Gary Marcus is a cognitive scientist, author, and one of the most visible critics of the claim that current deep learning is on a direct path to artificial general intelligence. He is a professor emeritus at New York University, founded the machine-learning startup Geometric Intelligence (acquired by Uber in 2016), and later founded Robust.AI. His books include “The Algebraic Mind” and, with Ernest Davis, “Rebooting AI.”
Marcus’s recurring argument is that deep neural networks, however impressive at perception and pattern completion, are weak at the things he associates with robust general intelligence: compositional reasoning, reliable handling of novel situations, explicit world models, and freedom from confident errors. In his 2018 paper “Deep Learning: A Critical Appraisal” he laid out ten concerns and argued that deep learning “must be supplemented by other techniques if we are to reach artificial general intelligence” - a position usually called neuro-symbolic or hybrid AI. He does not call for abandoning deep learning, but for combining it with the symbolic and structured-knowledge methods that connectionists had largely set aside.
Through his “Marcus on AI” newsletter and frequent public commentary he has become a leading skeptic of AGI hype cycles, warning about hallucination, unreliability, and overclaiming by labs. His debates with deep-learning advocates - among them Yann LeCun and Yoshua Bengio - are a long-running fixture of the field’s argument over how far scaling current methods can go. Supporters credit him with documenting failure modes that the mainstream downplayed; critics argue he understates how much capability raw scale has delivered.