PERSON
Yann LeCun
The French-American engineer who built the machines and refuses to be impressed by them—Turing Award laureate and architect of modern deep learning, and its most insistent critic, who argues that the road to genuine intelligence runs through world models and grounded perception, not through scaling language prediction.
Yann LeCun is a contrarian with receipts. He shared the 2018 Turing Award with Geoffrey Hinton and Yoshua Bengio for the deep learning methods that now power virtually every AI system on earth, and he has spent the years since arguing that the direction those methods have taken is a detour rather than a destination. He built
LeNet at Bell Laboratories in the late 1980s—a convolutional neural network that by the late 1990s was reading a significant share of American bank checks—and he held the idea through the long winter when the field had abandoned neural networks entirely, which is why he does not mistake majority views for settled truths. His defining question is the one he has returned to in every public talk for forty years: how could a machine learn as efficiently as a child? A toddler learns that objects fall from a handful of observations;
large