It may not be on the level of the Montagues and the Capulets, or the Sharks and the Jets, but in the world of geeks the rivalry is about as intense as it gets. For decades, two competing “tribes” of artificial intelligence experts have been furiously duelling with each other in research labs and conference halls around the world. But rather than swords or switchblades, they have wielded nothing more threatening than mathematical models and computer code.
On one side, the “connectionist” tribe believes that computers can learn behaviour in the same way as humans do, by processing a vast array of interconnected calculations. On the other, the “symbolists” argue that machines can only follow discrete rules. The machine’s instructions are contained in specific symbols, such as digits and letters.
After an initial burst of enthusiasm for connectionist thinking among AI pioneers, the symbolist tribe came to dominate as researchers realised that human and machine intelligence are very different attributes. But over the past decade the maverick connectionists have had their revenge, making many of the most striking advances in AI, such as speech and image recognition systems, conversational chatbots, semi-autonomous cars and the AlphaGo program that famously beat the world’s strongest champion in the fiendishly complex game of Go in 2016.
This intellectual counter-revolution that is now rapidly transforming the world is the subject of technology writer Cade Metz’s colourful and readable book, Genius Makers. As computers steadily encroach into almost every corner of our lives, these AI researchers are emerging as the architects of our algorithmic age, shaping the information we absorb and the decisions we make. It is important to know more about who they are and what they think.
As Metz explains, the connectionist concept of neural networks, that mimic the functioning of the human brain, dates back to the 1950s. But by the turn of the century, most AI researchers had abandoned the field as a dead end. Marvin Minsky, one of AI’s most formidable and intimidating pioneers, ridiculed its practitioners, telling one researcher: “How can an intelligent young man like you waste your time with something like this? ”
However, a “neural network underground”, as one researcher called it, kept the faith alive. One champion was Geoffrey Hinton, a scion of a British scientific family and professor at the University of Toronto, who reanimated the field with extraordinary vision and determination, in spite of enduring the loss of two wives to cancer and a back condition that left him unable to sit.
There was a scrambled to sign up the world’s top researchers, who could command salaries comparable with NFL quarterbacks
As computers became more powerful, data sets exploded in size, and algorithms became more sophisticated, deep learning researchers, such as Hinton, were able to produce ever more impressive results that could no longer be ignored by the mainstream AI community. Metz describes the personal histories and advances of several of these experts, including Yann LeCun, Yoshua Bengio, Andrew Ng and Demis Hassabis, the co-founder of the London-based DeepMind, which invented the AlphaGo program.
When the potential of these deep learning systems were realised, big tech companies such as Google, Facebook, Amazon, Microsoft and Baidu in China scrambled to sign up the top researchers, who could command salaries comparable with NFL quarterbacks’. Hinton ran an auction for his three-person start-up before selling to Google for $44m in 2012. Staff costs at DeepMind, also acquired by Google in 2014, run at about $371,000 per employee.
As you would expect from a New York Times technology reporter, Metz’s book draws on extensive access and meticulous research. Over the course of eight years, he interviewed more than 400 people. Yet, though discussed, some of the most contentious issues are hardly explored to the depth they deserve.
Metz records that Margaret Mitchell and Timnit Gebru, the AI ethics researchers who have just been controversially fired by Google, were complaining about the shocking lack of diversity in the field years ago. At one AI conference, Gebru counted just six black attendees, all of them men, out of 5,500 participants. But Metz does not fully spell out the consequences of this “sea of dudes” problem, as Mitchell describes it.
Nor does he sufficiently interrogate the top researchers about the limitations and voracious energy consumption of their models. Their stated ambition of achieving artificial general intelligence (when machine intelligence surpasses human intelligence across every domain) that scares the wits out of many observers is also largely unchallenged.
Genius Makers is a good read, as far as it goes, but the reader is left with the wish that Metz had displayed deeper learning.
Genius Makers: The Mavericks Who Brought AI to Google, Facebook and the World, by Cade Metz, Random House Business, RRP£20, 384 pages
John Thornhill is the FT’s innovation editor
Join our online book group on Facebook at FT Books Café