Science
Why a Leading Scientist Says AI Doesn't Actually 'Think' Like We Do
A chatbot can write a legal brief in seconds yet stumble on a puzzle a five-year-old solves easily. Computer scientist Melanie Mitchell argues this gap reveals we are measuring machine intelligence the wrong way.
70+ yearshow long scientists have debated how to define machine intelligence
The facts
- 1Computer scientist Melanie Mitchell argues that AI systems do not 'think' or 'reason' the way humans do, even when they pass human-designed tests.
- 2Many AI benchmarks were originally built to test humans, so a high score does not prove the same kind of understanding or true generalization.
- 3Mitchell studies analogy-making and abstraction, arguing these skills expose real reasoning better than pattern-matching across huge text datasets.
- 4Without sharper cognition tests, companies and governments risk overestimating how reliable AI decisions are in real-world use.
- 5Debate over how to define and measure machine intelligence has continued for more than 70 years, since Alan Turing's 1950 proposal, and remains unresolved.
Why it matters
How intelligence gets measured shapes how much trust society places in AI for medicine, hiring, and law, so flawed tests can lead to overconfident or unsafe decisions.
Sources
- Quanta Magazine
- Santa Fe Institute
