When we built Enginuity Search Engine to better understand and contextualize information the Bayesian approach simply does not work. We ended up building a different (and we think better) approach which works very well and used by top schools like Harvard.
I'll be interested to see how he gets on. I had a read of his book and it's a tricky to see if that hidden hierarchical markov models stuff is going to the answer in AI or not.