It's Like Déjà Vu All Over Again
"You could probably waste an entire day on the preceding links alone. But why take chances? We also give you Paul Snively..." — John Wiseman, lemonodor
Yet each of us "know" apples in a slightly different way. Most of the time, the way we each know something is so much alike, that we communicate well enough to accomplish our goal. Every now and then we have a communication breakdown, and so we try to fall back to a less powerful, but more common level of understanding. Then we build from there.
The challenge with a "semantic web" is to build powerful *and* common levels of "understanding". In this case, "understanding" means "same set of definitions for the symbols being exchanged". Each system's symbols have to be grounded in the same definitions. (And any of the systems may be human or non-human.)
Even more challenging is for a system to recognize a "communication breakdown" (i.e. the "symbol grounding problem") and then to repair the problem well enough to accomplish the goal.
Interesting comments. However I don't think it's a really big challenge to have this common level of "understanding". That is because the symbols (or the "things" I talked about) of the semantic web are URI's. This means that anyone can introduce symbols.
Which, at first blush, to me makes it sound like we'll end up with an enormous set of symbols that are distinct, but mean the same, or at least very similar, things. It's a better argument against commonality emerging than in favor of it.
What I think what will happen is that companies and individuals will introduce task specific symbols. Simply because it makes work much easier.
Aggregation is indeed helpful if only by dramatically narrowing the domain that one has to model, but the risk is of creating artificial archipelagos.
Then this company or individual, or maybe someone else, can provide a grounding (transformation) into a set of symbols that is in common use within the appropriate community. And a smart person within that community may provide a transformation to symbols from another community.
The point is that a commonly-grounded symbol doesn't need transformation, and you rarely really get grounding out of a transformation process because "something gets lost in the translation." Having said that, sometimes (most of the time?) what has to happen is that transformation is done, and over use and time, one party determines that the other symbol works better for them, and eventually a consensus is reached on a single symbol, whether it's one of the initial symbols, or a new one that somehow amalgamates the initial two.
But just like spoken languages, there's not reason to have a basic level. Whenever there's a transformation from one symbol to another, or several others, someone will provide it. And when there is no transformation of symbol A into language B, there's no reason to use symbol A in language B. Spoken languages regularly borrow words from other languages too.
Sure, but once again, "Mensch" in German does not have the same connotations as "Mensch" in Yiddish. There's much more to words than the words.
I see more challenge in building the global system that takes care of all those tranformations. The most logic systems I've seen assume total knowledge. There has to be infrastructure to acquire required knowledge. Anyone has some docs?
No worthwhile knowledge representation system in the past decade plus has made the closed-world assumption; it's been too clear that reasoning under uncertainty is the goal for too long. Keep your eye on OpenCyc and check out PowerLOOM. And the best-founded system I'm aware of is OSCAR, but sadly, OSCAR is held hostage by patents.
The best available text on AI is still Russell and Norvig's "Artificial Intelligence: A Modern Approach", which deals with Decision Theory extremely well, and also with machine learning topics. A must have for anyone who's even slightly curious, as it manages to be both definitive and highly approachable at the same time, a combination of qualities so rare as to induce anaphylactic shock when encountered as they are here.
7:15:16 AM
