50 million bits is enough to specify a human-level intelligence, but only if one has the appropriate machine. In terms of constructing an AI, the human genome is doing little more than taking clever advantage of ways to cause things in the physical world to self-organize (as per the book, Notes on the Synthesis of Form). Physical reality itself is the execution machine, is responsible for much of the end product, and has way more than 50 million bits of complexity.
If we are thinking of running our first AI on something resembling current computer hardware, then we are going to run it in a extremely impoverished environment, as compared with actual physical reality. This means that the AI-specifying algorithm will have less environmental complexity to work with than the human-intelligence algorithm does, and this in turn means that the AI-specification will need to be longer and more complex.
So from a practical standpoint, while one could in theory specify an AI in 50 million bits, in practice the first AI will need a much larger specification, and/or will need to run on some very powerful computer hardware.
Just how powerful can probably be estimated, but it has little to do with the fact that the human genome contains 50 million bits of information. In terms of determining when AI will happen, the 50 million bit argument is a red herring.
I don't agree with Kurzweil on much, but the more I read the back and forth, the more it seemed to me that Myers was totally out of his depth. He completely misquoted what Kurzweil said, and then when that came to light he followed it up with fairly thorough misinterpretations of what Kurzweil was saying. I don't think he was doing that intentionally: he just didn't understand what he was saying. On my reading, Kurzweil wasn't really delving into genomics (this was entirely a red herring): he was staying on firm ground in information theory, cognitive science, and computer science, and in these areas it was Myers who showed a fair lack of acknowledgment for his own ignorance in the field.
2 comments
[ 3.3 ms ] story [ 18.5 ms ] threadIf we are thinking of running our first AI on something resembling current computer hardware, then we are going to run it in a extremely impoverished environment, as compared with actual physical reality. This means that the AI-specifying algorithm will have less environmental complexity to work with than the human-intelligence algorithm does, and this in turn means that the AI-specification will need to be longer and more complex.
So from a practical standpoint, while one could in theory specify an AI in 50 million bits, in practice the first AI will need a much larger specification, and/or will need to run on some very powerful computer hardware.
Just how powerful can probably be estimated, but it has little to do with the fact that the human genome contains 50 million bits of information. In terms of determining when AI will happen, the 50 million bit argument is a red herring.