semantic web

[gusl]

This essay seems to argue that the dream of the semantic web is unrealistic.

It would be wise of me to ponder their points very seriously, and perhaps stop dreaming on these unfeasible dreams.

Comments

[jcreed.livejournal.com]

It is definitely an interesting essay - I have to give it some more time and thought, too. I've been habitually skeptical of classical AI, and somewhat so of the Semantic Web crowd, but I also don't really enjoy "Worse is Better" as a normative claim about how to design software and protocols and formats, though I do accept it as an observation about which things have tended to get accepted. I would love for the internet (and other things, like programs) to become less merely textual and more structured, and so I feel a little challenged by Shirky's implied belief that "adding meta-data" is a different kind of thing from "making the Semantic Web" as opposed to baby steps towards it.

[ikeepaleopard]

I actually think that particular criticism of the semantic web is misguided. I think it is silly for other reasons. The author seems to think its dumb because the people who are talking about it can't explain the point and the point they do explain is dumb. I could imagine several applications facilitated by having more semantic information available on the internet. The real problem is that no one will ever care enough to actually generate good semantic information, which means it will have to be reconstructed automatically, which means its what we have now.

[ikeepaleopard]

Oh actually, he makes that point too, I just didn't read far enough.

[easwaran.livejournal.com]

After reading this, the idea of the semantic web strikes me as much more like the "unity of science" and logical positivist stuff going on in the 1920s - following Frege, people (notably starting with Russell) thought that we should find an ideal language for science and translate everything into that language, and then things would develop nicely from there, once we got rid of the inefficiencies of natural language. Tarski was also thinking that way when he gave the first semantics that even approaches working for natural languages, except he thought language should be made to fit his semantics, rather than vice versa.

However, in the last 40 years or so, with the growth of linguistics (and especially since the work of Richard Montague) people have taken those old semantic frameworks and started seeing how they could make sense of actual language. I think many of the criticisms Shirky makes early on would be avoided if we could figure out the actual semantics of natural language. Some of those problems he mentions (like the case of generics, "people in Brooklyn speak with a Brooklyn accent" or "people in France speak French") are notoriously hard ones though, so of course not very much useful deductive inference can be done from them. But that problem is fixable.

I think he's right though towards the end when he says that the problem isn't to make people use some ideal standard in marking up their pages; rather we should make the machines better understand what people are naturally doing in marking up their pages. Some amount of standardization will be helpful of course (the way traditional search engines have taught people how to use keyword search) but the way he describes the semantic web project seems to assume that much more of this will happen than is actually plausible.

[williamallthing.livejournal.com]

I have loved this article ever since I first read it. Yes, the semantic web is unrealistic. It's ultimately a combination of some very naive ideas about semantics and a general misunderstanding of which things will be easy and which will be difficult.

Shirky makes two points very well:

1. Syllogisms are useless, because chaining them (the fundamental operation of the semantic web) produces garbage.

2. Even if they were useful, the actual hard parts---merging ontologies, and getting people to generate this stuff in the first place---are glossed over.

I will go further and make these points:

1. Regardless of whether they're expressed as syllogisms or not, the semantic web won't work because it assumes we can capture semantics by writing down a bunch of symbols that looks like words, and some binary relations between them. Yeah. That was the big idea of 1950's AI. It hasn't really gotten us anywhere since.

2. The web works because people are motivated to create content. They see the results immediately. No one is going to generate metadata for some vague process that may or may not work in the future, if enough other people have done it. There needs to be some incremental path, and the semantic web doesn't really provide one. (Shirky does make this point in later essays. He calls it the fallacy of "this will work because it would be good if it did.")

[gustavolacerda.livejournal.com]

I forgot to give you credit for the link.