I'd like to believe there's magic. Mama told me magic or illusion doesn't mean it's fake.
Robotics pioneer Rodney Brooks debunks AI hype seven ways
That's a great article, thanks for sharing it. I like Brooks a lot and I have a bunch of his books of both the lay and the professional flavor.
He makes a lot of good points directly and those are mostly true, but he also implies something else which is important and false. By providing a list of ways that people warning of a potential great leap forward in AI could be mistaken he strongly suggests that people who seriously consider that scenario are not thinking carefully, and he also implies that great leaps forward do not occur. This is demonstrably false, especially in the field of neural networks today, because we've seen them happen recently and the closer you look at the field the more of them you will see.
The 2012 ImageNet results were one of those great leaps forward. The world's assembled masters of vision processing technology stood agape as a technique which nearly all of them believed was not ever going to work posted a set of results that exceeded anything they expected to see for years, perhaps decades. And in the ensuing years this new NN technique time and again outstripped the most optimistic expectations of experts and led us into a new era where problems that have stood against the best efforts of our finest minds for *decades* are falling left and right just as soon as some grad student has the time to apply NN to the question.
DeepMind's work with RL applied to Atari video games made the cover of the most prestigious peer reviewed journal in the world not because it was amusing, or well presented, or a good human interest story. It took that honor because it was an astounding result, far ahead of anything that the field expected. Subsequently Deep Mind's work with AlphaGo Fan, AlphaGo Lee, AlphaGo Master, and now AlphagoZero has step-by-step crushed out expectations for what is possible. Yes - those are four different systems which have elements in common, but each of them include fundamental new capabilities and each represents a giant leap beyond the performance of it's predecessor.
In the world of AI today the people who want to argue that further 'miracles' will not appear now have the burden of proof to bear. Of course it is very hard to say exactly where this all leads because these giant leaps occur in a hard to guess sequence and often in hard to guess directions. But the idea that more giant leaps are not coming is the least likely thing of all.
