What Went Wrong With AI?

Neuroscientist
We've been throwing the smartest people on the planet at the problem of artificial intelligence since the 1960s, and all we have to show for it is the Roomba vacuum cleaner.
  • Transcript

TRANSCRIPT

David Eagleman: When I was a child growing up, I expected that we would have robots by this point, C-3P0, and we still don't. In fact, the smartest thing that we have is the Roomba vacuum cleaner, so that means that something went terribly wrong with AI. And it forces us to ask what went wrong because, essentially, we’ve been throwing the smartest people on the planet at that problem since the ‘60s and it’s been a failed problem. The answer, I think, clearly, is that we have to listen to what Mother Nature is doing. So since the ‘60s, people have thought, okay, look, there's all this wet biological gushy stuff, let’s put that aside and just figure out what would intelligence take, how could we just do this from scratch, and that hasn't gotten us anywhere. The game now is shifting to saying, "All right, how is Mother Nature accomplishing it?"

The reason AI got stuck is because programmers have been saying, "Look, how can we solve this part of the problem and then solve this little part of the problem and then this part of the problem?" So if you have a robot and you want it to pick up wooden blocks and stack them, programmers have traditionally tried, okay, let’s solve the, find the block problem, and then let’s, once we’ve done that, solve the pick up the block problem and then the stack the block problem and so on. And this sort of division of labor seems like a pretty good idea, but it’s never really gotten us anywhere. The difference with what Mother Nature does is she doesn't come up with solutions to problems and then check the box. The brain is an end product of a chronic process of reinvention that Mother Nature always does, so there's no single solution to anything. Instead, the brain is this collection of subpopulations that all solve problems in overlapping ways. There's no single solution. So, for example, the detection of motion by the visual system, there's not just one way that the brain does that. There are multiple ways that were invented at different points in evolutionary history and they all weigh in on the same problem using different mechanisms. The same with memory, we don't have a single memory system. There are lots of ways that memories get written in the brain.

 
If we want to solve the AI we have to do it the same that Mother Nature did, which is build a machine that runs on conflict, that runs on reinventing solutions and having overlapping solutions of the problem space. And that, I think, is our best hope for un-logjamming the field of artificial intelligence.

 
Directed/Produced by
Jonathan Fowler & Elizabeth Rodd


×