PERSON
Gary Marcus
The field's most disciplined skeptic—cognitive scientist, hybrid-AI advocate, and the man who said deep learning was hitting a wall and was punished for the timing before he was vindicated by the outcome.
Gary Marcus is the most demanding believer artificial intelligence has. Trained not as an engineer but as a cognitive scientist—he earned his doctorate under Steven Pinker studying how children acquire the past tense—he reads today's systems the way he once read a child saying
goed: as behavior that reveals an architecture. What he sees is a machine magnificent at interpolation and fragile at everything that matters: reasoning, causation, the handling of the genuinely new. In March 2022, at the peak of scaling euphoria, he published "Deep Learning Is Hitting a Wall," and the phrase became a meme his opponents waved at every new model. But the wall he described was never a wall that stopped progress; it was a wall that stopped a
particular kind of progress—the kind that mistakes fluency for understanding. His diagnosis runs parallel to
Judea Pearl's: that
large language models are sophisticated
curve-fitters, and that genuine intelligence demands the structure, symbols, and
causal reasoning that pure pattern-matching cannot