I just read a new paper by Jeff Shrager (currently at CommerceNet, which funded my work on Zocalo in 2005 and 2006; thanks to Tom Malone of MIT's Sloan School for the pointer to the paper) on a system designed to support group reasoning processes based on Bayesian principles. Having just read Eliezer Yudkowsky's post on Artificial Addition, I was sensitive to the notion that some older attempts at AI had failed to enforce any semantic relationship between the nodes that are intended to represent particular concepts.
The system described by Shrager et. al. uses a language understanding system (presumably based on the same tools as PowerSet relies on; I've been admitted to PowerSet's trials, but can't access the activation code, so I haven't tried their demo yet). This means that for the first time to my awareness, the system can help by finding semantic connections between assertions. This isn't trying to be AI, but computer support for collaborative reasoning is just as much on the path to enhancing our power to change the world as is AI.