Gary Marcus
Cognitive Scientist & Author
Cognitive scientist, author of Taming Silicon Valley, and the media’s go-to AI skeptic — he testified to the U.S. Senate that we have built “bulls in a china shop.”
Why this score
How the formula works →- 22%antiCalls AI overhyped
- 22%nuancedHolds a documented two-sided view
- 20%anti4 anti-AI statements (forceful)
- 18%nuancedCampaigns for AI safety rules
- 18%nuancedSupports AI only with conditions
Includes the ×1.5 multiplier because AI is their life’s work. Evidence splits 0% pro, 42% anti, 58% direction-free → Nuanced
In his words
We have built machines that are like bulls in a china shop — powerful, reckless, and difficult to control.
The choices we make now will have lasting effects for decades, maybe even centuries.
Biography
Gary Marcus (born 1970) is an American cognitive scientist, professor emeritus at NYU, and serial author-entrepreneur: he founded the machine-learning startup Geometric Intelligence, acquired by Uber in 2016, and wrote Rebooting AI (2019) and Taming Silicon Valley (2024). Trained under Steven Pinker, he has spent two decades arguing that pure neural networks are a dead end without symbolic reasoning — a position that made him deep learning’s in-house heckler long before ChatGPT.
Skeptic-in-chief
When generative AI exploded, Marcus became the press corps’ default counterweight — cataloguing hallucinations, debunking capability claims, and predicting on his widely read Substack that scaling alone would hit a wall. In May 2023 he testified before the Senate Judiciary subcommittee alongside Sam Altman, warning that “democracy itself is threatened” by machine-generated persuasion and calling for an FDA-style agency that would make AI developers prove safety before deployment.
Crucially, Marcus is not against artificial intelligence — he is against the current recipe for it. He argues large language models are unreliable, ungrounded, and overhyped, and that trustworthy AI will require hybrid, neurosymbolic approaches plus real regulation. This has earned him fire from both directions: boosters call him a perpetual naysayer who moves goalposts as models improve, while some critics of the industry find him too invested in AI succeeding at all.
His refrain — that we are deploying systems we do not understand into infrastructure we cannot afford to break — has proven durable, even as the systems themselves keep outrunning his more specific predictions.
Quote sources
- U.S. Senate testimony, 2023(techpolicy.press)
Sources & further reading
Canonical record: https://battlelines.ai/topic/gary-marcus
