Re: [agi] The role of incertainty
This is similar to the thread I was workign on recently about Goals that didnt get quite as far as I would have liked either. 1. For use as testing metrics or for our personal goals of What an AGI should achieve, or what is important these goals or classes of problems should be defined as well as we possibly can. A couple obvious upper level classes are Navigation - in a real or virtual world. Natural Language - speaking, reading, talking Basic Problems Solving - given a simple, problem find a solution. 2. More specific for your AGI, What do you see the virtual pets doing? Specifically as end user functions for the consumer, the selling points you would give them, and how the AGI would help these functions. Is it going to be a rich enough situation in general to display more than just a blocks world type of pet, go fetch me the blue ball over there James Ratcliff Benjamin Goertzel [EMAIL PROTECTED] wrote: On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote:No, I keep saying - I'm not asking for the odd narrowly-defined task - but rather defining CLASSES of specific problems that your/an AGI will be able to tackle. Well, we have thought a lot about -- virtual agent control in simulation worlds (both pets and humanlike avatars) -- natural language question answering -- recognition of patterns in large bodies of scientific data Part of the definition task should be to explain how if you can solve one kind of problem, then you will be able to solve other distinct kinds. We can certainly explain that re Novamente, but IMO it is not the best way to get across how the system works to others with a technical interest in AGI. It may well be a useful mode of description for marketing purposes, however. 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/?; ___ James Ratcliff - http://falazar.com Looking for something... - Never miss an email again! Yahoo! Toolbar alerts you the instant new Mail arrives. Check it out. - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
2. More specific for your AGI, What do you see the virtual pets doing? Specifically as end user functions for the consumer, the selling points you would give them, and how the AGI would help these functions. Is it going to be a rich enough situation in general to display more than just a blocks world type of pet, go fetch me the blue ball over there James Ratcliff Well, it's a commercial project so I can't really talk about what the capabilities of the version 1.0 virtual pets will be. But the idea will be to start simple, and then incrementally roll out smarter and smarter versions. And, the idea is to make the pets flexible and adaptive learning systems, rather than just following fixed behavior patterns. One practical limitation is that we need to host a lot of pets on each server... However, we can do some borg mind stuff to work around this problem -- so that each pet retains its own personality, yet benefits from collective learning done basis on the totality of all the pets' memories... -- 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Ben Goertzel writes: Well, it's a commercial project so I can't really talk about what the capabilities of the version 1.0 virtual pets will be. I did spend a few evenings looking around Second Life. From that experience, I think that virtual protitutes would be a more profitable product :) - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Second Life also has a teen grid, by the way, which is not very active right now, but which virtual pets could enhance significantly. Virtual prostitutes are not in the plans anytime soon ;-) On 5/4/07, Derek Zahn [EMAIL PROTECTED] wrote: Ben Goertzel writes: Well, it's a commercial project so I can't really talk about what the capabilities of the version 1.0 virtual pets will be. I did spend a few evenings looking around Second Life. From that experience, I think that virtual protitutes would be a more profitable product :) - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Is there any already existing competition in this area - virtual adaptive pets - that we can look at? - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Friday, May 04, 2007 4:46 PM Subject: Re: [agi] The role of incertainty 2. More specific for your AGI, What do you see the virtual pets doing? Specifically as end user functions for the consumer, the selling points you would give them, and how the AGI would help these functions. Is it going to be a rich enough situation in general to display more than just a blocks world type of pet, go fetch me the blue ball over there James Ratcliff Well, it's a commercial project so I can't really talk about what the capabilities of the version 1.0 virtual pets will be. But the idea will be to start simple, and then incrementally roll out smarter and smarter versions. And, the idea is to make the pets flexible and adaptive learning systems, rather than just following fixed behavior patterns. One practical limitation is that we need to host a lot of pets on each server... However, we can do some borg mind stuff to work around this problem -- so that each pet retains its own personality, yet benefits from collective learning done basis on the totality of all the pets' memories... -- 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/?; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.467 / Virus Database: 269.6.2/785 - Release Date: 02/05/2007 14:16 - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On a less joking note, I think your ideas about applying your cognitive engine to NPCs in RPG type games (online or otherwise) could work out really well. The AI behind the game entities that are supposedly people is depressingly stupid, and games are a bazillion-dollar business. I hope your business direction works out well for you! - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
