On Fri, Nov 12, 2021 at 6:27 AM John Rose <[email protected]> wrote:
> ... > > While these examples may sound edgy, often these incompleteness's are > where there is much to be learned. Exploring may help some understandings > especially, as James pointed out, that “AIXI = AIT⊗SDT = Algorithmic > Information Theory ⊗ Sequential Decision Theory". > AIXI *reduced* the parameter count of an AGI with unlimited computation but limited information. Before you jump all over the fact that it is necessary to limit the computation, we still need to talk about the remaining open parameters in AIXI. In AIT the open parameter is: "Which Turing Machine?" In SDT the open parameter is "Which Utility Function?" To answer "Which Turing Machine?" I've intuited an approach that Matt reduced to a pretty restrictive descriptive space of NOR DCGs. This reduces what might be thought of as the descriptive space of Turing Machines to what Matt formalized. It doesn't get rid of _all_ of the unknowns in that space, but it is far more rigorous than the descriptive space of all UTMs. There is a _lot_ of work to be done with this approach and advances will, IMNSHO, have immediate and profound application in logic design. To answer "Which Utility Function?" we must become a lot more philosophically serious than has heretofore been the case in all the brouhaha about "friendly AI". Hutter's paper "A Complete Theory of Everything (will be subjective) <https://arxiv.org/abs/0912.5434>" is his (still incomplete) approach to addressing what you refer to as "these incompleteness's". Now, having said all that: Yes, the measurement level of abstraction does get into the economics of computation resources and, yes, it would be nice to find approaches that obviate all of the above "incompletenesses", but you must do better than to redefine the words "lossy" and "lossless" compression as that merely hobbles an existing approach to these incompletenesses while at the same time threatening to hobble their practical applications by confusing the meanings of words. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T5ff6237e11d945fb-M8e52f64cbae47e56fc8a569a Delivery options: https://agi.topicbox.com/groups/agi/subscription
