Hi Ben, Got 5 so I thought I'd work through your points.
On Fri, May 22, 2015 at 3:33 AM, Benjamin Kapp <[email protected]> wrote: > I agree that we should do our best to avoid obscure terminology, because > it will make it easier for us to understand each other. > > Regarding surveying the field of AGI. It seems that everyone who does > such a survey talks about the cyc project (hard code all knowledge of the > world into your system). It seems like this information could be useful > for AGI, but of course the methodology seems fundamentally flawed since > that isn't how humans acquire their knowledge. > Here you have touched on something important and under-appreciated. The AGI program is less about the AGI knowing anything and more about 'finding out' from a point of ignorance. I see CYC and even Wikipedia as a new abstract compendium of knowledge that is utterly meaningless to any AGI that is not in the world like we are, and that must, to be a real AGI, be able to *add to*. Autonomously. Not via a human telling it what new knowledge looks like. The H-AGI substrate (wet or dry) is there to give the entity access to the world in a way that is more like humans. What way is that? Well that is for the IGI to investigate. That is what is un-explored in the AI and AGI programs to date. > But humans are born with some knowledge of the world, for example babies > are born with the ability to swim, and to avoid crawling off of cliffs and > such or to put it more simply humans are born with "instincts" which can be > thought of as a kind of knowledge. Perhaps we could give our AGI the cyc > knowledge as "instinctual" knowledge? But we wouldn't expect our AGI to > exclusively acquire knowledge in this fashion since, hardcoding all of > human knowledge isn't necessary, and difficult (if not impossible). > Although it is my understanding that the cyc project has begun to utilize > automated means of adding knowledge to their system, and as such it isn't > fair to say they only go about hardcoding knowledge of the world. > That CYC automation is, of course something created by a human, not something innate I in the C-AGI that CYC is. It may be useful! Worth doing. But not what an H-AGI is doing. An H-AGI may actually have such C-AGI knowledge hardwired that somehow might be called instinct and/or reflexes. I can see that in an H-AGI. > > Of course automated knowledge acquisition is precisely the method used by > IBM's Watson. For example it read all of wikipedia and generated knowledge > of the world from this. Interestingly they also read the urban dictionary, > but found that it degraded the performance of their system, and so they had > to "unlearn" this. Garbage in garbage out it would seem. An interesting > aspect of IBM's Watson is that they do indeed use a kind of hybrid system. > They have many algorithms which they use when they try to answer Jeopardy > questions, and they have another system that measures the confidence of the > results of each of these algorithms and the one with the highest confidence > is selected as the final answer. Of course they are just doing > computational linguistics, and so their system has no hope of working on > non text based modalities, which is a serious limitation, and certainly an > indication that it isn't really AGI. > > So it would seem that our system would have as a necessary requirement > that it can work with all kinds of input, be it sight, sound, touch, etc. > Although if it had the ability to query Watson as a module in some part of > the system, it isn't immediately obvious to me that this would be a bad > thing. It would just be the case that the Watson approach in and of its > self is not viable for creating real AGI. If i had the ability to query > google in my brain, surely google wouldn't become a kind of AGI even though > it would be part of an AGI (namely me). but the essential part of the > system that was creating the AGI would be my brain, not google. And so > perhaps it doesn't make much sense to just wire together a bunch of non agi > systems in the hopes that together they will create AGI? > I can see now you have 'got it'. These knowledge abstractions Watson. Google, Wikipedia CYC etc