However, we can do some borg mind stuff to work around this problem -- so that each pet retains its own personality, yet benefits from collective learning done basis on the totality of all the pets' memories... Nice! - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Friday, May 04, 2007 11:46 AM Subject: Re: [agi] The role of incertainty 2. More specific for your AGI, What do you see the virtual pets doing? Specifically as end user functions for the consumer, the selling points you would give them, and how the AGI would help these functions. Is it going to be a rich enough situation in general to display more than just a blocks world type of pet, go fetch me the blue ball over there James Ratcliff Well, it's a commercial project so I can't really talk about what the capabilities of the version 1.0 virtual pets will be. But the idea will be to start simple, and then incrementally roll out smarter and smarter versions. And, the idea is to make the pets flexible and adaptive learning systems, rather than just following fixed behavior patterns. One practical limitation is that we need to host a lot of pets on each server... However, we can do some borg mind stuff to work around this problem -- so that each pet retains its own personality, yet benefits from collective learning done basis on the totality of all the pets' memories... -- 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/?; - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
You can take NARS (http://nars.wang.googlepages.com/) as an example, starting at http://nars.wang.googlepages.com/wang.logic_intelligence.pdf Pei On 5/1/07, rooftop8000 [EMAIL PROTECTED] wrote: It seems a lot of posts on this list are about the properties an AGI should have. PLURALISTIC, OPEN-ENDED AGI, adaptive, sometimes irrational .. it can be useful to talk about them, but i'd rather hear about how this translates into real projects. How to make a program that can deal with uncertainty and is adaptive and can think irrationally at times.. Seems like an awful lot of things.. how should we organize all this? How do we take existing solutions for some of these problems and make sure new ones can get added .. --- Mike Tintner [EMAIL PROTECTED] wrote: Yes, you are very right. And my point is that there are absolutely major philosophical issues here - both the general philosophy of mind and epistemology, and the more specific philosophy of AI. In fact, I think my characterisation of the issue as one of monism [general - behavioural as well as of substance] vs pluralism [again general - not just cultural] is probably the best one. So do post further thoughts, esp. re AI./AGI - this is well worth pursuing and elaborating. - Original Message - From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, April 30, 2007 3:31 PM Subject: [agi] The role of incertainty The discussion of uncertainty reminds me of a story about Piaget that struck a chord with me. Apparently, when Piaget was but a pup, he had the job of scoring tests given to kids. His job was to count the correct answers, but he started getting interested in the wrong answers. When he mentioned to his bosses that the wrong answers looked really interesting in their wrongness, they got made at him and pointed out that wrong was just wrong, and all they were interested in was how to make the kids get more right answers. At that point, P had a revelation: looking at right answers told him nothing about the children, whereas all the information about what they were really thinking was buried in the wrong answers. So he dumped his dead-end job and became Jean Piaget, Famous Psychologist instead. When I read the story I had a similar feeling of Aha! Thinking isn't about a lot of Right Thinking sprinkled with the occasional annoying Mistake. Thinking is actually a seething cauldron of Mistakes, some of which get less egregious over time and become Not-Quite-So-Bad Mistakes, which we call rational thinking. I think this attitude to how the mind works, though it is painted in bright colors, is more healthy than the attitude that thinking is about reasoning modulated by uncertainty. (Perhaps this is what irritates me so much about the people who call themselves Bayesians: people so desperate to believe that they are perfect that they have made a religion out of telling each other that they think perfectly, when in fact they are just as irrational as any other religious fanatic). ;-) 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/?; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.467 / Virus Database: 269.6.2/780 - Release Date: 29/04/2007 06:30 - 