On Wed, Apr 3, 2013 at 6:43 PM, Jim Bromer <[email protected]> wrote:
> On Wed, Apr 3, 2013 at 7:31 AM, Andrew G. Babian <[email protected]>wrote: > >> ... it's very clear to me that language cannot be the bottom or basis of >> representation. A language system has to be a piece on top of the basic >> system. It may be the most important piece to us, because for interaction >> with us, and ability to use our body of written knowledge and contribute to >> it, a system will need to use language. But, that need in no way implies >> that you could ever get any intelligent behavior if you just start at the >> level of language. >> > > >> Our computers presently use symbolic language to do all sorts of > intelligent things and as time passes they can do more and more. Are you > saying that language cannot be the basis of representation for AGI? > > Isn't the internet a lovely place to repeat oneself! As I remarked in other ramblings, language is brilliant for most things and should be very close to the bottom of things - we do not unfairly or irresponsibly talk of a film's or composer's or painter's "vocabulary". Indeed, any reasonable degree of mastery over any domain comes with its own linguistics and we shouldn't shy away from the fact that ordinary human languages represent a victory of the species, a considerable mastery over space-time, psychology, biology and zoology at least. In all these domains- and more - considerable magic happens between the "cracks" of reality, the gaps and overlaps of symbols, rules, interactions etc. If it's dancing we are talking about then performing flawless jumps and turns and drops is not enough, the flow from one to another has to be managed as well, the rhythm etc. When trying to learn by reading/listening, your mileage will vary depending on the ability to ask questions and suggest rephrasings that will appropriately generalize or specialize the discourse. Of course I am a big fan of embodiment where the last bit of disambiguation happens, the reality check! Now, what would be the meaning of "starting at the level of language", would it be some kind of PureEnglish in which the state-of-the-world would be described, and then continuously refined with observations and experiments? Let's tie this down a bit, it would have to be the state-of-the-microworld, the state of the 10, 50, 1000 entities and 10 people I know well, right? And it would have to be probabilistic because it would not be terribly intelligent to assume I know what state of mind my long-lost friend is in, and it would be very naive to believe that the bicycle I left in a corner some days ago will be in the state I left it for much longer. So, immediately I am given an opportunity to either encode the world in ProbabilisticEnglish or perhaps in PureEnglish with probabilities added on top. Is this so impossibly clear and clearly impossible? I find it a very plausible way to go about things. Some other representations that come to mind like a sparse 3d or 4d matrix that holds information about the materials in that space (or the molecules, or the superstrings) sound terribly incompressible and uncomputable. OK, maybe you can do smarter things but language sounds terribly smart already. Now, I will try to recap my position: a learning or intelligent system is anything and everything but -- a human like intellect will need decent modelling of the human mind, mamallian locomotion and everyday objects and physics, and most importantly the transitions of these things. As much as I am not a fan of OO programming, I don't see reasoning happening without objects, object histories and person histories(and personalities of course). Ontologies should be built around these objects for extra productivity, and everything has to be kept somewhat flexible, whether fairies exist or not we know not, same for gravitons, our sensors will not have unlimited reliability and our interaction with sentient beings will include the systemic risks of lies and inaccuracies. Of course all this flexibility adds problem-space dimensionality and the real point of intelligence is to navigate without ending up in loops and dead ends. But will it be a problem if object histories are described in an English subset? I don't think so. It would be more of a problem if the proto-AGI was misbehaving and we lacked linguistic clues to debug it, for example if it was contemplating the prisoner's dilemma and fell into infinite loops. For all I know the various "failed" linguistic systems never tried to enter this high-dimensional probabilistic world - what is the point of story understanding (or CYC) if you can't factor in the probability that the opening sentence in a book is actually a spoon talking, especially if you've paid an engineer to code logical rules like "only humans talk". So, in my mind the problem was not language but dimensionality, more specifically dimensionality avoidance! AT ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
