Genius Makers

AI pioneer Geoff Hinton features prominently in Cade Metz’s new book, Genius Makers.

How does it feel to see your life’s work go up in smoke? In the early 2000s, the computational linguist Chris Brockett had a sudden panic attack when he realized that a new crop of machine learning methods would make his research obsolete. The anxiety set in when it dawned on him that he had wasted nearly seven years of his life writing down linguistic rules for natural language processing. His colleagues thought he was having a heart attack and rushed him to the hospital. “My fifty-two-year-old body had one of those moments when I saw a future where I wasn’t involved,” he later reflected.

Many AI researchers experienced a similar shock in 2012 when Geoff Hinton and two of his grad students showed that deep neural networks could beat state-of-the-art AI systems in image recognition. Hinton belonged to a small group of academic contrarians––the “neural network underground”––who bet their careers on a concept that was long dismissed as a technological dead end. “Neural networks are for people who don’t understand stats,” they were told. But Hinton’s gang had the last laugh––much to the dismay of their detractors who had invested themselves in “shallow learning” methods.

Progress, of course, didn’t stop with image recognition. Since 2012, neural networks have achieved similar breakthroughs across previously intractable problems, ranging from machine translation and language generation to solving the conundrum of protein folding. These advances have changed the technology industry in profound ways and set off a global arms race for top AI talent. It has also led to a fundamental shift in how software is being developed: instead of programming software by writing explicit instructions, we now increasingly train software by showing labeled examples. The new mantra is to throw just enough training data at a problem until it’s solved. I’ve witnessed this shift myself over the years when I co-founded a company with one of Hinton’s former doctoral students.

Cade Metz’s new book, Genius Makers, chronicles the AI miracles of the past decade from the vantage point of its creators. It’s a very readable and informative history of modern AI aimed at a general audience. The great strength of the book is that it avoids the common pitfall of tipping into hyperbole. Instead, it reminds us that technology always reflects the values, biases, and incentive systems of its makers. Although the narrative holds few groundbreaking revelations for people who are active in the field, it’s still fun to read about a subject when you’ve met many of the key protagonists in the flesh. And let’s be honest: Hinton’s oft-quoted wry sense of humor is worth the price of admission alone.