AI: The origin story
In this fascinating book on the history and evolution of Artificial Intelligence (AI), Cade Metz looks at all the twists and turns that the technology took to get to the point it has reached today — and the scientists, researchers and thinkers who made it all possible. Mr Metz, a Silicon Valley journalist who was earlier with Wired magazine and now works for the New York Times, explores not only the roots of the technology from its early days in the 1950s and the people who played a significant role in shaping the research but also the current worries, including bias and deep fakes, and the rivalries of different researchers to shape a technology that is clearly evolving faster than humans can cope.
What we call AI today started as the idea of neural networks in the 1940s — a study of how the neurons of the brain functioned and whether an electronic version of it could be created. In the mid-1950s, a young Cornell University professor of psychology named Frank Rosenblatt demonstrated how a computer could learn to distinguish simple patterns by itself. Though massively hyped in the US media in that era, the fact was that Rosenblatt’s creation — the Perceptron — was a primitive system incapable of any practical application.
Rosenblatt’s contemporary, Marvin Minsky, wrote a book that showed that self-learning systems supposedly mimicking the neural networks were useless. Minsky had also explored neural networks — in fact, he had probably built the first neural network machine even before Rosenblatt — but he had become convinced that this was not the right way. Minsky, John Mccarthy and Nathaniel Rochester proposed the term Artificial Intelligence in a convention, and Minsky was convinced that the way forward was not through neural networks but by using an approach called Symbolic AI. It would teach a computer to do specific things by giving very specific instructions.
Eventually, Minsky’s ideas carried the day — Rosenblatt had drowned in a boating accident — and AI research in the US would evolve using variations of the Symbolic AI approach for several decades. Funding for neural network research dried up and the field of study became pretty much a dead area.
One researcher who remained convinced about neural networks was a professor called Geoff Hinton, who moved from the US to teach at the University of Toronto, Canada. It was primarily his research, and those of his many students, that would result in the field of Deep Learning. Another researcher responsible for many breakthroughs and training students in the new methods was Yuan Lecun, a French origin computer scientist who moved to Silicon Valley.
Google would later hire Dr Hinton and many of his students. Facebook persuaded Lecun to set up its AI department. Baidu, from China, went on a war for talent with its Silicon Valley rivals. Though Microsoft was one of the first giants to spend millions on AI research and sponsor the research at many universities, it fell behind because it had initially backed the proponents of Symbolic AI and needed to catch up with Deep Learning.
The book looks at the breakthroughs — both in terms of AI research as well as hardware. GPUS gave AI research a huge boost even though they cost a lot. The head of Google search Amit Singhal was initially against Deep Learning because he felt his department would lose control. It happened anyway — and he left.
Elon Musk warned that AI, especially Deep Learning, was progressing too fast though he was not against using the technology for his own company. Google had taken an early lead in the war for talent by buying both Dr Hinton’s company as well as Deep Mind, which was working on Artificial General Intelligence. Google’s Google Brain and Deep Mind had a not-so-friendly rivalry. Google Brain, based in Silicon Valley, focused on using Deep Learning for technologies that could be rolled out quickly. Deep Mind, working as an independent lab and based in the UK, was focusing on more abstract work.
Facebook was trying to create a Deep Learning programme that would beat the world’s best Go player, but Deep Mind managed that feat much earlier.
Mr Musk eventually helped create Openai, an artificial intelligence lab that would be a counter balance to Google and Facebook and others but would give away its research free.
Mr Metz is a snappy storyteller. His book is as comprehensive as any one volume on AI history and development can be. Because of the lack of access, there is relatively less information on Chinese research, and that in EU or Japan, though they do make short appearances in the book. Though Amazon, too, features in the book, it probably did not give Mr Metz much access either.
The AI story is evolving. But for anyone interested in a volume on how AI reached the stage it has, this is the book with which to start.