🤖The Greats

Co-authored the 1986 paper that popularized backpropagation for training neural networks, then spent decades as one of deep learning's few consistent believers through two AI winters until his student's AlexNet won the 2012 ImageNet competition and reshaped the field almost overnight.

“Persisting with an unfashionable idea through repeated rejection can pay off once the evidence, not the fashion, catches up.”

Built the convolutional neural network at Bell Labs in the late 1980s and had it reading handwritten checks commercially by the 1990s — a working deep-learning deployment more than fifteen years before the field's mainstream breakthrough. Now Meta's chief AI scientist and a vocal skeptic that today's language models lead to general intelligence.

“A system proven in a real deployment teaches lessons no benchmark score can — even if the field has moved on.”

Advanced neural language modeling and, with colleagues, introduced the attention mechanism that fed directly into the Transformer architecture, while building the Université de Montréal's Mila into one of the world's largest concentrations of deep-learning researchers. Later became one of the field's most prominent voices calling for AI-risk regulation.

“Building a powerful technology and warning publicly about its risks are not contradictory positions to hold at once.”

A former chess prodigy and video-game designer who co-founded DeepMind in 2010, then led the teams that built AlphaGo, the first program to beat a top professional Go player, and AlphaFold, which predicted the 3D structure of nearly every known protein. Shared the 2024 Nobel Prize in Chemistry for AlphaFold.

“A system trained only on the rules of a narrow game can still surprise the humans who built it.”

Built ImageNet, a hand-labeled dataset of roughly 14 million everyday photographs, turning image recognition into a measurable annual competition; the AlexNet system that won it in 2012 is widely credited with starting the deep-learning boom. Now co-director of Stanford's Human-Centered AI Institute.

“A dataset can matter as much as an algorithm — sometimes the unglamorous infrastructure work is the real breakthrough.”

Coined the term 'artificial intelligence' while organizing the 1956 Dartmouth workshop, then invented the Lisp programming language the following year, giving the young field both its name and, for decades, its primary research tool. Won the 1971 Turing Award for founding the field's theoretical foundations.

“Naming a field decisively can shape its identity for decades, independent of what its founders accomplish that same year.”

Developed inverse document frequency in 1972 at Cambridge, the statistical insight behind TF-IDF that still underlies most search engines and much of natural-language processing, decades before the deep-learning models that now dominate the field. A lifelong, outspoken advocate for women entering computing.

“A single, precisely stated statistical idea can outlast entire architectures built decades later on top of it.”