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/?; __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Pei, Glad to see your input. I noticed NARS quite by accident many years ago remembered it as pos. v. important. You certainly are implementing the principles we have just been discussing - which is exciting. However, reading your papers Ben's, it's becoming clear that there may well be an industry-wide bad practice going on here. You guys all focus on how your systems WORK... The first thing anyone trying to understand your or any other system must know is what does it DO? What are the problems it addresses, and the kinds of solutions it provides? It should be commonly accepted that it is EXTREMELY BAD PRACTICE not to first define what problems your system is set up to solve. Imagine if I spent 100 pages writing up these intricate mechanisms of this new machine, with all these wonderful new wireless and heat and electroservo this and that principles involved,.. and then only at the v. end do I tell you that it's an apple-peeler. You'd find it a bit of a strain to read all that. The only difference between the above write-up and yours and Ben's is that we the readers never even get to find out that what you've got is an apple-peeler! I still don't know what your systems do. It may be good for grants to cover up what you do, but it's actually not good for you or your thinking or the progress of AI. I'd very much like to know what your NARS system DOES - is that possible? P.S. Minsky is much the same. - Original Message - From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 12:50 PM Subject: Re: [agi] The role of incertainty You can take NARS (http://nars.wang.googlepages.com/) as an example, starting at http://nars.wang.googlepages.com/wang.logic_intelligence.pdf Pei On 5/1/07, rooftop8000 [EMAIL PROTECTED] wrote: It seems a lot of posts on this list are about the properties an AGI should have. PLURALISTIC, OPEN-ENDED AGI, adaptive, sometimes irrational .. it can be useful to talk about them, but i'd rather hear about how this translates into real projects. How to make a program that can deal with uncertainty and is adaptive and can think irrationally at times.. Seems like an awful lot of things.. how should we organize all this? How do we take existing solutions for some of these problems and make sure new ones can get added .. --- Mike Tintner [EMAIL PROTECTED] wrote: Yes, you are very right. And my point is that there are absolutely major philosophical issues here - both the general philosophy of mind and epistemology, and the more specific philosophy of AI. In fact, I think my characterisation of the issue as one of monism [general - behavioural as well as of substance] vs pluralism [again general - not just cultural] is probably the best one. So do post further thoughts, esp. re AI./AGI - this is well worth pursuing and elaborating. - Original Message - From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, April 30, 2007 3:31 PM Subject: [agi] The role of incertainty The discussion of uncertainty reminds me of a story about Piaget that struck a chord with me. Apparently, when Piaget was but a pup, he had the job of scoring tests given to kids. His job was to count the correct answers, but he started getting interested in the wrong answers. When he mentioned to his bosses that the wrong answers looked really interesting in their wrongness, they got made at him and pointed out that wrong was just wrong, and all they were interested in was how to make the kids get more right answers. At that point, P had a revelation: looking at right answers told him nothing about the children, whereas all the information about what they were really thinking was buried in the wrong answers. So he dumped his dead-end job and became Jean Piaget, Famous Psychologist instead. When I read the story I had a similar feeling of Aha! Thinking isn't about a lot of Right Thinking sprinkled with the occasional annoying Mistake. Thinking is actually a seething cauldron of Mistakes, some of which get less egregious over time and become Not-Quite-So-Bad Mistakes, which we call rational thinking. I think this attitude to how the mind works, though it is painted in bright colors, is more healthy than the attitude that thinking is about reasoning modulated by uncertainty. (Perhaps this is what irritates me so much about the people who call themselves Bayesians: people so desperate to believe that they are perfect that they have made a religion out of telling each other that they think perfectly, when in fact they are just as irrational as any other religious fanatic). ;-) 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
