I made a feeble effort to define AGI-complete, the main idea being that, for some unknown and most likely unprovable reason, a system that handles a specific domain super-intelligently, for example visual scene analysis or audio source separation, could also handle most/all other domains at AGI levels. I have written elsewhere about my hunch that audio and video domains are indeed best handled with "hypothesis generation", you try a few different object configurations and then you conclude that one of them is the most likely explanation. I think it is a pattern that works in NLP too , for example "He entered the room" has to fire up guesses about who is he, what kind of room and even the different entries one can make. All in all, I would not be shocked if a "system" proves to be AGI-complete, but I'd probably be shocked if an equation or some other one-trick-pony is AGI-complete. As I explained before the problem with compression/statistics is the "entropic bias" and mean reversion, suggesting that what I will write next is one of the 1000 things I most often wrote before. Instead, symbolic/productive systems are counter-entropic, they are the proper way to produce one-in-a-zillion solutions to life's once-in-a-lifetime problems. Then again, the statistics and symbols will need a Bob Fosse choreography if they are going to work together.
It is a manifesto I am putting together, also meant to attract or otherwise my early collaborators, some who might also be business partners, but it is being fleshed out in a kind of continuous deployment fashion all the way to a possible specification document. It is meant to become a curated public wiki too, I just didn't like one thing or another with some wikis I briefly looked into, and it casually references entire research and business projects, for example social agents have one parameter "reputation", and maintaining that should be a whole business by itself. And as I am hinting from time to time, I am trying to define a research domain around what is necessary and sufficient in an agent in order to do physics (in whatever virtual or real environment). It is a very slippery subject but it would be quite the bomb if the first experimental results started to come in. AT On Wed, Mar 25, 2015 at 11:22 AM, Piaget Modeler <[email protected]> wrote: > *Corrections*: > > algorightms --> algorithms > > activities were people --> activitiies where people > > *Comments*: > > You should include Bob Fosse as one of Michael Jackson's influences for > dance in that collection of dancers. > > What does AGI-complete mean? > > Overall it's very high level, but it's a start. Just need to drill down > into the specifics. For example, there is an > Agent Framework. You also have a requirement for Natural Language > understanding and problem solving. > I'd take a look at some diagrams for NLU systems and problem solving > systems and combine them in some > fashion to get those features of your system, and then integrate them with > your agent framework. it could > be promising. > > Best of luck... > > ~PM > > ------------------------------ > From: [email protected] > Date: Wed, 25 Mar 2015 09:58:39 +0100 > Subject: Re: [agi] Vaporware HELLENE, this is it! > > To: [email protected] > > And part two http://www.hellene.eu/personml-mimicme , some typos were > also edited out of the intro. > > AT > > On Mon, Mar 23, 2015 at 10:36 AM, Anastasios Tsiolakidis < > [email protected]> wrote: > > I will prepare a diagram after defining some more acronyms. But the the > top-level control loop reads like Agents(updateBeliefs), one hopes with > perfect elasticity, meaning that the most important agents consume most of > the CPU cycles, and the rest almost nothing. The elasticity algorithm > should probably "evolve", having some copies of HELLENE "competing" against > each other with different settings. > > I value very highly the ease of distributing data and computations and > CouchDB scripts as well as Erlang bits are the most promising. I want the > codebase to pay the bills as well, being used early in commercial projects, > for which I envision a graph database as the most expressive interface. > Perhaps one of the CouchDB hybrid competitors or even parts of OpenCog can > do the job. > > Lastly, lower level physics, statistics and the like are supposed to be > transplanted as they are in their original languages and environments. > > AT > > On Mon, Mar 23, 2015 at 12:24 AM, Piaget Modeler via AGI <[email protected]> > wrote: > > Do you have an architecture diagram, and what language will you program it > in? > > Kindly advise. > > ~PM > > ------------------------------ > From: [email protected] > Date: Sun, 22 Mar 2015 22:18:57 +0100 > Subject: [agi] Vaporware HELLENE, this is it! > To: [email protected] > > I decided it was time to take a slightly more direct approach to my AGI > project, despite the somewhat "out there" nature of my intro paragraphs at > http://hellene-aghi.azurewebsites.net . If you have been my careful > reader you will find little new in that first page, or the other 10 that I > expect will conclude the theoretical framework of the project, basically my > gives and takes over prior art. I will probably mirror it somewhere where > it is easier to leave comments too. It is not a small undertaking at all, > so wish me luck! > > AT > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/19999924-4a978ccc> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/14050631-7d925eb1> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > > > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/19999924-4a978ccc> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/14050631-7d925eb1> | > 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
