Predictions about artificial intelligence (AI) have a patchy record. Any greybeard in the field will tell you tales of previous hype cycles in the 1970s and 1980s that crashed when their fabulous promises were not fulfilled.
Now, though, times are good again. A spurt of progress in machine learning, a sub-field of AI, has companies piling in. The technology is being used for everything from working out how best to aim advertisements at web-surfers to how to develop self-driving cars. A landmark was working out how to beat humans at Go, an East Asian strategy game that computers have historically found hard. An AI created by DeepMind, a British subsidiary of Google, beat a human champion of the game in 2015.
Can such progress continue? A group of researchers at the universities of Oxford and Yale decided to poll hundreds of attendees at two well-regarded AI conferences. First, the good news. The researchers believed that it would be around 125 years before computers were sufficiently advanced to be better than humans at all the suggested tasks. Those who worry about an AI “explosion”, in which clever computers design ones that are cleverer still, and do so far faster than humans can follow, can rest easy as well. The AI researchers reckoned that one of the last jobs to be automated would be “AI researcher” (although that might just suggest AI researchers are as prone to wishful thinking as anybody else).
Other milestones, though, will be reached sooner. The researchers thought a machine could win the World Series of Poker within three years, and that robots would be better than humans at folding laundry within half a decade (machines often struggle more with physical tasks than with abstract, intellectual ones). Lorry drivers will be obsolete within a decade or so, while retail workers have about 15 years left. Machines will be writing best-selling books within three decades.
Or at least, those are the averages. One of the most striking things about the survey was the wide range of predictions given by researchers. The most gung-ho thought humans would be entirely obsolete by the middle of the century; the most pessimistic (or optimistic) that it would not happen for 200 years or more. Much of the split depends on geography. Researchers from America were the most cautious. They thought, on average, that full automation would not arrive for another 169 years. Their counterparts in Asia gave it just 104.
VIDEO>> What will be the top ten stand-out moments in 2018?
The World in 2018 brings you tomorrow's water-cooler moments, today