You can generalize this to: Someone still has to understand everything in order to make it actually useful for humans. You can build a robot to do work previous

You can generalize this to: Someone still has to understand everything in order to make it actually useful for humans. You can build a robot to do work previously done by a human, but you can't make it actually understand when it needs to be working and on what. That would be an exponentially harder problem to make a computer solve. Making computers want interesting things is perhaps THE problem in artificial intelligence. If you can solve it, you've got AGI. You can hand a machine learning algorithm a very constrained environment and it can work out some basic goals, like making a high score go up. But increase the complexity of the environment and all of a sudden, the algorithm needs to learn how to simulate it, significantly ramping up the difficulty of what it has to do. Humans are going to start getting inexorably shifted out of things that humans really shouldn't be doing in the first place, to be replaced by the just-as-difficult task of commanding our new robot slaves. Maybe for a little while it'll be difficult for humans to find purchase in the brave new world, but the limitations of machines to literally read our brains and tell us what we want is going to quickly make themselves known and humans all over the world will take on the role of specifying the behavior of machines that are getting increasingly smarter. You're going to see the same sort of stratification of haves and have-nots you have now, I see little about the new production modes that's going to change the fundamentals of economics. But everybody is going to have more of it, because making things is going to get much, much easier, even easier than it is now, and it's already really easy.