Cognitive AI by specialization of several functionalities? It might work, still, from what we know of human minds, simple goal directed inference must fail since it is neither very robust or adaptable to open-world situations. In highly unintellectual terms, this model lacks the 'white noise' needed to reach goals by a process of goal oriented distributed chaos. I wonder how a mixed approach of partially organized neural nets might work.
Excellent work by Singh and Minsky.
