I still think there is much more work to be done on isolated systems and making them more adaptable than larger metasystems. I work on text classification problems and while the problem is still mostly implementation, I would like to solve the problem more robustly w/o having hand-analyze problems, construct models that are really hacks, etc. Basically work higher up the abstraction level in classification. What I am arguing is a bottom up view of AI.
This certainly sounds nice. It's easy to be a cynic though. There was a lot of theoretical and conceptual work on AI back in the day, but it didn't really lead to anything very effective. The problem of intelligent behavior was much, much more complex than anyone had realized.
This article is a very good statement that the problems are hard and complex, but just because you get a bunch of people together who want to solve a hard problem, doesn't mean they will.
One of the major problems IMHO with AI systems is that marginally working hacks work nearly as well for the most part. AI doesn't offer nearly the order of magnitude improvement over dumber systems that's been promised.
5 comments
[ 3.2 ms ] story [ 21.3 ms ] threadThis article is a very good statement that the problems are hard and complex, but just because you get a bunch of people together who want to solve a hard problem, doesn't mean they will.