Re: [agi] Wozniak's defn of intelligence
Charles D Hixson wrote: Richard Loosemore wrote: J Storrs Hall, PhD wrote: On Friday 08 February 2008 10:16:43 am, Richard Loosemore wrote: J Storrs Hall, PhD wrote: Any system builders here care to give a guess as to how long it will be before a robot, with your system as its controller, can walk into the average suburban home, find the kitchen, make coffee, and serve it? Eight years. My system, however, will go one better: it will be able to make a pot of the finest Broken Orange Pekoe and serve it. In the average suburban home? (No fair having the robot bring its own teabags, (or would it be loose tea and strainer?) or having a coffee machine built in, for that matter). It has to live off the land... Nope, no cheating. My assumptions are these. 1) A team size (very) approximately as follows: - Year 1: 10 - Year 2: 10 - Year 3: 100 - Year 4: 300 - Year 5: 800 - Year 6: 2000 - Year 7: 3000 - Year 8: 4000 2) Main Project(s) launched each year: - Year 1: AI software development environment - Year 2: AI software development environment - Year 3: Low-level cognitive mechanism experiments - Year 4: Global architecture experiments; Sensorimotor integration - Year 5: Motivational system and development tests - Year 6: (continuation of above) - Year 7: (continuation of above) - Year 8: Autonomous tests in real world situations The tests in Year 8 would be heavily supervised, but by that stage it should be possible for it to get on a bus, go to the suburban home, put the kettle on (if there was one: if not, go shopping to buy whatever supplies might be needed), then make the pot of tea (loose leaf of course: no robot of mine is going to be a barbarian tea-bag user) and serve it. FWIW, the average suburban home around here has coffee, but not tea. So you've now added the test of shopping in a local supermarket. I don't believe it. Not in eight years. It wouldn't be allowed past the cash register without human help. Note that this has nothing to do with how intelligent the system is. Maybe it would be intelligent enough, if it's environment were sane. But a robot? Either it would be seen as a Hollywood gimmick, or people would refuse to deal with it. Robots will first appear in controlled environments. Hospitals, home, stockrooms...other non-public-facing environments. (I'm excluding non-humanoid robots. Those, especially immobile forms, won't have the same level of resistance.) Well, I am not talking about the event proceeding without anyone noticing: I assume it will be done as a demonstration, so what the robot looks like will not matter. I imagine it would be followed by a press mob. The point is only whether the system could manage the problems involved in doing the shopping and then making the tea. And I think that other things will be happening at the same time anyway: I suspect that new medicines will already be coming out of the lab, from an immobile version of the same system. So if people are skittish about the tea-making robot, they will at least see that there are other, obviously beneficial products on the way. Really, though, the question is whether such a system could be built, from the technical point of view. My only point is that IF the resources were available, it could be done. That is based on my understanding of the timeline for my own project. 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=93208591-d770cb
[agi] History of MindForth
From the rewrite-in-progress of the User Manual -- 1.2 History of MindForth In the beginning was Mind.REXX on the Commodore Amiga, which the author Mentifex began coding in July of 1993, and publicizing in the Usenet comp.lang.rexx newsgroup. The late Pushpinder Singh of MIT sent e-mail expressing his amazement that anyone would try to do AI in REXX. Mentifex mailed back the entire Mind.REXX source code. Another fellow, an IBM mainframe programmer, tried to port the Amiga Rexxmind to run on his IBM mainframe -- which would have been a Kitty-Hawk-to-Concorde leap -- but the REXX AI code was not fit for IBM consumption. When Mind.REXX thought its first thought in late 1994, Mentifex posted news of the event in Usenet newsgroups for many of the most significant programming languages. Only the Forth community took up the AI challenge and expressed any interest in translating the AI program. A maker of Forth chips gave advice and counsel, and a maker of robots requested a copy of Mind.REXX for porting into the Forth in which he programmed his robots. Sorely disappointed at not having established a colony of AI Minds on IBM mainframes, Mentifex resolved to learn Forth on his own and assist in the porting of Mind.REXX into Mind.Forth for use in amateur robotics. Mentifex bought a copy of Starting Forth at a used book store and recorded his pilgrim's progress in the first volume of the Mind.Forth Programming Journal (MFPJ). The amateur robot-maker, a professional engineer, flew to Seattle on business with Boeing and visited Mentifex in his Vaierre apartment with a lesson on Forth coding. Another engineer, formerly with IBM and a REXX expert who had helped Mentifex in the coding of Mind.REXX AI, flew to the Bay area for a REXX