3

In my head, AI attempts get sorted into two categories. I'm wondering if the distinction is present in reality or just my imagination. If this it's real, I'd like to know of any terms for them. The two types are as follows.

The first (in an arbitrary order) is mostly seen in modern times. It consist of simulating the hardware of the brain, like artificial neural networks do. This is imitating the brain on a very low level, sort of like doing a physics simulation at the atom level, when you are trying to simulate, say, a ball rolling down a hill. It works great but it doesn't provide much insight about the nature of thought. The algorithm does not have to be a direct simulation of the brain, but it will always be relatively simple, removed from any high-level representation of cognitive processes, and the complexity of it's behavior will come from the data fed into it. To put it bluntly, a "simple" learning algorithm + a ton of data.

I encountered the second type only in older writings about AI. I'd say it's a semantic approach, to use quite a loaded word. An attempt is made to describe cognitive activity, usually all of it (aiming for a general/complete AI), on a high level. Words like "problem", "frame", "context", "agent", "rule" and the sort give it away. The task is to design an algorithm for thought, or rather to algorithmically describe thinking. I have not seen this approach yielding much results. Most AI programs nowadays use the first approach, at least that's what my confirmation bias says.

There are, of course, two competing theories of the brain, parallel to these two approaches, at least on a shallow level. One is evolutionary biology, mirroring the "semantic" variety. Not so much the evolutionary part, rather the idea that the brain is composed of mostly static modules which implement various functions of thought. I don't know of a name for the opposing view, but it's the belief that brain is a very efficient learning machine, with only a couple basic, tiny, modules for learning being built-in, and the rest acquired along the way. I learned of this distionction from this article: http://lesswrong.com/lw/md2/the_brain_as_a_universal_learning_machine/.

  • 1
    You have left out optimizing and game-theoretic AIs. ANNs don't scale to a full intelligence, one must note. – Deer Hunter Jul 27 '15 at 1:40
  • @DeerHunter Why don't ANN scale to full intelligence? If I put a big enough ANN into an environment (a simulation, likely) similar to those humans inhabit, would it not mold into a human-like mind? Probably not, but why exactly not? – user2244484 Jul 27 '15 at 23:40
3

You might consider that the first is related to machine learning, but the second is symbolic Artificial Intelligence.

J.Pitrat's blog has an entry: the future of AI is the Good Old Fashioned AI explaining why symbolic AI could still be relevant.

BTW, you might be interested by Artificial General Intelligence, as a research domain.

See also this & that answers.

Notice that AI experts are currently disagreeing on what AI really is...

  • Thank you so much! The term symbolic AI is exactly what I was looking for. Thanks for all the links, too. I'll check it all out. – user2244484 Jul 26 '15 at 12:31
  • The first is sub-symbolic AI. – Ergwun Oct 18 '16 at 3:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.