4 ms·

I'd be careful saying that one is succeeding and another failing, at least to the point of drawing very large conclusions about the nature of intelligence. Unde

by bbctol 10y ago

I'd be careful saying that one is succeeding and another failing, at least to the point of drawing very large conclusions about the nature of intelligence. Under our current level of technology and our current mathematical models, neural networks and other squishier techniques are outperforming rule-based agents in certain fields. Basic neural networks were proposed long ago, though, and Minsky and Papert made quite a convincing argument in the 70s that the success of syntactical AI over neural networks at many important tasks was a clear sign that rule-based logic was correct! We're still so far from general intelligence that it's hard to draw conclusions.

woodandsteel 10y ago
That's a good point. It makes me think that strong AI would have to have both neural network and rule-based components. However, I think that the meta-function that decides when to use one or the other would have to be a neural network. I think that's the way it works in human beings. I think that is so because the realities that intelligence is supposed to deal with are such that they can't be fully described by any single or multiple systems of precise concepts. That was Wittgenstein's point, and also many other philosophers such as Husserl, Heidegger, Merleau-Ponty, Ricoeur, Whitehead, Putnam, Strawson, Rorty, and all the Pragmatists.
- bbctol 10y ago
  I agree that a lack of precision with respect to correspondence with the real world is inevitable. I don't think that at all mandates a neural network (and artificial neural networks are only superficially similar to human neurons, and their performance often improves by making them even less analogous.) But the idea that the world can't be fully and accurately described by a simple system of precise concepts, as much as it would pain Leibniz, doesn't necessitate that an intelligent agent couldn't use strictly rule-based logic.

by bbctol 10y ago

woodandsteel 10y ago

That's a good point. It makes me think that strong AI would have to have both neural network and rule-based components. However, I think that the meta-function that decides when to use one or the other would have to be a neural network. I think that's the way it works in human beings. I think that is so because the realities that intelligence is supposed to deal with are such that they can't be fully described by any single or multiple systems of precise concepts. That was Wittgenstein's point, and also many other philosophers such as Husserl, Heidegger, Merleau-Ponty, Ricoeur, Whitehead, Putnam, Strawson, Rorty, and all the Pragmatists.

bbctol 10y ago
I agree that a lack of precision with respect to correspondence with the real world is inevitable. I don't think that at all mandates a neural network (and artificial neural networks are only superficially similar to human neurons, and their performance often improves by making them even less analogous.) But the idea that the world can't be fully and accurately described by a simple system of precise concepts, as much as it would pain Leibniz, doesn't necessitate that an intelligent agent couldn't use strictly rule-based logic.