> Where will AI be in a "full generation"? I don't know, but a good bet is getting very close to human-level intelligence

LOL, no. The amount AI would have to advance from its current state to "very close to human-level intelligence" would have to be orders of magnitude more than it has evolved from its initial state to its present state (over the last 50-60 years). Pray tell, what technologies can you name that evolved faster in their second half-century than their first?

Does that mean we definitely won't have human-scale AI in 50 years? No, because revolutions are impossible to predict. What it does mean, however, is that human-scale AI is not the predictable outcome of existing evolutionary trends. Indeed, developments in AI more or less stalled out in the 1980's, most of what we've seen to date is new applications of techniques pioneered decades ago.

To use an analogy, what do you think CPU clockspeeds will be 10 years from now? It's possible that some breakthrough will happen and we'll have 20 GHz processors, but projecting existing trends suggests that the increase in clockspeeds from 2013 to 2023 will be a lesser percentage than what we saw from 1993 to 2003.

If you think AI stalled out in the 80s, you deserve a LOL right back. We didn't even have Bayesian nets in the eighties. AI has made huge progress since the eighties, look at Geoff Hinton's work, look at all Sebastian Thrun has done with automated cars - something many experts thought would take 50 years in the late nineties. AI is just now blossoming. It's a very exciting field at the moment.

Wikipedia puts the development of Bayesian networks in the 1980's. As for Hinton: I didn't mean to say we had zero progress since the 1980's (though admittedly, that's what "stalled" technically means), but rather that after a flurry of rapid development we hit a point of diminishing pace of development, as all technologies do.

I don't know if AI is blossoming right now, but all the interesting things I've seen in the field (Watson, self-driving cars) are not theoretical breakthroughs, but rather new applications. I don't think the technology in self-driving cars gets us fundamentally closer to human-scale AI.

I think what happened is that it fell out of popular usage to call most AI applications "AI", and a lot of the more extreme claims about the near-term potential of AI died out, so there is something of a perception in some corners that the field had died out.