RE: [agi] An AGI Test/Prize
Ben, That is sort of a neat kind of device. Will have to think about that as it is fairly dynamic I may have to look that one up and potentially experiment on it. The kinds of algebraic structures I'm talking about basically are as many as possible. Also things like sets w/o operators, and things not classified as algebraic but related. I can talk generally about this and then maybe specifics. The idea is that - let's say you are a developer and are writing say a web server. How do you go about it? First thing you do is scrounge the internet for snippets and source code, libraries, specs, etc.. The AGI I'm talking about is approached the same way cept' you scrounge mathematics publications generally dealing with abstract algebras. To start off though as there are hundreds of years of code snippets with proofs BTW but we start with simple stuff - groups, rings, fields, algebras, groupoids, etc. including sub-chunks and twists of these things. Sticking with discrete for starters except for some continuous here and there. One might ask why do it this way? The idea is that the framework is elaborate, universal, super powerful construct - basically all abstract math - defined by man cumulative over time, grounded in rigorous proofs and absolutes. The goal is to get everything into it meaning all data input is analyzed for algebraic structure and put into the thing. It's an algebraic superhighway mesh highly dense -yes you have to emulate it on digital computers - go from infinite algebraic mesh to physical real digital subset emulated BUT that's kind of what our brains do. We happen to live in (at least from day to day perspective) a very finite resource world. I'd like to delve deeper into digital physics but will not here J So there is a little background. All we are talking about is math and data and computer. So getting stuff into it? Think about it this way - built in lossy compression. Yes you have sensory memory duration gradations, example: photographic to skeletoid, but to get the algebraic structure is where the AI and stats tools get used. You can imagine how that works - but the goal is algebraic structure especially operators, magma detection, - imagine example a dog running look at all the cyclic groups going on - symmetry, sets, these are signatures, motion operators - subgroups of bodily movement definitions sampled is behavioral display, then put the dog into memory - morphisms storage - all dogs ever seen -think of a telescoping morphism tree index like structure. The AGI internals include morphism and functor networks kind of like analogy tree nets. Subgroups, subfields, etc. are very important as you leverage their structure defined onto their instance representations - Linguistic semantics? Same way. The AI and stats sensory has to break it up into algebraic structure. You need complexity detection. A view of a mountain and a view of a page of text have different complexity signatures. It detects text. The gradation from image to algebraic structure - the exploded text - sets and operators - processed according to its complexity sig, rips it apart put into the algebraic text structure mesh memory of built in telescoping morphism tree (or basically mossy or wormy structures at this point from a dimensional cross section view). The linguistic text structure is hierarchies of intersecting subsets and subgroups with morphic relational trees intersecting with cyclic group and subgroup indexors, etc.. tied into the KB through, once again algebraic structure. Knowledge is very compressed and cyclic group centric (seems like especially physicl world knowledge)- it sort of collapses with a self-organizing effect as more data is added where memories can be peeled off. Anyway, kind of understand where it's headed? John From: Benjamin Goertzel [mailto:[EMAIL PROTECTED] John Rose, As a long-lapsed mathematician, I'm curious about your system, but what you've said about it so far doesn't really tell me much... Do you have a mathematical description of your system? I did some theoretical work years ago representing complex systems dynamics in terms of abstract algebras. What I showed there was that you could represent a certain kind of multi-component system, with complex inter-component interactions, in such a way that its dynamic evolution over time is equivalent to the iteration of a quadratic function in a high-dimensional space with an eccentric multiplication table on it. The multiplication table basically encodes information of the form (component i) acts_on (component j) to produce (component k) where acts_on is the mult. operator So then complex systems dynamics all comes down to Julia sets and Mandelbrot sets on high-dimensional real algebras ;-) I never ended up making any use of this direction of thinking, but I found it interesting... This stuff made it into my 1997 book From Complexity to Creativity I believe... I am curious what
Re: [agi] An AGI Test/Prize
On Monday 22 October 2007 08:05:26 am, Benjamin Goertzel wrote: ... but dynamic long-term memory, in my view, is a wildly self-organizing mess, and would best be modeled algebraically as a quadratic iteration over a high-dimensional real non-division algebra whose multiplication table is evolving dynamically as the iteration proceeds Holy writhing Mandelbrot sets, Batman! Why real and non-division? I particularly don't like real -- my computer can't handle the precision :-) Josh - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56270025-9c1ac7
Re: [agi] An AGI Test/Prize
On 10/22/07, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: On Monday 22 October 2007 08:05:26 am, Benjamin Goertzel wrote: ... but dynamic long-term memory, in my view, is a wildly self-organizing mess, and would best be modeled algebraically as a quadratic iteration over a high-dimensional real non-division algebra whose multiplication table is evolving dynamically as the iteration proceeds Holy writhing Mandelbrot sets, Batman! Why real and non-division? I particularly don't like real -- my computer can't handle the precision :-) You need to get the new NVidia AIXI chip ... it's a bargain at $infinity.99 ;-) - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56281147-0ed02b
RE: [agi] An AGI Test/Prize
Holy writhing Mandelbrot sets, Batman! Why real and non-division? I particularly don't like real -- my computer can't handle the precision :-) Robin - forget all this digital stuff it's a trap, we need some analog nano-computers to help fight these crispy impostors! John - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56310633-8760ea
Re: [agi] An AGI Test/Prize
Benjamin Goertzel wrote: On 10/22/07, *J Storrs Hall, PhD* [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: On Monday 22 October 2007 08:05:26 am, Benjamin Goertzel wrote: ... but dynamic long-term memory, in my view, is a wildly self-organizing mess, and would best be modeled algebraically as a quadratic iteration over a high-dimensional real non-division algebra whose multiplication table is evolving dynamically as the iteration proceeds Holy writhing Mandelbrot sets, Batman! Why real and non-division? I particularly don't like real -- my computer can't handle the precision :-) You need to get the new NVidia AIXI chip ... it's a bargain at $infinity.99 ;-) Oh, it's not the price for the NVidia AIXI chip that bothers me, its the delivery: Amazon say that orders will be shipped when Hell reaches Zero Degrees Kelvin. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56326627-3df523
