I agree that "determinism" has much to do with the questions here, but to immediately pose a question about "the world" being deterministic might be putting the

I agree that "determinism" has much to do with the questions here, but to immediately pose a question about "the world" being deterministic might be putting the carriage before the horse. I think the the other, perhaps more prescient point here, is about whether or not that kind of exponentially-learning AGI being discussed is even possible given that the substrate such things would be embedded in (hardware, software) which are [or appear to be?] deterministic. One assumes that that kind of learning implies learning about or aquiring capabilites which have not already been determined or specified by a human. At some level of abstraction, humans _must_ determine that system. It is possible to concieve of a machine that is non-deterministic in that way? At what point does such a system actually leap beyond what has been specified by us? Does this imply a machine that not only rewrites itsel in softwate, but also remanufactures itself in hardware?