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: Pei, Glad to see your input. I noticed NARS quite by accident many years ago remembered it as pos. v. important. You certainly are implementing the principles we have just been discussing - which is exciting. However, reading your papers Ben's, it's becoming clear that there may well be an industry-wide bad practice going on here. You guys all focus on how your systems WORK... The first thing anyone trying to understand your or any other system must know is what does it DO? What are the problems it addresses, and the kinds of solutions it provides? Well, that is exactly the problem addressed in the paper I mentioned: my working definition of intelligence, and why I think it is a better understanding than the others. It should be commonly accepted that it is EXTREMELY BAD PRACTICE not to first define what problems your system is set up to solve. Agree. Imagine if I spent 100 pages writing up these intricate mechanisms of this new machine, with all these wonderful new wireless and heat and electroservo this and that principles involved,.. and then only at the v. end do I tell you that it's an apple-peeler. You'd find it a bit of a strain to read all that. Agree. The only difference between the above write-up and yours and Ben's is that we the readers never even get to find out that what you've got is an apple-peeler! I still don't know what your systems do. I wonder is you really read the paper I mentioned --- you can criticize it for all kinds of reasons, but you cannot say I didn't define the problem I'm working on, because that is what that paper is all about! If it is still not clear from that paper, you may also want to read http://nars.wang.googlepages.com/wang.AI_Definitions.pdf and http://nars.wang.googlepages.com/wang.WhatAIShouldBe.pdf It may be good for grants to cover up what you do, but it's actually not good for you or your thinking or the progress of AI. I'd very much like to know what your NARS system DOES - is that possible? I guess I don't understand what you mean by DOES. If you mean the goal of the project, then the above papers should be sufficient; if you mean how the system works, you need to try my demo at http://nars.wang.googlepages.com/nars%3Ademonstration ; if you mean what domain problems it can solve by design, then the answer is none, since it is not an expert system. Can you be more specific? Pei P.S. Minsky is much the same. - Original Message - From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 12:50 PM Subject: Re: [agi] The role of incertainty You can take NARS (http://nars.wang.googlepages.com/) as an example, starting at http://nars.wang.googlepages.com/wang.logic_intelligence.pdf Pei On 5/1/07, rooftop8000 [EMAIL PROTECTED] wrote: It seems a lot of posts on this list are about the properties an AGI should have. PLURALISTIC, OPEN-ENDED AGI, adaptive, sometimes irrational .. it can be useful to talk about them, but i'd rather hear about how this translates into real projects. How to make a program that can deal with uncertainty and is adaptive and can think irrationally at times.. Seems like an awful lot of things.. how should we organize all this? How do we take existing solutions for some of these problems and make sure new ones can get added .. --- Mike Tintner [EMAIL PROTECTED] wrote: Yes, you are very right. And my point is that there are absolutely major philosophical issues here - both the general philosophy of mind and epistemology, and the more specific philosophy of AI. In fact, I think my characterisation of the issue as one of monism [general - behavioural as well as of substance] vs pluralism [again general - not just cultural] is probably the best one. So do post further thoughts, esp. re AI./AGI - this is well worth pursuing and elaborating. - Original Message - From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, April 30, 2007 3:31 PM Subject: [agi] The role of incertainty The discussion of uncertainty reminds me of a story about Piaget that struck a chord with me. Apparently, when Piaget was but a pup, he had the job of scoring tests given to kids. His job was to count the correct answers, but he started getting interested in the wrong answers. When he mentioned to his bosses that the wrong answers looked really interesting in their wrongness, they got made at him and pointed out that wrong was just wrong, and all they were interested in was how to make the kids get more right answers. At that point, P had a revelation: looking at right answers told him nothing about the children, whereas all the information about what they were really thinking was buried in the wrong answers. So he dumped his dead-end job and became Jean Piaget, Famous Psychologist instead
Re: [agi] The role of incertainty
However, reading your papers Ben's, it's becoming clear that there may well be an industry-wide bad practice going on here. You guys all focus on how your systems WORK... The first thing anyone trying to understand your or any other system must know is what does it DO? What are the problems it addresses, and the kinds of solutions it provides? Hey Mike: What are the problems you address, and the kinds of solutions you provide? I could ask the same question about my 10 year old daughter ... or a newborn baby... My point is: an AGI system is by definition not restricted to a highly particular problem domain ... so the set of problems and solutions potentially addressable by any AGI system will be extremely broad Thinking in terms of incremental pathways to AGI, one may posit particular problem domains as targets for partial versions of one's AGI. But then there is a danger that people will see those interim problem domains and overgeneralize and believe that is what one's AGI system is all about. For instance, the first commercial manifestation of the Novamente AI Engine was