Use AI to Get Smarter

Last updated: September 2026 · Honest and hype-free — no miracle claims, no invented statistics.

Here is the honest headline, up front: no study shows that using AI raises IQ or general cognitive ability. If someone sells you "AI that makes you a genius," they are selling you something.

But here is the headline that is true — and it is the entire thesis of this site: AI won't raise your IQ score, but used well it raises what your IQ can accomplish. Used lazily, it does the opposite. AI is the first tool in history that doesn't just store your thinking (notebooks, search engines) — it does thinking with you. That makes it the most powerful cognitive amplifier ever built, and the most efficient thinking-replacement ever sold. Same tool, opposite outcomes. The difference is how you use it.

This guide gives you both sides: what the research actually found, the traps to avoid, and four evidence-aligned protocols that turn AI into a thinking coach instead of a thinking crutch.

The core idea: augmentation, not IQ points

In 1998, philosophers Andy Clark and David Chalmers argued ("The Extended Mind") that when an external process does work we'd call cognitive if done in the head, it can count as part of the cognitive system — the parity principle. A calculator never raised anyone's arithmetic IQ, yet numerate people with calculators vastly outperform those without. AI is the most aggressive version yet: not just storage but active reasoning, coupled to you.

The honest framing: AI multiplies your realized intelligence — what you can actually produce, solve, and learn — not your measured IQ. But the multiplier has a human-in-the-loop requirement. The research below is essentially the empirical shadow of this argument: the coupled system is brilliant until the human stops checking, at which point it degrades to whatever the AI hallucinates.

What the research actually says

Not hype, not doomerism — the studies:

Heavy AI use is linked to less critical thinking

Gerlich (2025, Societies, n=666): frequent AI-tool use significantly correlated with lower critical-thinking scores, mediated by cognitive offloading — steepest in younger users. Lee et al. (2025, Microsoft Research + Carnegie Mellon, n=319 knowledge workers): higher confidence in generative AI was associated with less critical thinking applied to its output, while confidence in one's own ability predicted more scrutiny. The authors warn of "deterioration of cognitive faculties that ought to be preserved."

Your brain on ChatGPT (MIT, 2025)

Kosmyna et al. (MIT Media Lab, n=54, EEG across 4 months of essay writing — preprint, small sample, one task): brain-only writers showed the strongest neural connectivity; ChatGPT users the weakest — and 83% of the ChatGPT group couldn't reproduce a single quote from their own essay minutes later (vs. 11% in the brain-only group). The crucial twist: when groups swapped, writers who struggled first and then used AI showed higher engagement than those who outsourced from the start. Lead author Nataliya Kosmyna has explicitly cautioned against "ChatGPT makes you dumb" headlines — this measures engagement during the task, not brain damage.

The crutch effect in AI tutoring

Bastani et al. (PNAS, ~1,000 students): unrestricted ChatGPT raised practice scores 48% — but students scored 17% worse on the unassisted exam than students who never had AI. A hints-not-answers tutor design raised practice 127% and erased most of the harm — yet still produced no positive exam effect vs. control. The takeaway: AI tutors reliably raise performance while the tool is present; evidence they raise learning that survives the tool's removal is thin to null.

The GPS precedent

Dahmani & Bohbot (2020): greater lifetime GPS use predicted worse spatial memory, with a 3-year follow-up showing steeper decline in hippocampus-dependent navigation — and GPS use drove the decline, not the reverse. Nobody's spatial memory loss made GPS a bad tool; it made navigating without ever practicing the risk. Same for AI.

Bottom line from the evidence: nothing here raises IQ. But the pattern across every study is identical — how you use the tool determines whether it builds thinking or replaces it. Which brings us to the protocols.

Protocol 1 — Struggle first, outsource second

This is the single highest-leverage habit in this guide, and it comes straight from the MIT crossover finding: people who wrestled with the problem before touching AI engaged more deeply and learned more than people who outsourced from the start.

The rule: draft your own answer — even a bad one — before you consult AI. Ten minutes of genuine effort is enough to change the dynamic completely. Then use the AI to critique and improve your attempt, not to generate it.

In practice: writing an argument? Write your three points first, then ask the AI to find the weakest one. Learning a concept? Explain it in your own words first, then ask the AI to correct you. The struggle creates the mental hooks; the AI hangs knowledge on them. Without the hooks, everything slides off — which is exactly what the 83% who couldn't quote their own essay demonstrated.