conference at S.L.A.C. and was treated to dinner by the maker of Forth chips. Unfortunately, Mentifex did not try hard enough to learn Forth and the Forthmind project languished in 1996 and 1997 -- while Netizens were attacking Mentifex for daring to claim that he had developed a theory of mind for AI. It gradually dawned on Mentifex that in every Usenet newsgroup related to AI or robotics, there was always one fellow who considered himself the ultimate authority on the subject matter of the newsfroup, and woe unto anyone, especially an independent scholar like Mentifex, who dared to make an extraordinary scientific claim (ESC) on so grave a matter as announcing actual progress in AI. When the alpha male of comp.robotics.misc (a really cool guy, by the way) bracchiated over to Mentifex in the group in 1997 and launched an unseemingly vicious ad hominem attack, Mentifex knew not how to defend himself and was overcome with feelings of immense gratitude when the foxie Forth chip maker smote the troublemaker a mighty blow in defense of Mentifex. Forthwith Mentifex took up Forth again and devoted the entire year of 1998 to porting Mind.REXX into the native language of telescopes and robots -- Forth. In Mind.REXX, Mentifex had gone overboard in creating variables for even the slightest chance that they might turn out to be useful. Nobody had ever written a True AI before, it was all uncharted territory, and it seemed better to err on the side of too many variables rather than too few. In Forth, however, variables are anathema. Forthers prefer to put a value on the stack instead of in a variable. Mentifex never became a genuine, maniacally obsessive Forth programmer, but chose to program his AI in Forth code that looked enough like other languages to be easy to understand and to be easy to port from Forth. While Mentifex moved his AI coding efforts from MVP-Forth on the Amiga to F-PC on IBM clones and finally to Win32Forth, he also in 2001 (a space odyssey) suddenly ported MindForth into JavaScript so that users could just click on a link and have the Tutorial AI Mind flit across the 'Net and and take up albeit brief residence on their MSIE computer. While Push Singh was simply amazed at doing AI in REXX, many Netizens openly laughed and sneered at the idea of coding an AI Mind in JavaScript, which was not by any means a traditional AI language. Mentifex, however, suspected that his Mind.html in JavaScript was slowly building the largest installed user base of any AI program in the world, because it was so easy to save-to-disk the Mind.html code and because Site Meter logs reported the spread of the AI. Mentifex fell into the practice of switching back and forth between coding AI in JavaScript for a while and then in Forth. In March of 2005 Mentifex began coding powerful diagnostic routines into MindForth. He began to find and eliminate bugs that he could not deal with earlier because he had not even suspected their existence. Meanwhile, Mr. Frank J. Russo began to code what became http://AIMind-i.com -- a version of the Forthmind with its own site on the Web and with special abilities far beyond those of
Re: [agi] Wozniak's defn of intelligence
It seems we have different ideas about what AGI is. It is not a product that you can make and sell. It is a service that will evolve from the desire to automate human labor, currently valued at $66 trillion per year. I outlined a design in http://www.mattmahoney.net/agi.html It consists of lots of narrow specialists and an infrastructure for routing messages to the right experts. Nobody will control it or own it. I am not going to build it. It will be more complex than any human is capable of understanding. But there is enough economic incentive that it will be built in some form, regardless. The major technical obstacle is natural language modeling, which is required by the protocol. (Thus, my research in text compression). I realize that a full (Turing test) model can only be learned by having a full range of human experiences in a human body. But AGI is not about reproducing human form or human thinking. We used human servants in the past because that was what was available, not because it was the best solution. The problem is not to build a robot to pour your coffee. The problem is time, money, Maslow's hierarchy of needs. A solution could just as well be coffee from a can, ready to drink. --- J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Hmmm. I'd suspect you'd spend all your time and effort organizing the people. Orgs can grow that fast if they're grocery stores or something else the new hires already pretty much understand, but I don't see that happening smoothly in a pure research setting. I'd claim to be able to do it in 10 years with 30 people with the following provisos: 1. same 30 people the whole time 2. ten teams of 3: researcher, programmer, systems guy 3. all 30 have IQ 150 4. big hardware budget, all we build is software ... but I expect that the hardware for a usable body will be there in 10 years, so just buy it. Project looks like this: yrs 1-5: getting the basic learning algs worked out and running yrs 6-10: teaching the robot to walk, manipulate, balance, pour, understand kitchens, make coffee It's totally worthless to build a robot that had to be programmed to be able to make coffee. One that can understand how to do it by watching people do so, however, is absolutely the key to an extremely valuable level of intelligence. Josh On Friday 08 February 2008 11:46:51 am, Richard Loosemore wrote: My assumptions are these. 