[Dragar]

Originally Posted by lisarea

Well, then, what is the technological singularity that we're in no danger of? The definition that most use is of 'smarter than human intelligence,' or of systems that function in ways we don't understand.

So what does that mean, anyway? Computers have been faster and more reliable than us at certain types of tasks pretty much forever. It's not that humangs--and I mean humangs in the universal or collective sense, not existential--are incapable of understanding how to do math, but we don't feel like it because math is boring and stupid so we let the boring and stupid computers do it instead. Even if a computer is doing some kind of calculation that we literally don't have the collective time for, like if every person in the world were to collaborate on it, it still doesn't make it unfathomable. It's not doing something totally inconceivable, it's just doing too many conceivable things for us to keep up with sans computers.

For technology to actually do something that is beyond human comprehension, it would have to employ some kind of supernatural force.

So again with language. Humans are capable of using natural human languages innately, but we are not capable of describing them. Early attempts at natural language processing were sort of brute force rule-describing programs, but there is only so far we could go with that because of the sheer volume of rules, and because we haven't articulated most of them yet. We could. There are rules. We don't have time for that shit is all. So modern NLP focuses more on machine learning. That is, rather than trying to write down all the rules that govern language, computers observe language as it's used and make their own observations just like human children do. Computers still can't do this at the level that humans do by a long stretch, but they're already doing things that people haven't fully articulated, similar to the way people do.

When most people refer to the singularity, they talk about the point machines become capable of designing more intelligent machines. Hence the rate at which computers progress diverges, or the timescale of development becomes singular (hence the name - a similar thing does happen with black holes, but not at the event horizon unless you work in silly coordinates).

I don't disagree we could make Chinese-room like constructions if we had the time and inclination, and thare are even some who make the case that this would be intelligence of a sort - perhaps even similar to human intelligence, if sufficiently sped up. But I agree with you that a better approach is the inference/machine learning one.

And yes, computers are getting better at that. And I think parallelism is a bigger paradigm shift than people realise when it comes to doing that sort of thing. But it's hard - really hard. And the successes we have are essentially brute force tecniques of doing what biological systems are clearly doing much, much better jobs of. With processor speed no longer doubling every couple of years (far from it), we can't just add more brute force computational power to solve the problem. I'm excited to see what happens next, but pessimistic it will be anything dramatic in machine intelligence within our lifetime.