etc can be integrated in as a sensory/perceptual system to the core biophysical substrate. That will add to knowledge acquired through the traditional sensory modes (vision, audition, etc). That makes it H-AGI. In an of themselves, Watson, google etc can be thought of as not being H-AGI, not because they are comparatively ignorant, but because humans are built into the process of finding out new knowledge or telling them what new knowledge looks like. Humans are a kind of 'intrinsic puppeteer'. In contrast an H-AGI will acquire new knowledge on its own. Like humans. Our human substrate is intrinsically able to *learn how to learn* new things. That is something a human baby has that no AI has. In a very real way the H-AGI program is about building such a baby, But not a human level baby. Maybe worm, ant or bee, first. The H-AGI idea is about knowledge dynamics (change), not any particular knowledge. Mathematically H-AGI is about dKnowledge(t)/dt , not Knowledge(t) itself. That's where the differences (I predict) will be scientifically measured (by the IGI!). That understanding may then feed back into traditional computer-only approaches. Don't know. We have to do it to find out. Am not sure how to express this concisely in the planned paper/document. It probably needs to be expressed somehow. I'll have a go and maybe we can workshop it into a formal position. > > > On Thu, May 21, 2015 at 12:29 PM, Benjamin Kapp <[email protected]> wrote: > >> How about we have a discussion sometime this weekend? We have some >> serious timezone issues to work around.. >> >> On Wed, May 20, 2015 at 10:48 PM, Mark Seveland <[email protected]> >> wrote: >> >>> Just a suggestion. Google+ Meetups are a good way for everyone to meet >>> each other, and in live voice and/or video chat discuss topics. >>> >>> On Wed, May 20, 2015 at 7:33 PM, Colin Hales <[email protected]> >>> wrote: >>> >>>> Hi Dorian et. al., >>>> I am having trouble getting time to properly participate here because >>>> of family stuff and my other commitments. I'm checking in to acknowledge >>>> how encouraging it is to see the activity is ongoing, and the birth of a >>>> possible paper that might underpin whatever this IGI initiative turns into. >>>> >>>> I'd like to focus my efforts on the paper primarily as a way to >>>> discover IGI directions. So if you could bear with a patchy contribution >>>> from me for a little while it would be greatly appreciated. I have a >>>> particularly difficult week ahead of me. There's no huge crashing need for >>>> speed here, so I'm hoping slow and steady might be OK. >>>> >>>> Whatever form this website takes: fantastic. It may only ever be a >>>> 'line in the sand'. But it's a significant one in the greater scheme of AGI >>>> futures and really good to see after being sidelined for so long. Yay! >>>> >>>> cheers >>>> Colin Hales >>>> >>>> >>>> >>>> On Thu, May 21, 2015 at 10:07 AM, Mike Archbold <[email protected]> >>>> wrote: >>>> >>>>> Why don't you just call it "AI" and if somebody asks THEN you can >>>>> clarify it? I mean, why be arcane about it? One of the reasons I got >>>>> into AI is because I don't like the way that people create things that >>>>> are intentionally difficult and known only to the in-group. Now here >>>>> you go with a boatload of new acronyms, known only to the select tiny >>>>> group that knows the secret meaning behind it. So, I guess I am >>>>> getting into Alan Grimes vent space with this. >>>>> >>>>> On 5/20/15, Dorian Aur <[email protected]> wrote: >>>>> > *Colin et al,* >>>>> > >>>>> > >>>>> > A possible plan for H-AGI towards S-AGI paper >>>>> > >>>>> > >>>>> > >>>>> > *Hybrid artificial general intelligent systems towards S-AGI* >>>>> > >>>>> > >>>>> > >>>>> > >>>>> > >>>>> > *Introduction* – a short presentation of AI systems and general goal >>>>> to >>>>> > build human general intelligence >>>>> > >>>>> > >>>>> > >>>>> > Why H-AGI? >>>>> > >>>>> > - Present different forms of computation , ( particular forms of >>>>> > computation analog, digital -Turing machines ) >>>>> > - Computations in the brain (examples of computations that are >>>>> hardly >>>>> > replicated on digital computers) >>>>> > - H-AGI can include all forms of computations, algorithmic / >>>>> > non-algorithmic, analog, digital,* quantum and classical *since >>>>> > biological structure is incorporated in the system >>>>> > >>>>> > >>>>> > >>>>> > *Steps to develop H-AGI* >>>>> > >>>>> > >>>>> > >>>>> > - A. Build the structure using either natural stem cells or >>>>> induced >>>>> > pluripotent cells a three-dimensional vascularized structure, >>>>> test 3D >>>>> > printing possibilities >>>>> > - Shape the structure and control spatial organization of cells >>>>> > - Detect the need of neurotrophic factors, nutrients and oxygen >>>>> ...use >>>>> > nanosensor devices, carbon nanotubes... >>>>> > - Regulate, control the entire phenomenon using a computer >>>>> interface, >>>>> > ability to use combine analog/digital and biophysical computations >>>>> > >>>>> > B. Train the hybrid system >>>>> > >>>>> > - Enhance bidirectional communication between biological >>>>> structure and >>>>> > computers >>>>> > - Create and use a virtual world to provide accelerated >>>>> training, use >>>>> > machine learning, DL, digital/algorithmic AI or AGI if >>>>> something is >>>>> > developed on digital systems >>>>> > - The interactive training system should also shape the evolution >>>>> of >>>>> > biological structure, natural language and visual information >>>>> can be >>>>> > progressively included >>>>> > >>>>> > see details in Can we build a conscious machine, >>>>> > http://arxiv.org/abs/1411.5224 >>>>> > >>>>> > >>>>> > *Goals of H-AGI* >>>>> > >>>>> > H-AGI can be seen as a transitional step required to understand >>>>> which >>>>> > parts can be fully replicated in a synthetic form to build a more >>>>> powerful >>>>> > system, >>>>> > >>>>> > · Natural language processing, robotics... >>>>> > >>>>> > · Space exploration, colonization..... etc >>>>> > >>>>> > · Techniques for therapy (brain diseases, cancer ....) since >>>>> we will >>>>> > learn how to shape biological structure >>>>> > >>>>> > >>>>> > >>>>> > >>>>> > Dorian >>>>> > >>>>> > >>>>> > PS This brief presentation may also provide an idea about possible >>>>> > collaboration list 1- list 3 >>>>> > >>>>> > >>>>> > >>>>> > On Tue, May 19, 2015 at 11:20 PM, Mike Archbold <[email protected] >>>>> > >>>>> > wrote: >>>>> > >>>>> >> > A summary ....we are looking at the idea that there are 2 >>>>> fundamental >>>>> >> kinds >>>>> >> > of putative AGI (1) & (3), and their hybrid (2) that forms a third >>>>> >> approach >>>>> >> > as follows: >>>>> >> > >>>>> >> > (1) C-AGI computer substrate only. Neuromorphic equivalents >>>>> of it. >>>>> >> > (2) H-AGI hybrid of (1) and (3). The inorganic version is a >>>>> new >>>>> >> > kind >>>>> >> > of neuromorphic chip. The organic version has ... erm... organics >>>>> in >>>>> >> > it. >>>>> >> > (3) S-AGI synthetic AGI. organic or inorganic. Natural brain >>>>> >> > physics >>>>> >> > only. No computer. >>>>> >> > >>>>> >> > (aside: S-AGI just came out of my fingers. I hope this is OK, >>>>> Dorian!) >>>>> >> > >>>>> >> >>>>> >> This is a cool idea, somewhat mind boggling in its possibilities. >>>>> >> Cool though! >>>>> >> >>>>> >> Personally I would favor something more like "EM-AGI" for >>>>> >> electromagnetic AGI. I mean, I don't understand the details of the >>>>> >> approach, only the generalities. But, "S" seems a bit >>>>> vague/ambiguous >>>>> >> while EM hits it more or less on target IMHO. >>>>> >> >>>>> >> MIke A >>>>> >> >>>>> >> >>>>> >> > Think this way: What we have now is 100% computer. S-AGI is 100% >>>>> >> > natural >>>>> >> > physics (organic or inorganic). H-AGI is set somewhere in between. >>>>> >> > It's >>>>> >> > the level of computer computation/natural computation that is at >>>>> issue. >>>>> >> All >>>>> >> > are computation. >>>>> >> > >>>>> >> > The human brain is a natural version of (3) with a >>>>> neuronal/astrocyte >>>>> >> > substrate. (3) has no computer whatever in it. it retains all the >>>>> >> natural >>>>> >> > physics (whatever that is). H-AGI targets the inclusion of the >>>>> >> > essential >>>>> >> > natural brain physics in the substrate of (2) and to incorporate >>>>> (1) >>>>> >> > computer-substrates and software to an extent to be determined. >>>>> In my >>>>> >> case >>>>> >> > an H-AGI would be inorganic. Others see differently. >>>>> >> > >>>>> >> > Where you might have a stake in this? >>>>> >> > >>>>> >> > The history of AGI can be summed up as an experiment that seeks >>>>> to see >>>>> >> > if >>>>> >> > the role of (1) C-AGI as a brain is fundamentally >>>>> indistinguishable >>>>> >> > from >>>>> >> > (3) S-AGI under all conditions. That is the hypothesis. The 65 >>>>> year old >>>>> >> bet >>>>> >> > that has attracted 100% of the investment to date. H-AGI does not >>>>> make >>>>> >> that >>>>> >> > presupposition and seeks to contrast (1) and (3) in revealing >>>>> ways that >>>>> >> > then allow us to speak authoritatively about the (1)/(3) >>>>> relationship >>>>> >> > in >>>>> >> > AGI potential. Only then will we really understand the difference >>>>> >> > between >>>>> >> > (1) and (3). So far that difference is entirely and intuition. A >>>>> good >>>>> >> one. >>>>> >> > But only intuition. Its time for that intuition to be turned into >>>>> >> science. >>>>> >> > Experiments in (1) have ruled to date. Now we seek to do some >>>>> (2)... >>>>> >> > E.E. >>>>> >> > we have 65 years of 'control' subject. H-AGI builds the first >>>>> 'test' >>>>> >> > subject. >>>>> >> > >>>>> >> > How about this? >>>>> >> > >>>>> >> > What would be super cool is if this mighty AGI beast you intend >>>>> making >>>>> >> > could be turned into the brain of a robot. Then we could contrast >>>>> what >>>>> >> > it >>>>> >> > does with what an IGI candidate brain does in an identical robot >>>>> in the >>>>> >> > same test. That kind of testing vision (as far off as it may >>>>> seem) is a >>>>> >> > potential way your work and the IGI might interface. Which >>>>> candidate >>>>> >> robot >>>>> >> > best encounters radical novelty, without any human >>>>> >> intervention/involvement >>>>> >> > whatever? .... is a really good question. To do this test you'd >>>>> not >>>>> >> > need >>>>> >> to >>>>> >> > reveal anything about its workings. Observed robot behaviour is >>>>> >> > decisive. >>>>> >> > >>>>> >> > It seems to me that whatever venture you plan, it might be wise >>>>> to keep >>>>> >> an >>>>> >> > eye on any (2)/(3) approaches. IGI or not. Because it is directly >>>>> >> informing >>>>> >> > expectations of outcomes in (1). We are currently asking the >>>>> question >>>>> >> "*If >>>>> >> > H-AGI were to be championed into existence, what would the first >>>>> >> > vehicle >>>>> >> > for that look like?*" If the enthusiasm maintains it will be >>>>> sketched >>>>> >> into >>>>> >> > a web page and we'll see what it tells us and what to do next. It >>>>> may >>>>> >> halt. >>>>> >> > It may go. I don't know. Worth a shot? You bet. >>>>> >> > >>>>> >> > With your (1) C-AGI glasses firmly strapped to your head, your >>>>> wisdom >>>>> >> > at >>>>> >> > all stages in this would be well received, whatever the messages. >>>>> So if >>>>> >> you >>>>> >> > have time to keep an eye on happenings, I for one would >>>>> appreciate it. >>>>> >> > >>>>> >> > regards >>>>> >> > >>>>> >> > Colin Hales >>>>> >> > >>>>> >> > >>>>> >> > >>>>> >> > On Wed, May 20, 2015 at 6:58 AM, Peter Voss <[email protected]> >>>>> wrote: >>>>> >> > >>>>> >> >> Thanks for asking. Haven’t followed the IGI discussions. >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> Is this about non-computer based approaches to AGI? If so, I >>>>> don’t >>>>> >> think >>>>> >> >> I have anything positive to contribute. >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> More generally, non-profit orgs need strong focus and >>>>> champions. And >>>>> >> >> specific goals. >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> *From:* Benjamin Kapp [mailto:[email protected]] >>>>> >> >> *Sent:* Tuesday, May 19, 2015 12:23 PM >>>>> >> >> *To:* AGI >>>>> >> >> *Subject:* Re: [agi] Institute of General Intelligence (IGI) >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> Mr. Voss, >>>>> >> >> >>>>> >> >> Given your understanding of the AGI community do you believe an >>>>> IGI >>>>> >> would >>>>> >> >> be redundant? Would your organization be open to collaborating >>>>> with >>>>> >> >> the >>>>> >> >> IGI? Do you have any advice for how we could be successful in >>>>> >> >> starting >>>>> >> >> up >>>>> >> >> this organization? Perhaps you would be open to being a member >>>>> of the >>>>> >> >> board? >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> On Tue, May 19, 2015 at 2:03 PM, Peter Voss <[email protected]> >>>>> wrote: >>>>> >> >> >>>>> >> >> Not something that can be adequately covered in a few words, >>>>> but…. >>>>> >> “We’re >>>>> >> >> building a fully integrated, top-down & bottom-up, real-time, >>>>> adaptive >>>>> >> >> knowledge (& skill) representation, learning and reasoning >>>>> engine. >>>>> >> >> We’re >>>>> >> >> using a combination of graph representation and NN techniques >>>>> overlaid >>>>> >> >> with >>>>> >> >> fuzzy, adaptive rule systems” – ha! >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> Here again are links for some clues: >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >>>>> http://www.kurzweilai.net/essentials-of-general-intelligence-the-direct-path-to-agi >>>>> >> >> >>>>> >> >> http://www.realagi.com/index.html >>>>> >> >> >>>>> >> >> https://www.facebook.com/groups/RealAGI/ >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> *From:* Benjamin Kapp [mailto:[email protected]] >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> Mr. Voss, >>>>> >> >> >>>>> >> >> Since you are the founder I'm certain you know what agi-3's >>>>> >> >> methodology >>>>> >> >> is. In a few words (maybe more?) could you share with us what >>>>> that >>>>> >> >> is? >>>>> >> >> >>>>> >> >> >>>>> >> >> >>>>> >> >> On Tue, May 19, 2015 at 1:24 PM, Peter Voss <[email protected]> >>>>> wrote: >>>>> >> >> >>>>> >> >> *>*http://www.agi-3.com They just glue together anything and >>>>> >> everything >>>>> >> >> that works. >>>>> >> >> >>>>> >> >> Actually, no. We have a very specific theory of AGI and >>>>> architecture >>>>> >> >> >>>>> >> >> *Peter Voss* >>>>> >> >> >>>>> >> >> *Founder, AGI Innovations Inc.* >>>>> >> >> >>>>> >> >> *AGI* | Archives < >>>>> https://www.listbox.com/member/archive/303/=now> >>>>> >> >> < >>>>> https://www.listbox.com/member/archive/rss/303/26973278-698fd9ee>| >>>>> >> >> 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/303/231420-b637a2b0 >>>>> >| >>>>> >> 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/303/11721311-20a65d4a> | >>>>> >> >> 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/303/11943661-d9279dae >>>>> >> > Modify Your Subscription: >>>>> >> > https://www.listbox.com/member/?&; >>>>> >> > Powered by Listbox: http://www.listbox.com >>>>> >> > >>>>> >> >>>>> >> >>>>> >> ------------------------------------------- >>>>> >> AGI >>>>> >> Archives: https://www.listbox.com/member/archive/303/=now >>>>> >> RSS Feed: >>>>> >> https://www.listbox.com/member/archive/rss/303/17795807-366cfa2a >>>>> >> Modify Your Subscription: >>>>> >> https://www.listbox.com/member/?&; >>>>> >> Powered by Listbox: http://www.listbox.com >>>>> >> >>>>> > >>>>> > >>>>> > >>>>> > ------------------------------------------- >>>>> > AGI >>>>> > Archives: https://www.listbox.com/member/archive/303/=now >>>>> > RSS Feed: >>>>> https://www.listbox.com/member/archive/rss/303/11943661-d9279dae >>>>> > Modify Your Subscription: >>>>> > https://www.listbox.com/member/?&; >>>>> > Powered by Listbox: http://www.listbox.com >>>>> > >>>>> >>>>> >>>>> ------------------------------------------- >>>>> AGI >>>>> Archives: https://www.listbox.com/member/archive/303/=now >>>>> RSS Feed: >>>>> https://www.listbox.com/member/archive/rss/303/11721311-20a65d4a >>>>> Modify Your Subscription: https://www.listbox.com/member/?&; >>>>> Powered by Listbox: http://www.listbox.com >>>>> >>>> >>>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>>> <https://www.listbox.com/member/archive/rss/303/27079473-66e47b26> | >>>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>>> <http://www.listbox.com> >>>> >>> >>> >>> >>> -- >>> Regards, >>> Mark Seveland >>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>> <https://www.listbox.com/member/archive/rss/303/26973278-698fd9ee> | >>> 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/303/11721311-20a65d4a> | > 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
