Wednesday, October 3, 2007

Why the machines are "stupid"

A bit of RL discourse - the human yesterday noticed an interesting article about ambiguous words by George Miller, one of the creators of the wordnet.

Interesting.

2 comments:

Melissa Yeuxdoux said...

I have to cringe at that.

The author seems to think that any word that does not refer to a unique entity is "polysemantic." By that criterion, mathematics is has infinitely polysemantic terms, like "real number" or "natural number"... but the definition of mathematical terms, aside from those that comprise the starting points of a system of mathematics, is of course quite without ambiguity. The set of triangles is perfectly well defined, despite having infinitely many elements.

Dalien said...

Well, yes and no :)

no, they are not ambiguous - because when you tell me "a real number" I understand quite easily what it means.

yes - because indeed, it is quite a hard thing to "teach" the machine.

OTOH, the math builds upon the interrelations between the abstract entities, which are defined as precisely as possible - so if we use a number (memory reference, ID, UUID, whatever) to represent each of those - we are able to encode these interrelations and make an illusion the machines "know" the math - most readily within the context of the manipulation with those entities.

but another example from "near-mathematics" area which are readily understood by humans - "very few", "a lot of", "not so many" are highly ambiguous. We have a "feeling" for them but can not express them (or do not bother, to avoid the exponential complexity?) One car accident per year is "too many", but one thousand concurrency in SL is "extremely few" :)