The field of A.I., of using computers to perform complex tasks as well as a human, is not new. People have been using deep learning with neural networks, a major subfield of A.I., for 40 years. A car used it to drive itself across the United States in 1995. But things started to change in 2010 and 2011 when a new set of academic papers identified the potential for machine learning models to become much better with scale: hundreds of times as many parameters for the algorithms and thousands of times as much data to train on. Computer systems working on a massively increased scale could produce not just quantitative growth, but a qualitative improvement in what machine learning could accomplish.
“Taking over the world is an intensely human thing to want to do,” says Astro Teller, in a short interview conducted at the 2017 Aspen Ideas Festival. At Google X, Teller studies and develops artificial intelligence. Here, he argues that current frenzy over the topic might be overblown.
When you consider the ways A.I. meets or supersedes human performance, it is a normal reaction to be anxious about the change that represents. As with most new technologies, when people sit and think about what’s going to happen, they don’t make lists of the good stuff. And we have to have compassion for that. We cannot ignore the challenges presented by A.I.
No one would go back 90 years and say, “Down with bulldozers!” and get out the shovels. Every time humanity has invented a new bulldozer, we’ve been able to go bigger, deeper, faster. A.I. will do the same. It will be a lever to help human minds solve problems the world faces.