Unsticking what's stuck...in business and beyond

The Semantic Argument Web

What really scares me.

By David Weinberger

June 14, 2002 — Tim Berners-Lee (TBL) is frustrated by his creation. It's turned out to be a good way for humans to post and read stuff. But we're overwhelmed by the humanness of it, its randomness, its capriciousness. We need help from our machines to make sense of what's out there, to get answers to our questions. And the way to do that -- or so says TBL -- is to put more smarts into the data.

It's interesting that TBL rules out the other obvious approach: make the machines smarter. But in his major work on the topic of the Semantic Web, he writes:

Leaving aside the artificial intelligence problem of training machines to behave like people, the Semantic Web approach instead develops languages for expressing information in a machine processable form.

Note the "instead" in that sentence. This doesn't mean that TBL thinks AI is pointless or evil. But it does clearly mean that the Semantic Web doesn't take the AI route.

And I'm glad to hear it. But as you read more of TBL's "roadmap," you find that he's drawn into one of the stickiest of the AI morasses: "knowledge representation." (And, here's an important note: Much of the discussion of the Semantic Web is over my head and is being conducted by certified geniuses who are much more likely to be right than I am.)

To make the data smarter, you have to put in metadata. TBL prefers RDF as the standard for representing metadata, but the tougher questions are: What metadata should be recorded, and how should it be tagged? So, even if we agree that we want to tag pages with, say, the last name of the people who created them, is the attribute labeled "lastname", "last_name", "name_last", "surname" or "nom_dernier"? If we can't agree on that, the machines aren't going to be able to help us. In fact, we have to agree not only on particular attribute names but on entire schema for categories of data: contact information, automotive parts lists, medical forms, Buffy fan pages, etc. The number of potential schema is huge. We are, in effect, back in the AI game of attempting to classify all knowledge. See Cyc. See Cyc not run.

I fear that the Semantic Web will go the way of SGML and for basically the same reason: normalization of metadata works real well in confined applications where the payoff is high, control is centralized and discipline can be enforced. In other words: not the Web.

I have long thought, however, that we contributors of Web pages could be quickly enticed into adding metadata that would be a lot better than nothing if the search engines were to announce that they're going to allow people to search on the following meta tags: author's name, language, self-rating, best if used by date, country of origin, Yahoo/DMOZ category, etc. Google ought to come up with a handful of them -- the Dublin Core offers a starting point -- and just do it. That'd quickly add enough semantics to the Web to help us search a hell of a lot better without having to have a global food fight over "copyright_holder" vs. "holder_of_copyright."

What scares you? Metadata? Artificial intelligence? The use of the word “semantics” in a title? Post your comments below.

	Discuss this article

June 14, 2002