RE: [agi] An AGI Test/Prize
Yeah I'm not really agreeing with you here. I feel that, though I haven't really studied other cognitive software structures, but I feel that they can built simpler and more efficient. But I shouldn't come out saying that unless I attack some of the details right? But that's a gut reaction I have after working on so many large software projects. And it does depend on the view of cognition. Some of cognition is just hype it depends on what you are trying to build. There are a lot of warm-fuzzies, Dr. Feelgood things going on with cognition. I like cognition as a machine, a systematic controlled complexity modeler, edge of chaos surfing, crystallographic, polytopical harmonic, probabilistic sort of morphism and structure pump, with SOM injection - yeah I want a machine that rips through the fabric of reality mesh. John From: Benjamin Goertzel [mailto:[EMAIL PROTECTED] Well the problem is that branches of algebra like universal algebra and category theory, that don't assume highly particular algebraic rules, don't really have any deep theorems that tell you anything... Whereas the branches of algebra that really give you deep information, all pertain to highly specialized structures that are very unlikely to be relevant to cognition... - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56339571-d001db
RE: [agi] An AGI Test/Prize
Vladimir, I'm using system as kind of a general word for a set and operator(s). You are understanding it correctly except templates is not right. The templates are actually a vast internal complex of structure which includes morphisms which are like templates. But you are right it does seem like a categorization approach. When you say categorization approach can you point out an example of that that I can look into? John From: Vladimir Nesov [mailto:[EMAIL PROTECTED] John, What do you mean by system? You imply that these objects have a structure, or equivalently are abstract models of original input. So, you take original input in whatever form it's coming in and based on it you create instances of abstract structures according to templates that are known to system. Is it essentially correct? If so, it's very similar to categorization approach: you observe experience indirectly, through categorization structure that current perception system produces for it. If you need to model a boolean based space for some sort of sampled data world it sees and correlates to that, the thing would generate a boolean algebra modeled and represented onto that informational structure for that particular space instance being studied. For example electronics theory - it would need to model that world as an instance based on electronics descriptor items and operators in that particular world or space set. Electronics theory world could be spat out as something very minor that it understands. So, it would assembled a description 'in place' from local rules, based on information provided by specific experience. Is it a correct restatement? Not sure if my terminology is very standard but do you understand the thinking? It may very well be morphic to other AGI structures or theories I don't know but I kind of like the way it works represented as such because it seems simple and not messy but very comprehensive and has other good qualities. It's very vague, but can with a stretch of imagination be mapped to many other views. It's unclear with this level of detail. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56546239-4cc4b3
RE: [agi] An AGI Test/Prize
Your busy and I'm busy, so we can wait for another topic before communicating next. But our communication on this topic has been interesting. Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] -Original Message- From: John G. Rose [mailto:[EMAIL PROTECTED] Sent: Saturday, October 20, 2007 10:58 PM To: agi@v2.listbox.com Subject: RE: [agi] An AGI Test/Prize Edward, Oops missed that - CA (cellular automata) is something that some other people on the list could really enlighten you on as it gets really deep and elaborate. CA for me is a potential toolset for some basic programming and logic constructs so far. Yes all those goodies get modeled. I suppose I need to elaborate hey wait a sec how did I become the theorist on all this crap? heh John From: Edward W. Porter [mailto:[EMAIL PROTECTED] Subject: RE: [agi] An AGI Test/Prize So, do you or don't you model uncertainty, contradictory evidence, degree of similarity, and all those good things? And what is a CA, or don't i want to know? _ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/? http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56054006-4da6e4
Re: [agi] An AGI Test/Prize
On 10/21/07, John G. Rose [EMAIL PROTECTED] wrote: Vladimir, That may very well be the case and something that I'm unaware of. The system I have in mind basically has I/O that is algebraic structures. Everything that it deals with is modeled this way. Any sort of system that it analyzes it converts to a particular structure that represents the data. All of its internal mechanisms are mathematically abstracted out – except for ancillary hard coded out of band assistors, AI, statistics, database, etc. The idea is to have a system that can understand systems and generate systems specifically. John, What do you mean by system? You imply that these objects have a structure, or equivalently are abstract models of original input. So, you take original input in whatever form it's coming in and based on it you create instances of abstract structures according to templates that are known to system. Is it essentially correct? If so, it's very similar to categorization approach: you observe experience indirectly, through categorization structure that current perception system produces for it. If you need to model a boolean based space for some sort of sampled data world it sees and correlates to that, the thing would generate a boolean algebra modeled and represented onto that informational structure for that particular space instance being studied. For example electronics theory – it would need to model that world as an instance based on electronics descriptor items and operators in that particular world or space set. Electronics theory world could be spat out as something very minor that it understands. So, it would assembled a description 'in place' from local rules, based on information provided by specific experience. Is it a correct restatement? Not sure if my terminology is very standard but do you understand the thinking? It may very well be morphic to other AGI structures or theories I don't know but I kind of like the way it works represented as such because it seems simple and not messy but very comprehensive and has other good qualities. It's very vague, but can with a stretch of imagination be mapped to many other views. It's unclear with this level of detail. -- Vladimir Nesovmailto:[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56066281-5a9cbe
Re: [agi] An AGI Test/Prize
Has anyone come across (or written) any papers that argue for particular low-level capabilities that any system capable of human-level intelligence must possess, and which posits particular tests for assessing whether a system possesses these prerequisites for intelligence? I'm looking for anything like this, or indeed anything that tries to lay out an incremental path toward AGI with testable benchmarks along the way. I'd be very appreciative if anyone could point me to any such work. Gabe On 10/19/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: I largely agree. It's worth pointing out that Carnot published Reflections on the Motive Power of Fire and established the science of thermodynamics more than a century after the first working steam engines were built. That said, I opine that an intuitive grasp of some of the important elements in what will ultimately become the science of intelligence is likely to be very useful to those inventing AGI. Yeah, most certainly However, an intuitive grasp -- and even a well-fleshed-out qualitative theory supplemented by heuristic back-of-the-envelope calculations and prototype results -- is very different from a defensible, rigorous theory that can stand up to the assaults of intelligent detractors I didn't start seriously trying to design implement AGI until I felt I had a solid intuitive grasp of all related issues. But I did make a conscious choice to devote more effort to utilizing my intuitive grasp to try to design and create AGI, rather than to creating better general AI theories Both are worthy pursuits, and both are difficult. I actually enjoy theory better. But my sense is that the heyday of AGI theorizing is gonna come after AGI experimentation has progressed a good bit further than it has today... -- Ben G -- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55733424-f2512b
Re: [agi] An AGI Test/Prize
On 10/19/07, Matt Mahoney [EMAIL PROTECTED] wrote: http://www.vetta.org/documents/ui_benelearn.pdf Unfortunately the test is not computable. True but how about testing intelligence by comparing the performance of an agent across several computable environments (randomly-generated finite games) to the performance of a random agent? I suspect this measure would provide a reasonable estimate of Hutter's definition and could be made arbitrarily more accurate by increasing the number and complexity of test environments. David - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55740663-05e954
RE: [agi] An AGI Test/Prize
I guess I am mundane. I dont spend a lot of time thinking about a definition of intelligence. Goertzels is good enough for me. Instead I think in terms of what I want these machines to do -- which includes human-level: -NL understanding and generation (including discourse level) -Speech recognition and generation (including appropriate pitch and volume modulation) -Non-speech auditory recognition and generation -Visual recognition and real time video generation -World-knowledge representation, understanding and reasoning -Computer program understanding and generation -Common sense reasoning -Cognition -Context sensitivity -Automatic learning -Intuition -Creativity -Inventiveness -Understanding human nature and human desires and goals(not expecting full human-level here) -Ability to scan and store and, over time, convert and incorporate into learned deep structure vast amounts of knowledge including ultimately all available recorded knowledge To do such thinking I have come up with a fairly uniform approach to all these tasks, so I guess you could call that approach something approaching a theory of intelligence. But I mainly think of it as a theory of how to get certain really cool things done. I dont expect to get what is listed all at once, but, barring some major set back, this will probably all happen (with perhaps partial exception on the last item) within twenty years, and with the right people getting big money most of it could substantially all happen in ten. In addition, as we get closer to the threshold I think intelligence (at least from our perspective) should include: -helping make individual people, human organizations, and human government more intelligent, happy, cooperative, and peaceful -helping creating a transition into the future that is satisfying for most humans Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] -Original Message- From: John G. Rose [mailto:[EMAIL PROTECTED] Sent: Saturday, October 20, 2007 1:27 PM To: agi@v2.listbox.com Subject: RE: [agi] An AGI Test/Prize Interesting background about on some thermodynamics history J. But basic definitions of intelligence, not talking about reinventing particle physics here, a basic, workable definition, not rigorous mathematical proof just something simple. AI, AGI cmon not asking for tooo much. In my mind it is not looking that sophisticated at the atomic level and it seems like it is VERY applicable for implementation if not required for testing. Though Hutter and Legg are apparently working diligently on this stuff and have a lot papers. John I largely agree. It's worth pointing out that Carnot published Reflections on the Motive Power of Fire and established the science of thermodynamics more than a century after the first working steam engines were built. That said, I opine that an intuitive grasp of some of the important elements in what will ultimately become the science of intelligence is likely to be very useful to those inventing AGI. Yeah, most certainly However, an intuitive grasp -- and even a well-fleshed-out qualitative theory supplemented by heuristic back-of-the-envelope calculations and prototype results -- is very different from a defensible, rigorous theory that can stand up to the assaults of intelligent detractors I didn't start seriously trying to design implement AGI until I felt I had a solid intuitive grasp of all related issues. But I did make a conscious choice to devote more effort to utilizing my intuitive grasp to try to design and create AGI, rather than to creating better general AI theories Both are worthy pursuits, and both are difficult. I actually enjoy theory better. But my sense is that the heyday of AGI theorizing is gonna come after AGI experimentation has progressed a good bit further than it has today... _ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/? http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55773358-059800
Re: [agi] An AGI Test/Prize
Regarding testing grounds for AGI. Personally I feel that ordinary computer games could provide an excellent proving ground for the early stages of AGI, or maybe even better if they are especially constructed. Computer games are usually especially designed to encourage the player towards creativity and exploration. Take a simple platform game for example, at every new stage new graphics and monsters are introduced, and in large, the player undergoes a continuous self training that last throughout the whole game. Game developers carefully distribute rewards and challenges to make this learning process as smooth as possible. But also I would like to say that given any proving ground for the first stages of AGI could be misused if AGI designers bring specialized code into their system. So if there is to be a competition for first generation AGI, there would have to be some referee that evaluates how much domain specific knowledge has been encoded to any given system. For the late development stages of AGI, where we basically have virtual human minds, then we could use so hard problems that specialized code could not help the AGI system anymore. But I guess that at that time we have basically already solved the problem of AGI, and competitions where AGI systems compete in writing essays on some subject, could only be used to polish some already outlined solution to AGI. I am a fan of Novamente, but for example when I watched the movie where they trained an AGI dog, I was left with the question about what parts of its cognition was specialization. For example, the human teacher used natural language to talk to the dog. Did the dog understand any of it, and in that case, was there any special language module involved? Also, training a dog is quite open ended, and it is difficult to assess what is progress. This shows just how difficult it is to demonstrate AGI. Any demonstration of AGI would have to support a list of what cognitive aspects are coded, and which are learnt. Only then you can understand whether it is impressive or not. Also, because we need to have firm rules about what can be pre-programmed, and what needs to be learnt, it is easier if we used some world with pretty simple mechanics. What I basically would like to see is an AGI learning to play a certain computer game, starting by learning the fundamentals, and then playing it to the end. Take an old videogame classic like The Legend of Zelda. http://www.zelda.com/universe/game/zelda/. I know a lot of you would say that this is a far to simplistic world for training an AGI, but not if you prohibit ANY pre-programmed knowledge. You only allow the AGI system to start with proto-knowledge representation, and basically hard-wire the in-game rewards and punishemnts to the goal of the AGI. The AGI system would then have to learn basic concepts such as: objects moving around on the screen which graphics correspond to yourself walls where you can go keys that opens doors the concept of coming to a new screen when walking of the edge of one how screens relate to each other teleportation (the flute for anyone who remembers) If the AGI system then can learn to play the game to the end and slay Ganon based on only proto-knowledge, then maybe we have some interesting going on. Such an AGI could maybe be compared to a rodent running in a maze, even if the motoric and vision system are more complicated. Then we are ready to increase the complexity of the computer game, adding communication with other characters, more complex concepts and puzzles, more dimensions, more motorics etc.. Basically, I would like to se Novamente and similar AGI systems play some goal oriented computer game, since AGI in itself needs to be goal oriented. /R 2007/10/20, Benjamin Goertzel [EMAIL PROTECTED]: I largely agree. It's worth pointing out that Carnot published Reflections on the Motive Power of Fire and established the science of thermodynamics more than a century after the first working steam engines were built. That said, I opine that an intuitive grasp of some of the important elements in what will ultimately become the science of intelligence is likely to be very useful to those inventing AGI. Yeah, most certainly However, an intuitive grasp -- and even a well-fleshed-out qualitative theory supplemented by heuristic back-of-the-envelope calculations and prototype results -- is very different from a defensible, rigorous theory that can stand up to the assaults of intelligent detractors I didn't start seriously trying to design implement AGI until I felt I had a solid intuitive grasp of all related issues. But I did make a conscious choice to devote more effort to utilizing my intuitive grasp to try to design and create AGI, rather than to creating better general AI theories Both are worthy pursuits, and both are difficult. I actually enjoy theory better. But my sense is that the heyday of AGI
RE: [agi] An AGI Test/Prize
No you are not mundane. All these things on the list (or most) are very well to be expected from a generally intelligent system or its derivatives. But I have this urge, being a software developer, to smash all these things up into their constituent components, partition commonalties, eliminate dupes, and perhaps further smash up into an atomic representation of intelligence as little intelligent engines that can be combined in various ways to build higher level functions. Kind of like a cellular automata approach and perhaps CA structures can be used. I really don't want to waste 10 years developing a giant piece of bloatage code that never fully works. Better to exhaust all possibilities in the mind and on paper as much as possible as software dev can be a giant PIA mess if not thought out beforehand as much as possible. Yes you can go so far before doing prototyping and testing but certain prototypes can take many months to build. Several on this email list have already gotten to this point and it may be more productive digesting their systems instead of reinventing. Even so that leaves many questions open about testing. Someone can claim they have AGI but how do you really know, could be just a highly sophisticated chatterbot. John From: Edward W. Porter [mailto:[EMAIL PROTECTED] I guess I am mundane. I don't spend a lot of time thinking about a definition of intelligence. Goertzel's is good enough for me. Instead I think in terms of what I want these machines to do -- which includes human-level: -NL understanding and generation (including discourse level) -Speech recognition and generation (including appropriate pitch and volume modulation) -Non-speech auditory recognition and generation -Visual recognition and real time video generation -World-knowledge representation, understanding and reasoning -Computer program understanding and generation -Common sense reasoning -Cognition -Context sensitivity -Automatic learning -Intuition -Creativity -Inventiveness -Understanding human nature and human desires and goals(not expecting full human-level here) -Ability to scan and store and, over time, convert and incorporate into learned deep structure vast amounts of knowledge including ultimately all available recorded knowledge . To do such thinking I have come up with a fairly uniform approach to all these tasks, so I guess you could call that approach something approaching a theory of intelligence. But I mainly think of it as a theory of how to get certain really cool things done. I don't expect to get what is listed all at once, but, barring some major set back, this will probably all happen (with perhaps partial exception on the last item) within twenty years, and with the right people getting big money most of it could substantially all happen in ten. In addition, as we get closer to the threshold I think intelligence (at least from our perspective) should include: -helping make individual people, human organizations, and human government more intelligent, happy, cooperative, and peaceful -helping creating a transition into the future that is satisfying for most humans Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55777112-33cf1e
RE: [agi] An AGI Test/Prize
Well I'm neck deep in 55,000 semi-colons of code in this AI app I'm building and need to get this bastich out the do' and it's probably going to grow to 80,000 before version 1.0. But at some point it needs to grow a brain. Yes I have my AGI design in mind since late 90's and had been watching what would happen with Intelligenesis. I particularly think along the lines of an abstract algebra based engine basically an algebraic structure pump. Everything is sets and operators with probability glue, a lot of SOM and AI sensory. But recent ideas in category theory are molding it and CA's are always rearing their tiny little heads.. But spending time digesting Novamente theory and then general AGI structure that's valuable. Things like - graph storage indexing methodology - really need to spend some time on that especially learning from the people with experience. John From: Benjamin Goertzel [mailto:[EMAIL PROTECTED] On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: John, So rather than a definition of intelligence you want a recipe for how to make a one? Goertzel's descriptions of Novamente in his two recent books are the closest, publicly-available approximation of that of which I currently know. Actually my book on how to build an AGI are not publicly available at this point ... but I'm strongly leaning toward making them so ... it's mostly just a matter of finding time to proofread them, remove obsolete ideas, etc. and generally turn them from draft manuscripts into finalized manuscripts. I have already let a bunch of people read the drafts... Of course, a problem with putting material like this in dead-tree form is that the ideas are evolving. We learn new stuff as we proceed through implementing the stuff in the books But the basic framework (knowledge rep, algorithms, cognitive architecture, teaching methodology) has not changed as we've proceed through the work so far, just some of the details (wherein the devil famously lies ;-) - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55787440-e8ac33
Re: [agi] An AGI Test/Prize
Ah, gotcha... The recent book Advances in Artificial General Intelligence gives a bunch more detail than those, actually (though not as much of the conceptual motivations as The Hidden Pattern) ... but not nearly as much as the not-yet-released stuff... -- Ben On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: Ben, The books I was referring to were The Hidden Pattern and Artificial General Intelligence, both of which I purchased from Amazon. I know you have a better description, but what is in these two books is quite helpful. Ed Porter -Original Message- *From:* Benjamin Goertzel [mailto:[EMAIL PROTECTED] *Sent:* Saturday, October 20, 2007 4:01 PM *To:* agi@v2.listbox.com *Subject:* Re: [agi] An AGI Test/Prize On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: John, So rather than a definition of intelligence you want a recipe for how to make a one? Goertzel's descriptions of Novamente in his two recent books are the closest, publicly-available approximation of that of which I currently know. Actually my book on how to build an AGI are not publicly available at this point ... but I'm strongly leaning toward making them so ... it's mostly just a matter of finding time to proofread them, remove obsolete ideas, etc. and generally turn them from draft manuscripts into finalized manuscripts. I have already let a bunch of people read the drafts... Of course, a problem with putting material like this in dead-tree form is that the ideas are evolving. We learn new stuff as we proceed through implementing the stuff in the books But the basic framework (knowledge rep, algorithms, cognitive architecture, teaching methodology) has not changed as we've proceed through the work so far, just some of the details (wherein the devil famously lies ;-) -- Ben Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] -Original Message- *From:* John G. Rose [mailto:[EMAIL PROTECTED] *Sent:* Saturday, October 20, 2007 3:16 PM *To:* agi@v2.listbox.com *Subject:* RE: [agi] An AGI Test/Prize No you are not mundane. All these things on the list (or most) are very well to be expected from a generally intelligent system or its derivatives. But I have this urge, being a software developer, to smash all these things up into their constituent components, partition commonalties, eliminate dupes, and perhaps further smash up into an atomic representation of intelligence as little intelligent engines that can be combined in various ways to build higher level functions. Kind of like a cellular automata approach and perhaps CA structures can be used. I really don't want to waste 10 years developing a giant piece of bloatage code that never fully works. Better to exhaust all possibilities in the mind and on paper as much as possible as software dev can be a giant PIA mess if not thought out beforehand as much as possible. Yes you can go so far before doing prototyping and testing but certain prototypes can take many months to build. Several on this email list have already gotten to this point and it may be more productive digesting their systems instead of reinventing… Even so that leaves many questions open about testing. Someone can claim they have AGI but how do you really know, could be just a highly sophisticated chatterbot. John *From:* Edward W. Porter [mailto:[EMAIL PROTECTED] I guess I am mundane. I don't spend a lot of time thinking about a definition of intelligence. Goertzel's is good enough for me. Instead I think in terms of what I want these machines to do -- which includes human-level: -NL understanding and generation (including discourse level) -Speech recognition and generation (including appropriate pitch and volume modulation) -Non-speech auditory recognition and generation -Visual recognition and real time video generation -World-knowledge representation, understanding and reasoning -Computer program understanding and generation -Common sense reasoning -Cognition -Context sensitivity -Automatic learning -Intuition -Creativity -Inventiveness -Understanding human nature and human desires and goals(not expecting full human-level here) -Ability to scan and store and, over time, convert and incorporate into learned deep structure vast amounts of knowledge including ultimately all available recorded knowledge . To do such thinking I have come up with a fairly uniform approach to all these tasks, so I guess you could call that approach something approaching a theory of intelligence. But I mainly think of it as a theory of how to get certain really cool things done. I don't expect to get what is listed all at once, but, barring some major set back, this will probably
RE: [agi] An AGI Test/Prize
John, [A]bstract algebra based engine thats basically an algebraic structure pump sounds really exotic. Im visualizing a robo-version of my ninth grade algebra teacher on speed. If its not giving away the crown jewels, what in the hell is it and how does it fit into to AGI? And what are the always rearing CAs, you know, the ones with the tiny little heads? Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] -Original Message- From: John G. Rose [mailto:[EMAIL PROTECTED] Sent: Saturday, October 20, 2007 4:44 PM To: agi@v2.listbox.com Subject: RE: [agi] An AGI Test/Prize Well Im neck deep in 55,000 semi-colons of code in this AI app Im building and need to get this bastich out the do and its probably going to grow to 80,000 before version 1.0. But at some point it needs to grow a brain. Yes I have my AGI design in mind since late 90s and had been watching what would happen with Intelligenesis. I particularly think along the lines of an abstract algebra based engine basically an algebraic structure pump. Everything is sets and operators with probability glue, a lot of SOM and AI sensory. But recent ideas in category theory are molding it and CAs are always rearing their tiny little heads . But spending time digesting Novamente theory and then general AGI structure thats valuable. Things like - graph storage indexing methodology - really need to spend some time on that especially learning from the people with experience. John From: Benjamin Goertzel [mailto:[EMAIL PROTECTED] On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: John, So rather than a definition of intelligence you want a recipe for how to make a one? Goertzel's descriptions of Novamente in his two recent books are the closest, publicly-available approximation of that of which I currently know. Actually my book on how to build an AGI are not publicly available at this point ... but I'm strongly leaning toward making them so ... it's mostly just a matter of finding time to proofread them, remove obsolete ideas, etc. and generally turn them from draft manuscripts into finalized manuscripts. I have already let a bunch of people read the drafts... Of course, a problem with putting material like this in dead-tree form is that the ideas are evolving. We learn new stuff as we proceed through implementing the stuff in the books But the basic framework (knowledge rep, algorithms, cognitive architecture, teaching methodology) has not changed as we've proceed through the work so far, just some of the details (wherein the devil famously lies ;-) _ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/? http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55814342-eb1811
Re: [agi] An AGI Test/Prize
On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: You mean I wasted my time and money by buying and reading the Novamente article in Artificial General Intelligence when I could have bought the new and improved Advances in Artificial General Intelligence. What a rip off! Ed ((( Bummer, eh? ;-) Seriously though: The articles on NM in the newer AGI edited volume don't review the overall NM architecture and design as thoroughly as the article on NM in the older AGI edited volume. We tried not to be redundant in writing the NM articles for the new volume. However, the articles in the new volume do go into more detail on various specific aspects of the NM system. One problem with the original (older) Artificial General Intelligence book is that the articles in it were actually written in 2002, but the book did not appear until 2006! This was because of various delays associated with the publishing process, which fortunately were not repeated with the newer volume... The good news is, the articles on NM in the newer AGI edited volume are available online at the AGIRI.org website, on the page devoted to the 2006 AGIRI workshop... http://www.agiri.org/forum/index.php?act=STf=21t=23 -- Ben -Original Message- *From:* Benjamin Goertzel [mailto:[EMAIL PROTECTED] *Sent:* Saturday, October 20, 2007 5:24 PM *To:* agi@v2.listbox.com *Subject:* Re: [agi] An AGI Test/Prize Ah, gotcha... The recent book Advances in Artificial General Intelligence gives a bunch more detail than those, actually (though not as much of the conceptual motivations as The Hidden Pattern) ... but not nearly as much as the not-yet-released stuff... -- Ben On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: Ben, The books I was referring to were The Hidden Pattern and Artificial General Intelligence, both of which I purchased from Amazon. I know you have a better description, but what is in these two books is quite helpful. Ed Porter -Original Message- *From:* Benjamin Goertzel [mailto:[EMAIL PROTECTED] *Sent:* Saturday, October 20, 2007 4:01 PM *To:* agi@v2.listbox.com *Subject:* Re: [agi] An AGI Test/Prize On 10/20/07, Edward W. Porter [EMAIL PROTECTED] wrote: John, So rather than a definition of intelligence you want a recipe for how to make a one? Goertzel's descriptions of Novamente in his two recent books are the closest, publicly-available approximation of that of which I currently know. Actually my book on how to build an AGI are not publicly available at this point ... but I'm strongly leaning toward making them so ... it's mostly just a matter of finding time to proofread them, remove obsolete ideas, etc. and generally turn them from draft manuscripts into finalized manuscripts. I have already let a bunch of people read the drafts... Of course, a problem with putting material like this in dead-tree form is that the ideas are evolving. We learn new stuff as we proceed through implementing the stuff in the books But the basic framework (knowledge rep, algorithms, cognitive architecture, teaching methodology) has not changed as we've proceed through the work so far, just some of the details (wherein the devil famously lies ;-) -- Ben Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] -Original Message- *From:* John G. Rose [mailto: [EMAIL PROTECTED] *Sent:* Saturday, October 20, 2007 3:16 PM *To:* agi@v2.listbox.com *Subject:* RE: [agi] An AGI Test/Prize No you are not mundane. All these things on the list (or most) are very well to be expected from a generally intelligent system or its derivatives. But I have this urge, being a software developer, to smash all these things up into their constituent components, partition commonalties, eliminate dupes, and perhaps further smash up into an atomic representation of intelligence as little intelligent engines that can be combined in various ways to build higher level functions. Kind of like a cellular automata approach and perhaps CA structures can be used. I really don't want to waste 10 years developing a giant piece of bloatage code that never fully works. Better to exhaust all possibilities in the mind and on paper as much as possible as software dev can be a giant PIA mess if not thought out beforehand as much as possible. Yes you can go so far before doing prototyping and testing but certain prototypes can take many months to build. Several on this email list have already gotten to this point and it may be more productive digesting their systems instead of reinventing… Even so that leaves many questions open about testing. Someone can claim they have AGI but how do you really know, could be just a highly sophisticated
RE: [agi] An AGI Test/Prize
http://en.wikipedia.org/wiki/Algebraic_structure http://en.wikipedia.org/wiki/Cellular_automata Start reading.. John From: Edward W. Porter [mailto:[EMAIL PROTECTED] John, [A]bstract algebra based engine that's basically an algebraic structure pump sounds really exotic. I'm visualizing a robo-version of my ninth grade algebra teacher on speed. If its not giving away the crown jewels, what in the hell is it and how does it fit into to AGI? And what are the always rearing CAs, you know, the ones with the tiny little heads? - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55818972-3fd4c5
Re: [agi] An AGI Test/Prize
On 10/21/07, John G. Rose [EMAIL PROTECTED] wrote: http://en.wikipedia.org/wiki/Algebraic_structure http://en.wikipedia.org/wiki/Cellular_automata Start reading…. John, It doesn't really help in understanding how system described by such terms is related to implementation of AGI. It sounds pretty much like I use a Turing Machine, but with more exotic equivalent. If you could be more specific, it'd be interesting to have at least a rough picture of what your approach is about. -- Vladimir Nesovmailto:[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55822610-45b851
RE: [agi] An AGI Test/Prize
Vladimir, That may very well be the case and something that I'm unaware of. The system I have in mind basically has I/O that is algebraic structures. Everything that it deals with is modeled this way. Any sort of system that it analyzes it converts to a particular structure that represents the data. All of its internal mechanisms are mathematically abstracted out - except for ancillary hard coded out of band assistors, AI, statistics, database, etc. The idea is to have a system that can understand systems and generate systems specifically. If you need to model a boolean based space for some sort of sampled data world it sees and correlates to that, the thing would generate a boolean algebra modeled and represented onto that informational structure for that particular space instance being studied. For example electronics theory - it would need to model that world as an instance based on electronics descriptor items and operators in that particular world or space set. Electronics theory world could be spat out as something very minor that it understands. Not sure if my terminology is very standard but do you understand the thinking? It may very well be morphic to other AGI structures or theories I don't know but I kind of like the way it works represented as such because it seems simple and not messy but very comprehensive and has other good qualities. John From: Vladimir Nesov [mailto:[EMAIL PROTECTED] John, It doesn't really help in understanding how system described by such terms is related to implementation of AGI. It sounds pretty much like I use a Turing Machine, but with more exotic equivalent. If you could be more specific, it'd be interesting to have at least a rough picture of what your approach is about. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55824739-ba7a29
RE: [agi] An AGI Test/Prize
Hi Edward, I don't see any problems dealing with either discrete or continuous. In fact in some ways it'd be nice to eliminate discrete and just operate in continuous mode. But discrete maps very well with binary computers. Continuous is just a lot of discrete, the density depending on resources or defined as ranges in sets, other descriptors, etc. different ways. I'm not really well versed on NARS and Novamente so can't comment on them and they are light years down the road. They are basically in implementation stage, closer to realized utility, more than just theories. Oh those 55(80),000 lines of code are an AI product I am making so it is not AGI but the thing has basically stubs for AGI or could be used by AGI. But the methodology I am talking about seems to be very well workable with data from the real world. It's hard for me to find things that it doesn't work with although real tests need to be performed. BTW this type of thinking I'm sure is well analyzed by many abstract algebra mathematicians. Computability issues exist and these may make the theory not workable to a certain degree. I actually don't know enough about a lot of this math to really work it through deeply for a feasibility study (yet) and much of it is still up in the air. John What I found interesting is that, described at this very general level, what this is saying is actually related to my view of AGI, except that it appears to be based on a totally crisp, 1 or 0 view of the world. If that is correct, it may be very valuable in certain domains, with are themselves totally or almost totally crisp, but it won't work for most human-like thinking, because most human concepts and what they describe in the real world are not crisp. THAT IS, UNLESS, YOU PLAN TO MODEL CONCEPTUAL FLUIDITY, ITSELF, IN A TOTALLY CRISP, UNCERTAINTY-BASED, WAY, which is obviously doable at some level. I guess that is what you are referring to by saying our mind does crisp thinking all the time. Even most of us anti-crispies, plan to implement our fluid system on digital machinery using binary representation, which we hope will be crisp (but at the 22nm node it might be a little less than totally crisp.) But the issue is: do your crisp techniques efficiently learn and represent the fluidity of mental concepts, the non-literal similarity, and the many apparent contradictions, and the uncertainty that dominate in human thinking and sensory information about the real world? And if so, how is your approach different than that of the Novamente/Pei Wang-like approaches? And if so, how well are your (was it) 80,000 lines of code of working at actually representing and making sense of the shadows projected on the walls of your AGI's cave by sensations (or data) from the real world. Ed Porter, P.S. Re CA: maybe I am well versed in them but I don't know what the acronym stands for. If it wouldn't be too much trouble could you please educate me on the subject? - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56005098-c2de21
RE: [agi] An AGI Test/Prize
So, do you or don't you model uncertainty, contradictory evidence, degree of similarity, and all those good things? And what is a CA, or don't i want to know? Edward W. Porter Porter Associates 24 String Bridge S12 Exeter, NH 03833 (617) 494-1722 Fax (617) 494-1822 [EMAIL PROTECTED] -Original Message- From: John G. Rose [mailto:[EMAIL PROTECTED] Sent: Saturday, October 20, 2007 10:39 PM To: agi@v2.listbox.com Subject: RE: [agi] An AGI Test/Prize Hi Edward, I dont see any problems dealing with either discrete or continuous. In fact in some ways itd be nice to eliminate discrete and just operate in continuous mode. But discrete maps very well with binary computers. Continuous is just a lot of discrete, the density depending on resources or defined as ranges in sets, other descriptors, etc. different ways. Im not really well versed on NARS and Novamente so cant comment on them and they are light years down the road. They are basically in implementation stage, closer to realized utility, more than just theories. Oh those 55(80),000 lines of code are an AI product I am making so it is not AGI but the thing has basically stubs for AGI or could be used by AGI. But the methodology I am talking about seems to be very well workable with data from the real world. Its hard for me to find things that it doesnt work with although real tests need to be performed. BTW this type of thinking Im sure is well analyzed by many abstract algebra mathematicians. Computability issues exist and these may make the theory not workable to a certain degree. I actually dont know enough about a lot of this math to really work it through deeply for a feasibility study (yet) and much of it is still up in the air John What I found interesting is that, described at this very general level, what this is saying is actually related to my view of AGI, except that it appears to be based on a totally crisp, 1 or 0 view of the world. If that is correct, it may be very valuable in certain domains, with are themselves totally or almost totally crisp, but it wont work for most human-like thinking, because most human concepts and what they describe in the real world are not crisp. THAT IS, UNLESS, YOU PLAN TO MODEL CONCEPTUAL FLUIDITY, ITSELF, IN A TOTALLY CRISP, UNCERTAINTY-BASED, WAY, which is obviously doable at some level. I guess that is what you are referring to by saying our mind does crisp thinking all the time. Even most of us anti-crispies, plan to implement our fluid system on digital machinery using binary representation, which we hope will be crisp (but at the 22nm node it might be a little less than totally crisp.) But the issue is: do your crisp techniques efficiently learn and represent the fluidity of mental concepts, the non-literal similarity, and the many apparent contradictions, and the uncertainty that dominate in human thinking and sensory information about the real world? And if so, how is your approach different than that of the Novamente/Pei Wang-like approaches? And if so, how well are your (was it) 80,000 lines of code of working at actually representing and making sense of the shadows projected on the walls of your AGIs cave by sensations (or data) from the real world. Ed Porter, P.S. Re CA: maybe I am well versed in them but I dont know what the acronym stands for. If it wouldnt be too much trouble could you please educate me on the subject? _ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/? http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=56007871-ae3472
Re: [agi] An AGI Test/Prize
Well, one problem is that the current mathematical definition of general intelligence is exactly that -- a definition of totally general intelligence, which is unachievable by any finite-resources AGI system... On the other hand, IQ tests and such measure domain-specific capabiities as much as general learning ability So human-oriented IQ tests are not so important I tend to think there should be some kind of test for general intelligence that is based on the requirement for self-understanding Humans have fairly rich dynamic internal models of themselves, cockroaches don't, and dogs have only pretty lame ones... Perhaps there could be a test that tries to measure the ability of a system to predict its own reaction to various novel situations? This would require the system to be able to model itself internally... However, it's still hard to make this kind of test objective in any sense, as different AGI systems will be adapted to different kinds of environments... Still, this is an interesting measure by which we could compare the self-understanding of different systems that live in the same environments... But, still, this is not really an objective measure for intelligence ... just another sorta-interesting sort of test... -- Ben On 10/19/07, John G. Rose [EMAIL PROTECTED] wrote: I think that there really needs to be more very specifically defined quantitative measures of intelligence. If there were questions that could be asked of an AGI that would require x units of intelligence to solve otherwise they would be unsolvable. I know that this is a hopeless foray on this list as there seems to be no basic mathematical definition of intelligence (someone correct me if I'm wrong). I do have in my mind a picture of what a minimalistic intelligence engine would look like so I feel strongly that there is a basic definition and the questions to test it would be in the form of feeding it successive layers of known background information and having it derive specific answers to questions based on that. Other qualities like creativity and imagination would need to be measured in other ways. John - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55532708-296773
Re: [agi] An AGI Test/Prize
John: I think that there really needs to be more very specifically defined quantitative measures of intelligence. ...Other qualities like creativity and imagination would need to be measured in other ways. The only kind of intelligence you can measure with any precision is narrow AI - convergent intelligence. That's what education marks out of 100 in tests with right/wrong answers. The other kind - AGI - divergent intelligence - can't be marked mathematically - can only be graded. That's what education gives grades to - Excellent/ Very Good/ Good/ Poor etc. You can't mark essays/ projects, for example, with precision, (or indeed the socially creative projects like novels/ new business plans/.(Navemente/A2I)). And they're half of education and half of intelligence. Sorry if that distresses those who can't live without maths - but that's life. P.S. It's worth pointing out that there are TWO kinds of intelligence - and there can be NO ARGUMENT about that here. You can argue about their definitions, not about their twoness. So Ben's and Pei's mono definitions of intelligence, for example, are a priori wrong. How can one be so dogmatic? Well, you guys have, for a start, to be able to distinguish between narrow AI and AGI, (which by my maths makes two kinds of intelligence) - otherwise you might as well cut your own throats professionally speaking. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55528758-3ab744
RE: [agi] An AGI Test/Prize
I think that there really needs to be more very specifically defined quantitative measures of intelligence. If there were questions that could be asked of an AGI that would require x units of intelligence to solve otherwise they would be unsolvable. I know that this is a hopeless foray on this list as there seems to be no basic mathematical definition of intelligence (someone correct me if I'm wrong). I do have in my mind a picture of what a minimalistic intelligence engine would look like so I feel strongly that there is a basic definition and the questions to test it would be in the form of feeding it successive layers of known background information and having it derive specific answers to questions based on that. Other qualities like creativity and imagination would need to be measured in other ways. John - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55519579-9983dc
Re: [agi] An AGI Test/Prize
I guess, off the top of my head, the conversational equivalent might be a Story Challenge - asking your AGI to tell some explanatory story about a problem that had occurred to it recently, (designated by the tester), and then perhaps asking it to devise a solution. Just my first thought - but the point is any AGI Test should be a focussed challenge, like the ICRA, not a vague Interview Test or similar. Hmmm... the storytelling direction is interesting. E.g., you could tell the first half of a story to the test-taker, and ask them to finish it... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54984670-ce964d
Re: [agi] An AGI Test/Prize
On 10/18/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: Hmmm... the storytelling direction is interesting. E.g., you could tell the first half of a story to the test-taker, and ask them to finish it... Or better, draw an animation of (both halves of) it. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=54994261-20c078
Re: [agi] An AGI Test/Prize
I think AGI test should fundamentally be a learning ability test. When there's a specified domain in which the system should demonstrate it competency (like 'chatting' or 'playing Go'), it's likely easier to write narrow solution. If system is not a RSI AI already, resulted competency depends on quirks of given domain too much, and it's unclear how improvements in general learning ability translate in competency. I see such test along the lines of feeding the system a stream of frame-like representations, and then it should be able to fill in the blanks in incomplete representations based on analogies. It's general enough to be AGI-complete, and simple enough to test existing narrow AI systems. Depending on supplied data it can be taken out of reach of algorithms which are too biased towards their narrow domain. Frame-like representations allow to construct tasks of different complexity according to human intuition, and likewise test their feasibility. This input stream shouldn't be too cluttered (it shouldn't include things like cyc database, wikipedia, etc.), but should assume zero knowledge. -- Vladimir Nesovmailto:[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=55007138-dd3f75