1) A team size (very) approximately as follows: - Year 1: 10 - Year 2: 10 - Year 3: 100 - Year 4: 300 - Year 5: 800 - Year 6: 2000 - Year 7: 3000 - Year 8: 4000 2) Main Project(s) launched each year: - Year 1: AI software development environment - Year 2: AI software development environment - Year 3: Low-level cognitive mechanism experiments - Year 4: Global architecture experiments; Sensorimotor integration - Year 5: Motivational system and development tests - Year 6: (continuation of above) - Year 7: (continuation of above) - Year 8: Autonomous tests in real world situations The tests in Year 8 would be heavily supervised, but by that stage it should be possible for it to get on a bus, go to the suburban home, put the kettle on (if there was one: if not, go shopping to buy whatever supplies might be needed), then make the pot of tea (loose leaf of course: no robot of mine is going to be a barbarian tea-bag user) and serve 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/?; -- Matt Mahoney, [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=94603346-a08d2f
Re: [agi] Wozniak's defn of intelligence
For the immediate future I think we are going to be seeing robots which are either directly programmed to perform tasks (expert systems on wheels) or which are taught by direct human supervision. In the human supervision scenario the robot is walked through a series of steps which it has to perform to complete a task. This could mean manually guiding its actuators, but the most practical way to do this is via teleoperation. So, after a a few supervised examples the robot is able to perform the same task autonomously, abstracting out variations in human performance. This type of training already goes on for industrial applications. Seegrid have a technology which they call walk through then work. Within the next ten years or so I think what we're going to see is this type of automation gradually moving into the consumer realm due to the falling price/performance ratio. This doesn't necessarily mean AGI in your home, but it does mean a lot of things will change. The idea that robotics is only about software is fiction. Good automation involves cooperation between software, electrical and mechanical engineers. In some cases problems are much better solved electromechanically than by software. For example, no matter how smart the software controlling it, a two fingered gripper will only be able to deal with a limited sub-set of manipulation tasks. Likewise a great deal of computation can be avoided by introducing variable compliance, and making clever use of materials to juggle energy around the system (biological creatures use these tricks all the time). Some aspects of the problem are within the realm of pure software, such as visual perception, navigation and mapping. Also, the idea that you can suspend real world testing until the end of the project is a recipe for disaster, unless your environment simulators are highly realistic, which at present involves substantial computing power. For more intelligent types of learning by imitation you really have to get into the business of mirror neurons, and ideas of selfhood. This means having the robot learn its own dynamics and being able to find mappings between these and the dynamics of objects which it observes. However, this can only be achieved if good perception systems are already developed and working. On 10/02/2008, J Storrs Hall, PhD [EMAIL PROTECTED] wrote: Hmmm. I'd suspect you'd spend all your time and effort organizing the people. Orgs can grow that fast if they're grocery stores or something else the new hires already pretty much understand, but I don't see that happening smoothly in a pure research setting. - 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=94603346-a08d2f
Re: [agi] Wozniak's defn of intelligence
Personally I would rather shoot for a world where the ever present nano-swarm saw that I wanted a cup of good coffee and effectively created one out of thin air on the spot, cup and all. Assuming I still took pleasure in such archaic practices and ways of changing my internal state of course. :-) I am not well qualified to give a good guess on the original question. But given the intersection of current progress in general environment comprehension and navigation, better robotic bodies, common sense databases, current task training by example and guesses on learning algorithm advancement I would be surprised if a robot with such ability was more than a decade out. - samantha - 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=94603346-a08d2f
Re: [agi] Wozniak's defn of intelligence
