The Deep Learning Revolution
Terrence J. Sejnowski, MIT, 352 pages, $29.95. Artificial intelligence has many definitions, but broadly it refers to software that perceives the world or makes decisions. It uses algorithms, or step-by-step instructions. Within AI is an area called machine learning, in which algorithms are not hand-coded but trained. And within machine learning is deep learning, which uses algorithms loosely modeled on the brain. So-called neural networks pass data among many connected nodes, each performing a bit of computation, like the brain's neurons. Deep learning is behind self-driving cars, speech recognition and superhuman players of Go and poker. In "The Deep Learning Revolution," one of its pioneers — Terrence J. Sejnowski, who created new forms of information processing in computers as a computational neuroscientist at the Salk Institute for Biological Studies and president of NeurIPS— traces its history. One is struck by how badly even experts misjudge the progress of this technology. A key ancestor of deep learning was a one-neuron algorithm developed in the 1950s called a perceptron. Some influential researchers in 1969 argued that multiple layers of perceptrons might not be trainable, and their pessimism caused a "winter" in the field that lasted until the 1980s. Sejnowski's book is part textbook and part memoir, with varying levels of accessibility. Those with an existing interest in the topic will be charmed and enlightened. Some anecdotes have a you-had-to-be-there quality. Much separates AI from people, but Sejnowski has hope for our ability to reverse-engineer the brain.