3 ms·
> Ok, but why are biological neurons exempted from this argument? Right. That's the "humans are magic" argument. It's a rehash of the "first cause" argument fr
by Animats 4y ago
> Ok, but why are biological neurons exempted from this argument?
Right. That's the "humans are magic" argument. It's a rehash of the "first cause" argument from theology for the existence of a creator. ("But who created God?" "Shut up, kid.")
It's pretty clear now that AI via correlation can do a lot, but it has limits. Deep neural nets seem to have a generic problem of getting the right answer most of the time, getting a really wrong answer some of the time, and not being able to detect the difference. We're still missing something.
What we're missing, I point out occasionally, is "common sense", defined as getting through the next 30 seconds of life without making a big mistake. We need to get to at least the mouse/squirrel level in that area, and we're not close. There was an article on HN a few days ago about someone having discovered the neural basis of panic flight escape route planning in mice. That's an important result. It's an indication that there's common sense hardware in mammal brains, and the beginning of understanding how it works.
This is a problem which is both very hard and not too rewarding, either financially or academically. Some years ago, I met Rod Brooks, back when he'd done his insect robots and was proposing Cog, a jump to human level AI which involved a human-like robot head. I asked him, why try to jump to human-level AI in one step? A robot mouse might be within reach. He said, "Because I don't want to go down in history as the man who created the world's best robot mouse." Cog was a failure, and he went down in history as the man who created the robot vacuum cleaner.
It's one of those thankless enterprises like OpenWorm, which is trying to get a simulation of the simplest known organism with a nervous system, a nematode with 959 cells. After over 20 years, it still doesn't work very well. Yet someone has to do that kind of ground work.
- foldr 4y ago> Right. That's the "humans are magic" argument. It's a rehash of the "first cause" argument from theology for the existence of a creator. ("But who created God?" "Shut up, kid.") I don't think that's a fair summary of the argument. The conclusion of the argument is a dilemma: either (i) humans are not conscious or (ii) humans are conscious but so are lots of other things (like rocks) that we don't usually take to be conscious. If you aren't willing to accept (i) or (ii), then you can either look for flaws in the argument that led to the unwelcome conclusion, or consider the possibility that there is more to consciousness than just executing a particular program. We know virtually nothing about consciousness, so we can't confidently eliminate the possibility that certain physical properties of neurons are a necessary prerequisite for it. Being sensibly open-minded about this possibility is not akin to a belief in magic. I would rather say that the magical thinkers are those who have an unwavering faith in the ability of computer programs to conjure up conscious experience in the absence of any known causal mechanisms.
- Animats 4y agoWorrying about "consciousness" is a dead end for now. We lack basic mechanisms for getting stuff done in an uncontrolled environment. That's why artificial intelligence is still too stupid to trust with anything important.
- astrange 4y ago"Humans are conscious but rocks aren't" shouldn't be any more surprising than "humans are conscious but dead humans aren't". Sometimes stuff's just dead. In fact, most of the time it is.
- YeGoblynQueenne 4y ago>> It's one of those thankless enterprises like OpenWorm, which is trying to get a simulation of the simplest known organism with a nervous system, a nematode with 959 cells. After over 20 years, it still doesn't work very well. Yet someone has to do that kind of ground work. I was considering a research project like that, on insect intelligence (I like bugs) (and bugs like me) but I was discouraged from it, the discouraging agent noting the extreme inter-disciplinary nature of such research. That's another roadblock I guess. Another form of this is that modern AI researchers tend to be completely clueless about cognitive science, psychology, and anything that has remotely to do with animal intelligence, other than computational neuroscience of the connectionist kind. People are trying to figure out how to make computers exhibit intelligent behaviour like that of animals, without having any idea what animal intelligence even looks like. Of course that's going to stall. Then again I was in an AI workshop at the Royal Society in London recently, with many illustrious folks from all the big unis in the US and EU, and after a big discussion about the capabilities of large language models, the subject of trying to simulate animal intelligence came up and was discussed in some seriousness, I think.