By Cade Metz
The people behind the AI revolution
Subject
Themes
A small group of researchers nurtured an idea for decades, often in the face of ridicule, before it suddenly came of age and they were pulled into the machinery of some of the largest companies on Earth. The idea was the neural network, a mathematical system modelled on the brain that could learn from data rather than follow hand-coded rules. Frank Rosenblatt built the first one in 1958, but Marvin Minsky's critique drove the technology into obscurity. Geoff Hinton, born into a distinguished British scientific family, kept the faith through two AI winters, developing backpropagation with David Rumelhart and rebranding the field as "deep learning." Yann LeCun built convolutional neural networks at Bell Labs that read over 10 percent of American cheques. Yoshua Bengio carried the torch at the University of Montreal. For years, the three worked with almost no funding and widespread hostility from the academic establishment.
A small group of researchers nurtured an idea for decades, often in the face of ridicule, before it suddenly came of age and they were pulled into the machinery of some of the largest companies on Earth. The idea was the neural network, a mathematical system modelled on the brain that could learn…
Genius Makers arrived at a pivotal moment in the public understanding of artificial intelligence, published in 2021 just as the technology was about to undergo yet another dramatic acceleration with the emergence of large language models like GPT-3 and the subsequent explosion of generative AI. Cade…
The most striking pattern in the development of artificial intelligence is how a small group of researchers sustained their belief in neural networks through decades of institutional hostility. Frank Rosenblatt faced Marvin Minsky's public ridicule and saw his life's work consigned to obscurity.…
The acquisition of Hinton's three-person start-up for $44 million set off a chain reaction that transformed the technology industry into a battlefield for human capital. Within months, Google paid $650 million for DeepMind's fifty employees. Facebook built FAIR by personally courting researchers.…
When Google Photos labelled eighty photos of Jacky Alcine's Black friend as "gorillas" in 2015, the company's response was revealing: it removed the gorilla tag entirely rather than fixing the underlying problem. Five years later, the service still blocked anyone from searching on the word. This…
From Frank Rosenblatt's 1958 prediction that the Perceptron would "walk, talk, see, write, reproduce itself, and be conscious of its existence" to Sam Altman's declaration that OpenAI would build AGI for $25 to $50 billion, the history of artificial intelligence has been defined by cycles of…
DeepMind's founders insisted, from the very first line of their business plan, that they were building artificial general intelligence. They also insisted, from the very beginning, that this technology could be dangerous. When they sold to Google, they demanded two contractual conditions: no…
In 1958, a twenty-nine-year-old Cornell professor named Frank Rosenblatt stood inside the United States Weather Bureau in Washington, D.C., demonstrating a machine that could learn to distinguish marked cards. He called it the Perceptron, a mathematical system modelled on the web of neurons in the…
Stanford professor Andrew Ng pitched deep learning to Google CEO Larry Page over lunch in late 2010, sketching a line graph with his hands: as neural networks consumed more data, they grew more accurate, and Google had more data than anyone. Project Marvin launched inside Google X, named with…
Elon Musk warned that AI was "potentially more dangerous than nukes," pointing to DeepMind as evidence that danger was five to ten years away. At a dinner at Zuckerberg's Palo Alto home, the Facebook team tried to convince Musk his fears were overblown. Musk was unmoved. Yet days later he phoned…
Ian Goodfellow invented generative adversarial networks while slightly drunk at a Montreal bar called Les 3 Brasseurs. His colleagues argued about generating photo-realistic images; Goodfellow proposed pitting two neural networks against each other, one creating images and the other judging them.…
Zuckerberg testified before Congress wearing his "I'm sorry suit" after Cambridge Analytica harvested data from 50 million Facebook profiles. He promised AI would solve the fake news problem, claiming it already caught 99 percent of ISIS content. But the technology could not distinguish a…
Inside Geoff Hinton's office on the fifteenth floor of the Google building in downtown Toronto, two white blocks sat on the cabinet near the window. Each was about the size of a shoebox, oblong and triangular. He would hand them to visitors and ask them to reassemble the pyramid. Two tenured MIT…
By Cade Metz
Subject
Themes
A small group of researchers nurtured an idea for decades, often in the face of ridicule, before it suddenly came of age and they were pulled into the machinery of some of the largest companies on Earth. The idea was the neural network, a mathematical system modelled on the brain that could learn from data rather than follow hand-coded rules. Frank Rosenblatt built the first one in 1958, but Marvin Minsky's critique drove the technology into obscurity. Geoff Hinton, born into a distinguished British scientific family, kept the faith through two AI winters, developing backpropagation with David Rumelhart and rebranding the field as "deep learning." Yann LeCun built convolutional neural networks at Bell Labs that read over 10 percent of American cheques. Yoshua Bengio carried the torch at the University of Montreal. For years, the three worked with almost no funding and widespread hostility from the academic establishment.