a Socratic tutor. Ask me questions that help me understand this concept. Don't move on until I've proven I get it.” Research by Benjamin Bloom found that one-on-one tutoring beats classroom learning by two standard deviations. We can finally access expert tutors across every field at a moment's notice—but only if we prompt them right (or use the “study and learn” mode).
- Don't wait for your org's AI tools: They're slow and often outdated. Instead, experiment yourself. Most AI tools are free or $20/month (though the best tiers cost ~$200, and the level of quality you get is often much higher, and if you use the API, you can pay per use via a tool like OpenRouter; here’s how).
Our favorite insight: LLMs can't learn on the job over months like humans, but they can instantly absorb your entire institutional knowledge before every single response… a superpower even humans don't have.
The Bottom Line
In the Judgment Economy, your value isn't what you know—it's your Taste (knowing what to build), your Agency (driving results), and your Learning Velocity.
If you let AI do all the thinking, you are bankrupting your future self. But if you use AI to make the struggle more efficient, you build a moat that no algorithm can cross.
