True Knowledge is apparently a new service that enables you to ask questions in plain language and get direct answers. It's an obvious idealization of how Internet search should work.
So why isn't everyone doing it?
This topic is commonly discussed in early AI courses and is considered intractable (theres no efficient way to do it).
For starters, a service like this is heavily language dependent. Natural language processing is a huge problem because of all the meanings/connotations implied by context, structure, sarcasm, and other dimensions of a language. Right now, translation engines interpret language statistically, but they are far from extracting the true meaning from any given sentence. To get around this, True Knowledge engineers likely code keywords/phrases into the search and use those to invoke different subsystems to answer the question. Good luck rolling it out to other languages.
Also, the type of inference they are doing requires a huge (possibly hierarchical) knowledgebase of propositional (or higher order) logic symbols. Again this has a large dependence on language, but more importantly, it only does well with logical questions. It probably wouldn't do well in answering "What is peanut butter jelly time?"
You know, I realized I'm wasting WAY too much time on this post.
Cliff notes:
Big problem.
Limited use.
Currently requires a lot of human intervention.
Difficult to port to new languages.
blah blah, here's their video... (still pretty cool)
Tuesday, February 19, 2008
True Knowledge
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment