> > Perhaps we **MUST** be as complex as we are, just to be able to gather > enough data to reduce the possibility-space down to a tractable size?!!! > This could bode poorly for small implementations.
We evolved gradually from a lower intelligence level. That means there are functional intermediate steps. On Thu, Dec 19, 2013 at 3:43 PM, Steve Richfield <[email protected]>wrote: > AT, > > I suspect that there is some sort of new logic at work in Google, machine > vision and other areas of pre-AGI, specifically... > > Google FINALLY made their search engine look for all synonyms and > variations of each word (unless you surround it with "quotes"). Machine > vision often becomes much more useful when a "recognition" is redefined as > an INability to eliminate the prospect of an object being present. This > handily solves the partial obscurement challenges, unusual orientations, > etc. > > I suspect that AGI-logic will become the process of manipulating > POSSIbilities rather than manipulating Bayesian PROBAbilities. There is > 100% probability of whatever is present being there, and 0% probability of > objects that are not present being there. Only in VERY familiar > circumstances are Chi Square computed conditional probabilities, Bayesian > computations, etc. worth anything at all. > > Then, the process of learning becomes the discovery of which possibilities > *might* be relevant, and which possibilities are clearly irrelevant. This > has the possibility of sidestepping the "probabilities of probabilities of > probabilities" conundrum, where you are trying to manipulate probabilities > through a haphazard process that itself is full of probabilities of being > grossly wrong. This has apparently sunk the implementation of practical > learning of Bayesian computations. > > When you have eliminated the impossible, whatever remains, *however > improbable*, must be the truth. > Sherlock Holmes > > This approach runs into problems when there is insufficient information to > eliminate multiple "truths" **AND** those truths lead to different actions > that have different-valued outcomes. Perhaps we **MUST** be as complex as > we are, just to be able to gather enough data to reduce the > possibility-space down to a tractable size?!!! This could bode poorly for > small implementations. > > Any thoughts? > > Steve > ======================= > On Wed, Dec 18, 2013 at 1:54 PM, Anastasios Tsiolakidis < > [email protected]> wrote: > >> I am in favour of all kinds of Products on top of the current AGI >> codebases, as opposed to Ben who has OKed only some kinds of Products. But >> I see no place for a Minimum Viable Product as scalability is they key >> unknown of search space algorithms, if I may call it that. Perhaps Siri is >> a brilliant NLP assistant or Siri 3 will be a brilliant NLP assistant, >> mastering 1000 words, but as I have pointed out before, by no means being >> the first one to say so, full-blown language depends on a constant >> goertzelification, constantly redefining words and occasionally creating >> new ones, trying to produce a tight fitting film over a much more expansive >> reality, a bit like the rubber case for your smartphone. Reaching out to >> the millions of words and meanings and uses and deciding which ones will do >> the trick or which ones to modify to do the trick, well, even in terms of >> parallel programming and complexity metrics is a bit of a nightmare. I do >> believe however that the products will show the way, just like I expect >> robotics with their cumbersome dancing routines and other human-inspired >> but limited and primitive repertoires to slowly but inexorably advance >> towards MVP-like states and beyond. Especially by focusing on the >> *transitions* between repertories or domains etc, for example imagine how >> well a chatbot would do if it could appropriately and seamlessly transition >> between certain dialogue modes like "me", "wikipedia", "joke", >> "arithmetic", "common sense" - it would probably pass the Turing test >> already. >> >> I was going to jump the gun and say I know what is not an MVP for AGI, >> "machine vision would not be an MVP", but then I had to remind myself that >> in my very own analysis a human level vision system would probably need to >> ask itself questions such as "could it be that there is an irregularly >> shaped black and white table partially obstructing a black and white cat?" >> in order to do visual object recognition. I cannot prove that the >> compositionality inherent in machine vision is of the same order as the >> compositionality/productivity of natural language or that of "physical >> existence" (meaning the compositionality offered by the material world >> itself once you actually interact with it, with its tremendous potential >> for emergence/surprise, such as getting an electric shock by touching the >> millionth piece of metal you encountered while all the previous touches >> gave no such shock). I think it would be OK if machine vision is an order >> magnitude below the complexity of language too, it would still be an MVP >> and quite a formidable one! >> >> Although my terminology is a bit more standard than the acrobatics of our >> usual suspects on this list, I am perfectly aware the intuitions will not >> be clear to the majority of AGIers, and certainly they are intuitions that >> relate weakly to a lot of academic but "unreal" AGI like Hutter's. I will >> try to see if watching Hofstadter can equip us with a few updated, powerful >> shared terms/metaphors, but at the end of the day the intuitions are for >> those who understand them and are willing and able to do three things: >> build, build, build! >> >> AT >> >> >> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >> <https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac> | >> Modify <https://www.listbox.com/member/?&;> Your Subscription >> <http://www.listbox.com> >> > > > > -- > Full employment can be had with the stoke of a pen. Simply institute a six > hour workday. That will easily create enough new jobs to bring back full > employment. > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/23050605-2da819ff> | > Modify<https://www.listbox.com/member/?&;>Your Subscription > <http://www.listbox.com> > ------------------------------------------- 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