the use of some of its learning routines inside the Biomind ArrayGenius product for gene expression microarray data analysis. So what? The next commercial manifestation may well be for controlling virtual pets operative within Second Life and/or other virtual worlds. Again, so what? This doesn't mean that Novamente is basically a virtual pet controller any more than it's basically a bioinformatics analysis tool. Once the end goal of AGI is reached, AGI systems will be able to do anything humans can do plus way more. And what AGI systems happen to be used to do on the incremental pathway there, doesn't really tell you that much about the ultimate nature of the AGI systems. (Similarly, e.g., the early applications that the Internet was used for in the 1970's don't really tell you much about the ultimate nature of the Internet. They tell you something of course, but they leave a lot out, as anyone can see now) -- 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/?member_id=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: Define the type of problems it addresses which might be [for all I know] *understanding and precis-ing a set of newspaper stories about politics or sewage *solving a crime of murder - starting with limited evidence *designing new types of buildings - starting from a limited knowledge of other types of buildings *navigate an agent through a domestic/ office environment with furniture strewn everywhere, animals and people milling around - limited rules for navigation... *learn to identify every plant and tree in a complex jungle scene - starting with limited knowledge and access to public databases Ideally I'd like to see in there, how having learned to solve one class of problems, it is going to solve new related classes of problems - to get an EDUCATION in various spheres ... (as distinct say from just learning in one sphere) - If one of the above problem is solved by an AGI system, it should be the result of learning of the system, rather than an innate capability built into the system. Furthermore, I don't think any of them is a necessary capability of an AGI system. If we are taking about possibility, I believe NARS has the potential for each of them, though the system isn't at that stage yet. I don't see why it is impossible, though I cannot show you that it has been done. so the navigational agent would have to be able, having mastered a domestic environment, to learn to navigate a jungle or forest environment. I want to see some PROBLEM ( education) EXAMPLES. That's all, really. (You say that your system is attending to all these different problems more or less simultaneously, or interweavingly, but you don't say what the problems are). Me, the general public, ( AGI people I suspect), have v. little idea of what AGI can actually do or is even trying to do right now I'm sorry if I make you disappointed --- I don't think any AGI has reached the stage of practical application. I don't want the general public to think that the AGI problem has been solved. Instead, I want they to think that AGI is an important and interesting research that should be supported, or at least tolerated. P.S. There are two sides to talking about knowledge/ intelligence/ problems etc. There is the side of the subject - the thinker, the brain manipulating ideas, the user of different techniques, logical, mathematical etc. bits, bytes etc And there is the side of the object(s) of knowledge - the crimes being solved, the buildings being constructed, the genes and society being learned about.. - what all that knowledge and those problems are about. There is the subjective side of the mirror reflecting and there is the objective side of the scene being reflected. You guys tend to be massively leaning over in describing everything in terms of the subjective side. But it's only when you describe intelligence problemsolving in terms of what you're trying to know/ solve problems about that things start to make sense. To me, this is a major reason that why we don't have a thinking machine yet --- people (both the general public and the mainstream AI researchers) believe that general intelligence can be reached by solving domain problems one after another. I won't try to convince you, but to point out that this belief is probably not as self-evident as many people assume. Before a good theory is established about AI, there won't be any good application. Pei - Original Message - From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 2:30 PM Subject: Re: [agi] The role of incertainty On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: Pei, Glad to see your input. I noticed NARS quite by accident many years ago remembered it as pos. v. important. You certainly are implementing the principles we have just been discussing - which is exciting. However, reading your papers Ben's, it's becoming clear that there may well be an industry-wide bad practice going on here. You guys all focus on how your systems WORK... The first thing anyone trying to understand your or any other system must know is what does it DO? What are the problems it addresses, and the kinds of solutions it provides? Well, that is exactly the problem addressed in the paper I mentioned: my working definition of intelligence, and why I think it is a better understanding than the others. It should be commonly accepted that it is EXTREMELY BAD PRACTICE not to first define what problems your system is set up to solve. Agree. Imagine if I spent 100 pages writing up these intricate mechanisms of this new machine, with all these wonderful new wireless and heat and electroservo this and that principles involved,.. and then only at the v. end do I tell you that it's an apple-peeler. You'd find it a bit of a strain to read all that. Agree. The only difference between the above write-up and yours and Ben's is that we the readers
