Sunday, April 8, 2007

AI, Language and Symbols

AI, Language and Symbols:
However, when it comes to communicating with other agents in what we perceive to be the real world, we have created an interface that does appear to have all of these nice qualities: symbols, structure, stereotypes and so on are all used to externalize our thoughts and as an input mechanism to grasp the inner workings of our fellow beings. And while it is attractive to believe in the emergence of intelligence via huge data sets and massive but simple processing power, that intelligence will arise from the simplest machines if only we throw enough data at them - the fact of the matter is that much of what we learn as humans we do so by the consumption of structured symbols of various types.

I didn't say not do I believe that "intelligence will arise from the simplest machines if only we throw enough data at them." However, Matt's comments seem to indicate a distortedly anthropocentric view of intelligence. Anyone who reads experimentally-grounded work on animal cognition and behavior from the last couple of decades will know that our non-linguistic relatives exhibit important aspects of intelligence that are unlikely to be based on (whatever passes for) symbol-mediated computation. Here a few highlights from my reading:

  • The Number Sense, Stanislas Dehane
  • Primates and Philosophers, Frans de Waal
  • Toward an Evolutionary Biology of Language, Philip Lieberman

As for "much of what we learn as humans we do so by the consumption of structured symbols of various types," have you counted? We don't know much about the evolutionary emergence of language except for some very tentative and controversial dates, and we certainly do not know which if any critical mutations might have been associated with it. It is at least worth considering that what Matt calls "human" is really "post-writing settled division-of-labor-based." The low reproduction error rate and long-term durability of writing relative to the evanescence of speech and human memory have played an important role in the “symbolization” of our culture. Long arguments and proofs need external memory. Euclid's “Elements” was a book, and Leibniz's formalism follows the explosive development of the printing press. We know much less about what “intelligence” was before writing, let along before speech. But we do know that “educated” people, even today, have difficulty in attributing equal intelligence to pre-literate cultures or illiterate individuals. That is, Matt's “intelligence” is a loaded term in this debate. It is (Humpty Dumpty comes to mind) what his literate and formalistic assumptions say it is rather than what experimental, operational evidence might indicate.

But how did we get from the genomic data - represented as simple sequence - to the problem of finding patterns in it? That requires all the symbolic, hierarchical structured knowledge: the genetic model.

This is the old standard canard against artificial intelligence: it was merely implanted by the programmer. My point is that the hypotheses generated by these programs were not anticipated by their creators. In science, the highest flattery is “I wish I had thought of that.’ These programs “thought” of hypotheses that their creators and others never considered. The “genetic model” is as determinant of the predictions as general relativity and quantum mechanics are determinant of the distribution of mass and energy in our current universe. Maybe totally in an idealized Laplacian universe, but irrelevantly so from the point of view of scientific discovery.

In (partial) answer to Fernando's question - clearly the parallelism of the brain is considerable, but that is not the type of scale that Larry Page is talking about (that is to say, the symbols - or units/mechanism of representation - and operations involved are quite different).

That's another assertion that needs evidence, isn't it? Matt and I can speculate endlessly of what is significant or insignificant in alleged differences between representations and computational models, but the reality is that neither he nor I know nearly enough to decide the issue. Neither does anyone else as far as I know. I repeat that I prefer experimental evidence. And the experimental evidence is on my side, in the sense that almost all the practical progress in artificial intelligence over the last twenty years has been based on improvements in methods and tools for extracting generalizations from data.

No comments: