Slashdot Mirror
Going from a 'Web of links' to a 'Web of meaning'
neutron_p writes "Computer scientists from Lehigh University are building the Semantic Web, which will handle more data, resolve contradictions and draw inferences from users' queries. The new improved Web will also combine pieces of information from multiple sites in order to find answers to questions."
Sounds like a recipe for disaster to me.
Does everything include nothing?
Covered not long ago - an interview with Berners-Lee regarding the Semantic Web.
People at DERI in Ireland's Galway are also working on the Semantic Web (see http://www.deri.ie/). I thought lots of people are...
I'll have to rtfa to see what they propose, but just the principle of resolving contradictions is a really difficult one, and most theories of knowledge (which are essentially networks of facts) aren't terribly robust, and contradiction repair, which involves running the entire network to find invalid assumptions, and then propigating the changes is NP complete :| i'm not positive that contradiction resolution is a reasonable thing to expect out of a massive distributed network.
There are lives at stake here!
You gotta understand that "meaning" has no meaning at all to machines, at least not yet.
And even for humans, the "meaning" of a certain thing can be different thing to different people !
Although I applaud the job they are doing for Semantic Web, I wonder how they can inject "meaning" into the whole thing.
My biggest fear is the 1984-like "my meaning is THE meaning and you canna have any other meaning" thing.
Muchas Gracias, Señor Edward Snowden !
..from user's queries.
Clippy..? Is that you?
There's a Starman, waiting in the sky / He'd like to come and meet us, but he hasn't got the time.
The same can be said about any semantic web technology - whether it's FOAF (an RDF vocab for describing people and their interests) or a vocabulary for reviews. As soon as major authoring tools (i.e. both web editors and content management systems) start integrating these technologies, people will use them if they are useful. Do not expect web designers or bloggers to have a clue about all the great things that the semantic web can do - give them one useful thing which they understand, package it in a pretty UI, and they will start using it.
The semantic web is a pretty popular area of research right now and its far from being "built by computer scientists at Lehigh University", in fact I could have done an undergrad dissertation on the semantic web, and there were numerous phD positions being advertised at uni's around the world researching about the semantic web.
Whichever lehigh uni professor submitted this is stooping pretty low trying to raise publicity (and hence finance) I would think!
I have discovered a truly remarkable sig which this post is too small to contain.
It seems to be a common mistake for computer scientists to think that it's possible to make systems that "understand" the world (both real and abstract knowledge), with all its complexity and ambiguity, in the same way that humans do. I feel that there is a fundamental difference between using computers to enable humans to organize stuff, and having computers automatically do it. Every single attempt at getting computers to be "smart" about infering human intentions has ended up as an irritating impediment to using the system - look at clippy, Bob, "intelligent" voice systems that try to "help" you by stopping you from talking to a real person... what computers are very, very good at is amplifying and enabling human intelligence. Computers are not themselves intelligent, and (my personal opinion) I don't think they ever will be - unless we manage to "grow" them using processes that we probably won't fully understand. You can't construct something that is as complex as the human mind through deterministic (i.e. consciously designed architectural) means - all you'll end up with, at best, is a very complex rule inference engine that is limited by the rules you gave it. Every "holy grail" of intelligent programming that has come along - neural nets, genetic programming etc - has turned out to be very limited (though very useful in special situations).
I also feel that talking about automatically organizing the world's knowledge in a semantic web is just more of the same hot air that we've been hearing from AI departments for the last few decades. You can't automatically allocate meaning to something unless you have the capability for "common sense" reasoning, and the world knowledge at your fingertips to be able to interpret the data intelligently, like a human would. And even then, different humans would interpret it differently... so there are multiple meanings, and anyway, how to allocate "meaning" to something abstract such as a poem or piece of art?
And if we require real people to add metadata to everything... well, it just ain't going to happen, in my humble opinion. Adding meta data is a pain in the ass, since you have to define the categories of object, agree on meanings for all the different taxonomies that will have to be used to describe the world... then there's the potential for abuse, as spammers will inevitably seed their documents with inappropriate metadata. So, the "honest" people can't be bothered, and the dishonest people will wreck anything that does get built. So, it ain't gonna happen.
The beauty of google (not that I love google, but they did hit a nail on the head) is that it requires no effort or "machine intelligence", beyond a very simple algorithm that depends not on AI but rather real, tangible relationships between words and documents (proximity and links). This is something that computers can be really good at.
Just my opinion... obviously there will be others out there who will vehemently disagree, and that's fine! Go ahead and try, you'll learn a lot in the process and you will probably come out with some tangential technology that you never thought of initially but is useful nonetheless.
A message has "meaning" if you can make special use of it.
Normal web pages have meaning for browsers, it's just that that meaning is limited to "how to draw words for the user."
What we're doing, is making it so that your computer can make special use of messages on the web, to do smarter things.
It would be scary if the Semantic Web were about "my meaning is THE meaning." But it is explicitely not like that. In fact, one of the main things about it is that anyone can make up their own languages, their own way of modelling the world.
There are tools that make it so you can say, "My word X is sort of like their word Y," but it's acknowledged that such translations will be imperfect. Likely, fuzzy logic, and systems that are able to ask for clarification (and remember responses), will be used to mediate that sort of things.
You may also be interested in my favorite page on AI by Open Mind. The Semantic Web isn't explicitely about AI, but it opens the door for a lot of AI work.
Semantic Web is the most ridiculous idea I've ever heard. The problem with meaning isn't representation -- English represents meaning just fine. The problem is meaning itself -- it doesn't matter if you figure out a way to encode it in some XML language, for every bit that it's easier for computers to use, it will carry that much less meaning.
Another way of putting it is, any program capable of extracting the same meaning from XML that humans can, should be able to understand English without much trouble. It's the whole Intelligence-complete" thing. Like NP-complete, there seem to be a class of problems which can only be solved by real intelligence, and they're all pretty much equivalent in that with real intelligence, you can solve them all.
"If you look 'round the table and can't tell who the sucker is, it's you." -- Quiz Show
Google works because it is largely a statistical tool that uses some meta-information.
While I could see frameworks being used for very specific purposes, like searching a homogeneous (e.g., slashdot, pubmed, nytimes) web-site where all content is controlled. But extending these ideas to a heterogenous web that would no doubt take advantages of such a volunteer system is ludicrous.
I also take issue with the top-down mind-state that they will be able to predict what is useful to the user. This is why statistical importance and quantity is the only realistic method for such a massive undertaking (which google is still actively researching).
I think that the only useful research to come out of such an endeavor would be to have news-sites, as mentioned above, implement and be scanned using an ontological browser. Of course, I am not sure how this would be different than Lexus-Nexus (sp?).