There is no clear notion about what intelligence is. Computer science attacks the goal to produce “artificial intelligence” with mixed success. It falls short with respect to the goals it has set itself, on the other hand there are impressive results with respect to goals like “speech recognition”.
Intelligence is usually measured by tests. The tests are useful but questionable. They contain rather easy tasks that have to be solved under time constraints. So what is actually measured is how quickly people can solve routine tasks. Of course one can also prepare for tests and train to get better results. This doesn’t mean a proportional increase in intelligence.
Real world problems are rather tough. In conflicts you need “social intelligence” but it doesn’t mean that this is a time problem. A minute or two, a day or two days usually don’t make a difference. “global warming” is a nice problem, or “how to make my software competitive”. The question is whether or not you find a solution at all (or divorce with your wife, for example).
“Intelligence is the ability to solve problems that one has not been trained to solve.”
This probably implies the ability to learn on your own.
This means that a “chess program” does not show intelligence, a computer that you can communicate with and that can learn to - however badly - play simple games like “hide and seek” or “poker” would.
In YesAiCanExist, I wrote that I think evolutionary algorithms applied over semantic networks (rather than simple sensor & motor networks) can learn how to play hide & seek, or poker, where it wasn’t programmed in before.
It would have to learn a bunch of stuff beforehand- the concept of a “game,” and “stages of a game,” and “taking turns,” and “hiding” and “seeking” and so on. But I see no reason why it couldn’t learn (evolve explanations inside itself for) these ideas.
Though, I suspect that the growth path of the machine intelligences will be very different than the growth path of the human intelligences.
Specifically, I think we will teach these concepts to machines directly. The concepts of “game” or “stages of a game” would be gut concepts, like knowledge of how to build a web in a spider, rather than learned concepts, like they are for humans.
But I don’t think this will stop them from making up new ideas, and using them.
I think sex is a gut concept for humans, but it doesn’t mean we can’t teach ourselves how to discover math, and so on.
I suspect that a lot of concepts the AIs teach themselves, will be things that you and I know and understand. And I suspect that a lot of the concepts the AIs teach themselves, will be things that you and I will never be able to understand. [1] 1
I didn’t say that intelligence is not possible to machines, quite to the contrary, I suggest a definition that gives machines a chance.
But on the other hand I state the that AI research has failed to identify the key research targets, the general concepts of intelligence. It is similar to NASA, which wasted billions of $ in the race to the moon and will waste them again for Mars - loosing decades - instead of a reasonable step-by-step exploration of space.
I think this is on the right track. And, this definition applies to other organisms besides humans. Which is fine with me. I have also always thought that a requirement for human intelligence is “self awareness”. The awareness of one’s self and one’s mortality. It might be possible that humans are the only organisms that possess this awareness. Perhaps I am wrong about that. We may yet learn that other organisms possess this awareness. Self-awareness comes from a neuro-system that possesses enough memory and prediction to be able to reasonably model what is going on in the brain of another organism, IMO.
Ha!
My thoughts on NASA: If their core question (ThePowerOfQuestions) is “How do we explore the solar system?” … They should be studying cybernetics, full time. It’ll be trivial to send a brain into orbit around Mars: No need to send showers, far less food required, far less electricity required, just bottle up the O2 in a giant vat, and so on. Further, it would promote so many other benefits, local-side, to economies and what not. It’d also likely put a complete end to GlobalWarming?, since there wouldn’t be all these people driving cars everywhere.
This is no joke. I am not kidding.
My thinking about AI in general, is that the real essential questions are being asked (and answered) by people in the hallways, and by the water coolers. It’s detail-work that’s done in the formal research projects, and I don’t think they dare write the answers that they’re getting by the water cooler, into those papers.
I’ve talked with several acedemics who have expressed those kinds of thoughts to me; That that is a reasonable way to think about their institutions and customs.
General concepts of intelligence I think are much more likely to come from the systems theory folk, and general theorists, and cybernetics thinkers, than from the research papers. I think it’s largely MetaPhysics, and I think that the metaphysics will work, to create functional AI.
We will see an “emergence” (perhaps emergency) of AI, at some point, in the future, as a number of systems are hooked together.
I suspect that over the next few decades, we will see smarter and smarter automated systems, queued into greater and greater subtleties.
It’s also entirely plausible that the public will largely not notice. [2] (That’s been the way, mainly, so far.) Though, I suspect it won’t be quite like that- I think future studies will part of basic education, (which I suspect will be what we today call “informal,”) and that OrganizedCulture will happen. Perhaps people will prepare for it. I don’t know. We have a lot of questions coming up.
Sorry; I get into prophet mode.
Two good books that connect dots:
The futures wiki are interesting, too.
Self-awareness is simple enough, I think: There are hoards of robots that must take into account their position and shape.
The future will simply be the SemanticWeb equivalent of that.
When you are interacting with people, to properly run simulations, you have to have some concept of how they understand you.
If you go out naked, you have to understand why people are looking at you funny, and talking funny. Otherwise, you are going to be very inefficient.
Just a question - for greater clarity:
“Intelligence is the ability to solve problems that one has not been trained to solve.”
Would this exclude humans?
Because: You could argue that the rewards and punishment systems of “the world” are the trainer.
Because I’m thinking: “Hm; Neural Networks figure out things, even though we didn’t tell them the process of how to figure them out.” But there’s that word- “trainer.”
We talk about “trainers” for Neural Networks, and so maybe that’s a problem for calling them AI.
But then I think, “Now wait a second- that’s what we, us, humans, are doing, when we get ourselves into situations, and then figure out how to get out of them, or whatever.”
The only difference, it seems to me, is that our situations are much more complex. But we’ve still got the “trainers” to work against. If we didn’t have a trainer to “work against,” then there’d be no development of whatever skill, to get out of the situation.
So I think we need some greater clarity on what this means.
Define external redirect: GlobalWarming