Protocol 2 — Make the AI Socratic, not oracular

The Bastani study's one bright spot: the tutor that gave hints instead of answers avoided the learning harm. You can force any AI into that mode with how you prompt. Stop asking for answers; start asking for coaching.

Copy-paste prompts that work:

• "Don't give me the answer. Ask me one question at a time that helps me find it myself."
• "Here is my reasoning: [paste it]. Attack it — find the weakest step and tell me why it fails."
• "What am I missing? Give me the strongest counterargument to my position."
• "Quiz me on this topic with 5 questions, one at a time. Don't move on until I answer."

Notice what these have in common: the AI does the structuring, you do the thinking. That division of labor is the whole trick — offload the scaffolding, never the reasoning.

Protocol 3 — Explain it back (the Feynman lock-in)

After the AI helps you, close the tab and restate the reasoning in your own words — out loud or on paper. If you can't, you didn't learn it; you offloaded it. This is the cheapest self-test in existence, and it directly targets the failure mode the MIT study measured: people who "wrote" essays they couldn't reproduce.

Make it a rule: no AI-assisted learning session ends without a 60-second explain-back. What survives the explain-back is yours. What doesn't was never yours — it was the AI's, temporarily rented.

Protocol 4 — Verify on purpose

The Lee et al. finding is the one to tattoo on your brain: confidence in the tool is exactly what switches critical thinking off. The more impressive the AI seems, the less you check it — and that's when errors sail through.

The habit: treat every AI output as a draft to check. Spot-check one claim per answer — one date, one number, one citation. It takes thirty seconds and it keeps the verification muscle alive. Professionals who work with AI daily do this reflexively; amateurs accept first drafts. The gap between those two habits is the gap between augmentation and atrophy.

A sample AI thinking session

Here's what all four protocols look like in one 20-minute session — say you're trying to actually understand Bayes' theorem instead of just nodding at it:

1. Struggle first (5 min). Write your best explanation of Bayes' theorem from memory, even if it's wrong. Note exactly where you get stuck — that stuck-point is the most valuable thing you'll produce today.
2. Socratic AI (10 min). Paste your attempt and prompt: "Correct my explanation, but don't rewrite it — ask me questions that expose my errors one at a time." Work through its questions.
3. Explain it back (3 min). Close the tab. Explain Bayes' theorem to an imaginary friend, out loud. Stumble? Reopen, target the stumble, close again.
4. Verify (2 min). Ask the AI for its key example's numbers, then check one of them independently. Done — the concept is now yours, not rented.

Compare with the default session: "explain Bayes' theorem" → skim → feel smart → retain nothing. Same 20 minutes, opposite outcomes.

Warning signs you're offloading too much

Check yourself honestly against this list. Two or more means the tool is driving and you're cargo:

• You can't summarize what you "learned" in an AI session an hour later.
• You accept first drafts without changing anything.
• You feel uneasy starting a thinking task without opening the AI first.
• Your notes are mostly pasted, not written.
• You've stopped noticing when the AI is vague — fluency feels like truth.

The fix is never "quit AI" — it's Protocols 1–4. The GPS didn't make navigators obsolete; it made practicing navigation optional, and the people who stopped practicing lost the skill. Don't stop practicing thinking.

FAQ

Does using AI raise my IQ?

No study shows AI use raising IQ or general cognitive ability. What it raises — when used well — is your realized intelligence: the quality of what you produce, the speed you learn, the problems you can solve. Think "bigger engine output," not "bigger engine."

Which AI should I use?

Any frontier model works. The technique matters far more than the model — a Socratic prompting habit on a mid-tier model beats passive answer-copying on the best model ever built.

Is using AI like this cheating?

Depends on the context. For learning, it's the opposite of cheating — it's the most effective study partner ever built, as long as you do the thinking. For credentialing (exams, applications that must reflect your unaided ability), outsourcing is misrepresentation. Know which game you're playing.

Will AI make me dumber?

Only the passive pattern does, per the research — and "dumber" here means disengaged and deskilled in the moment, not brain damage. The active pattern (struggle first, Socratic prompting, explain-back, verify) shows the opposite signature: deeper engagement.

Want to go deeper on the tool side? Books on artificial intelligence on Amazon.