It turns out that interest in, hype about and spending on artificial intelligence (AI) is cyclical, boom-and-bust, peaks-and-valleys. When the excitement is high, it’s called “AI summer.” And when the excitement is low, it’s called “AI winter.”
Although AI seems to have become widespread and mainstream in the past three years, in fact AI summers and winters have come and gone several times in the past.
Applying seasons to artificial intelligence interest started in 1984 at a meeting of the American Association of Artificial Intelligence (AAAI). They described the 1970s as an “AI winter,” where experts got pessimistic, the press turned negative and research funding dried up. Then the 80s were an “AI summer,” where hype around AI got out of hand and the claims and promised went too far, but the money was flowing.
Yada, yada, yada — the last three years were an even brighter summer than the 80s, with AI becoming the ultimate marketing buzzword, with claims about AI as the solution to every human problem (even the problem of Facebook) and the dollars spent completely out of control.
But in the past year, a few voices have been warning about the coming AI winter, including Thomas Nield, Kathleen Walch and most recently, the BBC’s Sam Shead.
I also think the AI winter is coming, in part because AI has failed to live up to our expectations. The problems is that last 1%. I described this phenomenon in a column in Computerworld called “Why autonomous cars won’t be autonomous.“
I remember the first DARPA Grand challenge — the Pentagon’s first contest for self-driving cars in 2004. I was there, and saw the cars roll off the starting line and immediately start crashing into things. A self-driving motorcycle just lunch forward, then fell over. None of the vehicles made it to the finish line.
The next challenge was totally different. Just one year later, all but one of the entries finished the track. Everybody thought — wow, at this rate we’ll have fully autonomous cars on the road by 2010.
Nope. The problem is that (as with any super complicated AI task), fast progress at the beginning doesn’t prepare you for the extremely slow progress at the end. It turns out that full autonomy is probably still a decade away.
People think self-driving cars are already here. But the real-world examples involve control rooms of people to compensate for unexpected problems that always crop up. And the same goes for most ultra-compex AI problems, like human-like virtual assistants. The companies hyping AI try hard to conceal the degree to which humans intervention is still required.
So the AI winter is coming. It’s good that maybe the hype will die down. It’s bad that the funding may dry up. But good or bad, winter is coming.