The idea is not new indeed, Andy Clarks's review paper was very inspiring to us. But it has not been detailed to a point of implementation/scaling. PVM is an at

The idea is not new indeed, Andy Clarks's review paper was very inspiring to us. But it has not been detailed to a point of implementation/scaling. PVM is an attempt to implement it in a "connectionist" way, but frankly all I need are associative memories, and how they are implemented I don't care. So a "probabilistic PVM" is totally feasible. In fact we discuss in the paper various possibilities in which a PVM like meta-architecture can be implemented.