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: Well, that really frustrates me. You just can't produce a machine that's going to work, unless you start with its goal/function. I think you are making an error of projecting the methodologies that are appropriate for narrow-purpose-specific machines, onto the quite different problem of designing AGIs... My colleagues at Novamente LLC have built plenty of purpose-specific software systems for customers, so it's not as though we're unable to work in the manner you're suggesting. We just find it inappropriate for the AGI task. The obvious and most basic type of adaptive problem it seems to me that agents/ robots should start with is navigational. Navigation IMO is a relatively narrow problem that can likely be solved by narrow-AI methods pretty effectively, without need for a really broad and robust AGI. So I don't view it as a great incremental problem for AGI. On the other hand, for instance Learning the rules of new games via communication with humans, and then being able to play these games effectively does seem to me like an appropriate incremental problem to orient one's work toward, on the gradual path toward AGI. However, I note that we will likely be approaching the navigation problem w/ Novamente during the next year, due to our intended business course of applying our proto-AGI system to control virtual agents in simulation worlds. -- 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/?member_id=231415user_secret=fabd7936
Re: [agi] The role of incertainty
IN the final analysis, Ben, you're giving me excuses rather than solutions. Your pet control program is a start - at least I have a vague, still v. vague idea of what you might be doing. You could (I'm guessing) say : this AGI is designed to control a pet which will have to solve adaptive problems like a) hide in surprising places within a complex environment and b) negotiate a complex environment strewn with obstacles and find new ways to destinations That more particular problem or type of problem, can then be generalised into vast classes of problems about agents finding new ways about complex environments - from searching buildings, to playing soccer or field games, to shopping in supermarkets or malls, etc. (In the end, you could probably generalise that class to include ALL problems period - including searching through information environments on the Net and in conversations). (Actually your mission statement should start the other way round - with the general class of problems/ activities you envisage - and then the particular examples that your AGI is going to concentrate on first.. and then indicate how you think it might progress). P.S. This is a truly weird conversation. It's like you're saying..Hell it's a box, why should I have to tell you what my box does? Only insiders care what's inside the box. The rest of the world wants to know what it does - and that's the only way they'll buy it and pay attention to it - and the only reason they should. Life's short. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 2:39 PM Subject: Re: [agi] The role of incertainty However, reading your papers Ben's, it's becoming clear that there may well be an industry-wide bad practice going on here. You guys all focus on how your systems WORK... The first thing anyone trying to understand your or any other system must know is what does it DO? What are the problems it addresses, and the kinds of solutions it provides? Hey Mike: What are the problems you address, and the kinds of solutions you provide? I could ask the same question about my 10 year old daughter ... or a newborn baby... My point is: an AGI system is by definition not restricted to a highly particular problem domain ... so the set of problems and solutions potentially addressable by any AGI system will be extremely broad Thinking in terms of incremental pathways to AGI, one may posit particular problem domains as targets for partial versions of one's AGI. But then there is a danger that people will see those interim problem domains and overgeneralize and believe that is what one's AGI system is all about. For instance, the first commercial manifestation of the Novamente AI Engine was the use of some of its learning routines inside the Biomind ArrayGenius product for gene expression microarray data analysis. So what? The next commercial manifestation may well be for controlling virtual pets operative within Second Life and/or other virtual worlds. Again, so what? This doesn't mean that Novamente is basically a virtual pet controller any more than it's basically a bioinformatics analysis tool. Once the end goal of AGI is reached, AGI systems will be able to do anything humans can do plus way more. And what AGI systems happen to be used to do on the incremental pathway there, doesn't really tell you that much about the ultimate nature of the AGI systems. (Similarly, e.g., the early applications that the Internet was used for in the 1970's don't really tell you much about the ultimate nature of the Internet. They tell you something of course, but they leave a lot out, as anyone can see now) -- 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/?; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.467 / Virus Database: 269.6.2/782 - Release Date: 01/05/2007 02:10 - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
