New Directions in Bayesian AI

A colleague points me a May 3 NYT article: Pursuing the Next Level of Artificial Intelligence, which describes some of the work of Daphne Koller at Stanford. Specifically Bayesian techniques that introduce more quantification and uncertainty in artificial intelligence models. The only thing I disagree with in the article is it's saying that in the original AI methods used in the 80s practitioners avoided quantification. This is not true, methods we used did address both quantification and uncertainty. They were often hung on a base of logical constructs. Now the reverse seems to be more popular. I like to see that the article says there is a revival occurring in AI, though I have not seen much of it in industry. The work being done by Koller and others should improve the tool set available to solve problems. Here is more on her research.