Why Today's Humanoids Won't Learn Dexterity

Summary

In this post I explain why today’s humanoid robots will not learn how to be dexterous despite the hundreds of millions, or perhaps many billions of dollars, being donated by VCs and major tech companies to pay for their training. At the end of the post, after I have completed my argument on this point, I have included two more short pieces. The first is on the problems still to be solved for two legged humanoid robots to be safe for humans to be near them when they walk. The second is how we will have plenty of humanoid robots fifteen years from now, but they will look like neither today’s humanoid robots nor humans. [[My side commentaries look like this.]] 1. Prolog Artificial Intelligence researchers have been trying to get robot arms and hands to carry out manipulation of objects for over 65 years; since just a few years after the term Artificial Intelligence first appeared in a proposal for a 1956 “Dartmouth Summer Research Project on Artificial Intelligence”. By 1961 Heinrich Ernst had produced a PhD thesis describing a computer controlled arm and hand that he had connected to the TX-0 computer at MIT, and had it picking up blocks and stacking them, and stunningly there is a video. His advisor was Claude Shannon, and he also thanked Marvin Minsky for his guidance, thus naming two of the four authors of the Dartmouth AI proposal. This led to industrial robots, which were and are computer controlled arms with various “end effectors”, think primitive hands, that have been used in factories around the world for sixty years Recently a new generation has stumbled upon the idea of building humanoid robots and you may have noticed just a little bit of hype about it. Gartner says it is early days and we are nowhere near maximum hype yet. This diagram is just a year old, and humanoids are at the very beginning of the cycle, while generative AI is over the hump and heading down to the doldrums: The idea is that humanoid robots will share the same body plan as humans, and ...