What is Moravec’s paradox? Moravec’s paradox is that AI machines find it very easy to do things that we find difficult, such as chess, but very difficult to do things that we find easy, such as tying a shoelace.
The paradox has Hans Moravec’s name, but he wasn’t the only person to notice this. He is a computer scientist and futurist who was born in Austria, but currently lives in America. He was involved with AI research in the 1970s when this paradox was first suggested. At the time, AI was in its infancy, but the paradox is even clearer today. AI can do staggeringly impossible things at ridiculous speeds. It can create artwork that would take the best of us months to do. It can write stories in seconds. It can consistently win chess games against even the best chess grandmasters, and it can strategize to levels that humans can only dream of. Yet, despite all of this capacity, AI cannot do things that even a one-year-old can do. It cannot pick up a piece of string and push it through a hole. It would have trouble putting the correct shaped blocks in the correct holes on that popular children’s toy. It cannot pick up a ball and put it in the correct box. Why is that? Well, when you think about it, it is fairly easy to understand.
The tasks that AI does well at but we are not so good at have very clearly defined rules. They involve crunching enormous amounts of data and analyzing millions of potential routes to find the correct one. When it comes to chess, AI will always beat humans because it has the ability to analyze any move to the Nth degree. Humans don’t have the ability to think that far ahead because there are too many possibilities. Grandmasters can think further ahead than the average person, but even they cannot analyze as far ahead as an AI can. Every pair of moves in a chess game has 103 possible outcomes. And those numbers start to multiply so that in the average chess game there are over 10,000 possible unique games. Chess grandmasters cannot think that far ahead, so they use previous games and experience to help them make choices of what to play. AI can think that far ahead, but, not only that, AI can also get better with experience. Humans play other humans to get better at chess. AI can play games with itself, and it can play games at supersonic speeds to rapidly improve. AI can do this with any kind of skill that requires reasoning and planning.
The tasks that we do well at but AI has trouble with involve perception, motion, and sensory data. You can pick up a toothpaste cap that has rolled into the sink and screw it back onto the toothpaste tube without even thinking about it. That would be a mammoth task for an AI. There is so much sensory data that needs to be processed and so many fine motor skills that it would be impossible. We can do these kinds of tasks without even thinking about them. Walking up stairs. Picking up a cup of coffee and carrying it without spilling any. Tying a shoelace.
According to Moravec, the main reason for this is that we have spent millions of years evolving the ability to do these things. Our sight, hearing, motion, and all of our other skills have been finely honed by evolution to the point they are at now. They are not the best in the world, but they enable us to do a huge number of beneficial tasks with ease. However, skills like reasoning and planning are relatively recent in terms of evolution. We probably evolved the ability to reason 100,000 years ago. We haven’t had as long to get better at it. The older a skill is, the more effortlessly we can do it, and the more recent a skill is, the harder it is for us to do. This is the exact opposite for AI.
All of this being said, AI is advancing at an alarming rate and may soon be able to do all of these things. However, just because it can do them, doesn’t mean it will do them. To build an AI robot that can pick up a cup of coffee and carry it to a table would cost a billion times more than just paying a person to do it. In the end, it will probably come down to economics, and we are cheaper. And this is what I learned today.
Try these next:
Sources
https://en.wikipedia.org/wiki/Moravec%27s_paradox
https://en.wikipedia.org/wiki/Hans_Moravec
https://techevaluate.com/is-ai-better-than-humans-at-chess
Photo by Pavel Danilyuk: https://www.pexels.com/photo/elderly-man-thinking-while-looking-at-a-chessboard-8438918/