- From: Drew McDermott <drew.mcdermott@yale.edu>
- Date: Tue, 21 Aug 2001 11:12:45 -0400 (EDT)
- To: www-rdf-logic@w3.org
[<Patrick.Stickler@nokia.com>] > > The core mechanisms of RDF *must* preserve the integrity of all data. > [Tom Passin] Aha! I'm going to strongly disagree with you here. One of the features of the current web is that it is not self-consistent, .... ... There is no way the SW is going to be any more self-consistent or stable than what we have today. Now, what's the difference between inconsistent or changing data, and mechanisms that don't "preserve the integrity of all data"? Nothing, really, it's just a matter at what point inconsistencies creep in. In either case, our systems are going to have to deal with it. At the risk of starting (or restarting) a discussion on an issue too fuzzy ever to be resolved --- I don't like the tactic of refuting every argument by saying, "Oh, well, the SW is going to be inconsistent anyway." I have several reasons: 1) Some of the same people who say this also say the SW will provide formal proofs of statements such as "You owe me 3000 Deutschmarks." I don't see how this is possible in an inconsistent system. (Rather, I don't see why anyone should pay any heed to formal proofs in an inconsistent system.) 2) When inconsistencies arise (as I agree they will), it will usually be because two or more datasets from different sources are combined. The combining agent gets a little too trusting, and then realizes it has an inconsistency on its hands. Exactly how it deals with it is a difficult issue, but it always has the fallback position of backing out of its tentative commitment to this collection of datasets. 3) Note that, as Patrick implied, this sort of inconsistency is very different from what happens when I take a *single* dataset and misconstrue it. That ought to be avoidable --- a program's reach should not exceed its grasp, else what's metadata for? 4) The whole mindset of those who play the inconsistency card is too anthropomorphic for me. I believe they are thinking along the following lines: People can cope with inconsistency, so our programs should be able to as well, not like those "brittle" formal systems. To quote from Tom's posting (italics mine): The web is extensible without central repositories or contracts in large part because it isn't required to be self-consistent. But *we* learn to deal with it anyway. After all, no two *people* have exactly the same definitions of or connotations for any word, yet somehow *we* communicate and get things done. It will have to be like that with the SW, I imagine. This style of reasoning is a perennial blind alley in AI, especially, but not only, among novices. The problem with it is that you don't get anywhere by observing what people do; you get somewhere by proposing an algorithm for a task. I don't see many inconsistent-SW fans proposing any algorithms. -- Drew McDermott
Received on Tuesday, 21 August 2001 11:12:46 UTC