Can AI Have an IQ? What Happens When Chatbots Take IQ Tests

Quick Answer

AI language models can answer IQ-test questions and sometimes score high, but the resulting number isn't a real IQ. IQ scores are normed against human age groups and assume the test-taker hasn't memorized the material — assumptions AI breaks completely, since models may have trained on the very questions and lack the general embodied cognition the score is meant to summarize.

Why an 'AI IQ Score' Is Misleading

An IQ score means one thing: how a person performed relative to a human norming sample of their age. Feeding the same questions to an AI produces a number without that meaning. Worse, public IQ-test items circulate widely online, so a model may have effectively seen the answers during training — the equivalent of grading a student who memorized the answer key.

Researchers who evaluate AI reasoning use benchmarks built to resist memorization, like the ARC (Abstraction and Reasoning Corpus), which tests novel pattern induction. AI systems have improved rapidly on such benchmarks, but their failure patterns remain very unlike human ones — models can solve graduate-level problems while fumbling puzzles most children find trivial.

What AI Performance Does Tell Us

The interesting finding isn't the score — it's the profile. Language models excel at verbal analogies, vocabulary, and general knowledge (crystallized-intelligence analogues) while remaining less reliable at genuinely novel visual and spatial reasoning, precisely the fluid-intelligence tasks IQ tests weight most heavily for humans.

For human test-takers, the takeaway is reassuring: your IQ score describes your standing among people. No chatbot's benchmark performance changes what your 115 or 130 means, any more than a calculator's arithmetic invalidates a math grade.

Frequently Asked Questions

What is ChatGPT's IQ?

Headlines assigning ChatGPT an IQ (often 120–155) come from feeding it test questions it may have seen in training, scored against human norms that don't apply to software. AI researchers instead use contamination-resistant benchmarks like ARC to measure machine reasoning.

Can AI pass a Mensa test?

Modern models answer many Mensa-style verbal and logical questions correctly, and some vendors have claimed passing-level performance. But Mensa admission requires supervised human testing — and a score normed on humans has no defined meaning for a machine.

Will AI make IQ tests obsolete?

Unlikely. IQ tests exist to compare humans with humans for education, diagnosis, and research. AI progress changes none of that, though it has pushed test publishers to guard item security and researchers to design memorization-proof reasoning benchmarks.