On 10/02/2008, Matt Mahoney [EMAIL PROTECTED] wrote: It seems we have different ideas about what AGI is. It is not a product that you can make and sell. It is a service that will evolve from the desire to automate human labor, currently valued at $66 trillion per year. Yes. I think the best way to think about the sort of robotics that we can reasonably expect to see in the near future is as physical artifacts which provide a service. Most robotics intelligence will be provided as remotely hosted services, because this means that you can build the physical machine very cheaply with minimal hardware onboard, and also to a large extent make it future-proof. It also enables the kinds of collective subconscious which Ben has talked about in the context of Second Life agents. As more computational intelligence comes online a dumb robot just subscribes to the new service (at a cost to the user, of course) and with no hardware changes it's suddenly smarter and able to do more stuff. - 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=94603346-a08d2f
Re: [agi] Wozniak's defn of intelligence
--- Mike Tintner [EMAIL PROTECTED] wrote: Matt: I realize that a full (Turing test) model can only be learned by having a full range of human experiences in a human body. Pray expand. I thought v. few here think that. Your definition seems to imply AGI must inevitably be embodied. It also implies an evolutionary model of embodied AGI - - a lower intelligence animal-level model will have to have a proportionately lower agility animal body. It also prompts the v. interesting speculation - (and has it ever been discussed on either forum?) - of what kind of superbody a superagi would have to have? (I would personally find *that* area of future speculation interesting if not super). Thoughts there too? No superhero fans around? A superagi would have billions of sensors and actuators all over the world -- keyboards, cameras, microphones, speakers, display devices, robotic manipulators, direct brain interfaces, etc. My claim is that an ideal language model (not AGI) requires human embodiment. But we don't need -- or want -- an ideal model. Turing realized that passing the imitation game requires duplicating human weaknesses as well as strengths. From his famous 1950 paper: Q: Please write me a sonnet on the subject of the Forth Bridge. A: Count me out on this one. I never could write poetry. Q: Add 34957 to 70764. A: (Pause about 30 seconds and then give as answer) 105621. Q: Do you play chess? A: Yes. Q: I have K at my K1, and no other pieces. You have only K at K6 and R at R1. It is your move. What do you play? A: (After a pause of 15 seconds) R-R8 mate. Why would we want to do that? We can do better. -- Matt Mahoney, [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=94603346-a08d2f
Re: [agi] What is MindForth?
On Feb 9, 2008 11:53 PM, A. T. Murray [EMAIL PROTECTED] wrote: It is not a chatbot. The AI engine is arguably the first True AI. It is immortal. Cool! What has it done to convince you that its truly intelligent? -J - 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=94603346-a08d2f
Re: [agi] What is MindForth?
Joseph Gentle wrote on Sun, 10 Feb 2008, in a message now at http://www.mail-archive.com/agi@v2.listbox.com/msg09803.html On Feb 9, 2008 11:53 PM, A. T. Murray [EMAIL PROTECTED] wrote: It is not a chatbot. The AI engine is arguably the first True AI. It is immortal. Cool! What has it done to convince you that its truly intelligent? -J Intelligent means understanding. When MindForth receives a sentence of English input (in the proper subject-verb-object format, for now), it understands the sentence by creating concept-nodes for the English words and by creating associative tags to link one concept to another. Thus the AI Mind knows the information asserted by the English sentence, and can include the asserted idea in its own thinking. Now for a miniature progress report on Mentifex AI. http://mind.sourceforge.net/audstm.html has been updated with a name-change to audSTM Auditory Short Term Memory module of free open-source MindForth True AI with the complete Table of Contents of the Mind.Forth User Manual listed at page-bottom with active URL-links. We shall see if happenstance websurfers decide to try out any of the AI features as listed in the Mind.Forth User Manual. Gentlemen, mesdames, brace yourselves for a ballooning Technological Singularity. ATM -- http://mind.sourceforge.net/mind4th.html http://mind.sourceforge.net/m4thuser.html - 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=94603346-a08d2f
Re: [agi] Wozniak's defn of intelligence
Matt: I realize that a full (Turing test) model can only be learned by having a full range of human experiences in a human body. Pray expand. I thought v. few here think that. Your definition seems to imply AGI must inevitably be embodied. It also implies an evolutionary model of embodied AGI - - a lower intelligence animal-level model will have to have a proportionately lower agility animal body. It also prompts the v. interesting speculation - (and has it ever been discussed on either forum?) - of what kind of superbody a superagi would have to have? (I would personally find *that* area of future speculation interesting if not super). Thoughts there too? No superhero fans around? - 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=94603346-a08d2f
