What they are promising to achieve is simply impossible to do. It assumes you could do logical and mathematical inferrence at the level of a Ph.D student...
... oh wait. I'm actually the one who wrote a review of Mathematica for MacWorld 15 years ago and said that it can reason to a level as good as a Ph.D student.
My opinion: there is about one person on earth who can pull this off, and this is the team. So I have great expectations.
I would argue that the hard problem is not "reasoning" (taking absolute truths and deriving other truths); but rather dealing with noise.
The former problem -- logic -- ended up giving us the AI winter, and it's the later (with machine learning) that is driving Google and modern robotics.
If this is just Mathematica + Prolog + A giant hand tuned database, I think it'll be a disappointment. (Look at the Cyc project -- led by a former Stanford professor, lots of funding, supposed to revolutionize human knowledge -- where is it now?)
