September 19, 2002
Bloated Incomprehensible Catastrophe
That's a description of my last social outing involving women.
No, actually, it's from this article by Bradley Sherman, which pretty much sums up my thoughts on the Semantic Web and RDF [via Jorn Barger]:
Jorn Barger <email@example.com> wrote:
>So I don't believe TimBL&co are going to be able to define any
>useful semantic labels for the bulk of webpage markup. (If they
>could, it would be more useful in a META header than in phony
This is an understatement. There are thousands of talented
people attempting to categorize knowlege and they are
*all* failing. XML is a disaster. Ontologies are a disaster.
XML plus Ontologies is synergistic: the combination is
a bloated incomprehensible catastrophe. The Semantic Web
is a red herring.
In my experience, even small ontologies with a lot of domain specifity are hard to get right. And small changes in the domain seem to often force significant restructuring of the ontology.
I've always been suspicious of Semantic Web people because it seems like I've never heard one talk about just how hard this is. Or about all the previous failed attempts at it that lots of AI researchers have made (at this point, you're thinking about Cyc, aren't you? Admit it. Well, is there any evidence that Cyc has been successful?)
Posted by jjwiseman at September 19, 2002 02:02 PM
Ok, I know smarter people than me are working on this but I still feel the need to pontificate.
I also don't think you can encode all semantics using only syntax, which, as I understand it, is the purpose behind the Semantic Web stuff. The reason is that Godel proved in the 1930s that there are true statements that can be expressed in any system than can not be proved true within that system. Sure you can go further than people have so far, but you'll never be able to get all the way, and you're likely to end up with something as incomprehensible as Principia Mathematica by Whitehead and Russell.
That's one objection where I'll side with SemWeb people. That there are theoretical limitations in the system doesn't mean there's not a whole lot of useful work that can be done. I mean, people are still earning degrees in math, right?
Yep. But you sometimes get the impression, and perhaps this is just throught the business exploitation side of this stuff, that all problems with integrating systems will soon be solved by XML Schema or some other scheme that has just been announced.
I agree that there is lots of useful work that can be done. Standardization of some of this stuff would greatly help me at work. However, it's sometimes presented as the solution to all our integration problems and I'm just getting pretty fed up with it.
I know a semantic web person who used to work at Cyc. Not only did he never admit that anything was hard -- he was notorious for going around saying things like "I could do that in 2 lines of code!"
Tim told me once that the kinds of people who are interested in these kinds of problems are going to also be the kinds of people who have little or no grip on reality, because anyone who had even a glancing acquaintance with reality would be able to tell how hard the problems were and would get discouraged and give up.
People are still earning math degrees because it was proven that a simple algorithmic system can't prove the truth of all true mathematical statements. :)
Yes, the mathematicians got off reaaaally lucky on that one.
Troutgirl, yours are exactly the sort of comments I like to see. Insider gossip and grounded, pithy statements.
I like what Francis Crick said about this (in "What Mad Pursuit"):
"It is interesting to note the curious mental attitude of scientists working on 'hopeless' subjects. Contrary to what one might at first expect, they are all buoyed up by irrepressible optimism. I believe there is a simple explanation for this. Anyone without such optimism simply leaves the field and takes up another line of work. Only the optimists remain. So one has the curious phenomenon that workers in subjects in which the prize is big but the prospects of success very small always appear very optimistic. And this in spite of the fact that, although plenty appears to be going on, they never seem to get appreciably nearer their goal."
This sounds all applicable to the speech recognition field: lots of optimism, XML is the saviour, and Microsoft in the boxing riing as well.
I think there actually is reason to be optimistic about speech recognition; progress seems to have been slow though steady for many years, and it's a problem that benefits from having the fastest processors available (unlike AI, for example).
If you're talking about dialogue management stuff like VoiceXML, that's different. I don't think it currently aims high enough in its goals for anyone to worry too much about unjustified optimism.