P.S. This is a truly weird conversation. It's like you're saying..Hell it's a box, why should I have to tell you what my box does? Only insiders care what's inside the box. The rest of the world wants to know what it does - and that's the only way they'll buy it and pay attention to it - and the only reason they should. Life's short. Well, I am not trying to sell the Novamente Cognition Engine to the average Joe as ANYTHING, because it is not finished. When it is finished, I will still not try to sell it to the average Joe (or Mike ;-) as a purpose-specific product, because it is not one. What I will try to sell to people are purpose-specific products, such as virtual pets that they can train, or software systems they can use (if they're biologists) to find patterns in their data, etc. I understand that what people want to pay for, are purpose-specific products. However, what will enable the construction of a wide variety of purpose-specific products, is a general-purpose AGI engine... To use a rough analogy, suppose it was a long time ago and I was developing the world's first internal combustion engine. Then we could argue... Mike: What are you working on, Ben? Ben: I'm building an internal combustion engine Mike: What does it do? Ben: Well, it's a device in which rapid oxidation, of gas and air occurs in a confined space called a combustion chamber. This exothermic reaction of a fuel with an oxidizer creates gases of high temperature and pressure, which are permitted to expand. The defining feature of an internal combustion engine is that useful work is performed by the expanding hot gases acting directly to cause pressure, further causing movement of the piston inside the cylinder. Mike: What? Ben: Well, you burn stuff in a closed chamber and it makes pistons move up and down Mike: Oh. Well who the hell would want to buy something that does that? No one wants to watch pistons move up and down, at least not in my neck of the woods. Ben: Well you can use it for all sorts of different things Mike: Like what? Ben: Well, to power a car, or a locomotive, or an electrical generator ... or even a backpack helicopter. Maybe a robot. A lawnmower. Mike: Ok, so if you want to get your engine built, you need to set a specific goal. For instance, your goal could be to build a lawnmower. Ben: Well, that could be a good incremental goal -- to make a small version of my engine to power a lawnmower. But no particular goal is going to encapsulate all the applications of the engine. The main point is that I'm building an engine that lets you burn fuel and thus create mechanical work -- and this can be used for all sorts of different things. Mike: But, if you want people to buy it, you have to tell them what it will do for them. No one wants to buy a machine that sits in their livingroom and makes pistons bob up and down. Ben: Ok, look. This conversation is getting frustrating. I'm going to close the email window and get back to work. Mike: Darn, this conversation is getting frustrating. I don't want to buy a bunch of exothermic reactions, I want to buy something that does something specific for me. *** -- Ben You could ask them for the specific purpose of the generator, and they would say: Well, it can be used to power light bulbs, or computers, or cars, or refrigerators, or etc. etc. etc. But yet, none of these particular applications summarizes what it does. What it does is to generate electricity, which then can be used in a lot of applications. - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Nah, analogy doesn't quite work - though could be useful. An engine is used to move things... many different things - wheels, levers, etc. So if you've got an engine that is twenty times more powerful, sure you don't need to tell me what particular things it is going to move. It's generally accepted that it can move millions of things.. The difficulty here is that the problems to be solved by an AI or AGI machine are NOT accepted, well-defined. We cannot just take Pei's NARS, say, or NOvaemnte, and say well obviously it will apply to all these different kinds of problems. No doubt it will apply to many. But you have to explain. You have to classify the problems. Indeed, you will at some point be able to (or can already) describe different AI architectures almost as engines - but it's bringing all those problems together - which is a mixture of a psychological and philosophical problem. Background here: the fact that psychologists are still arguing about whether g exists - general intelligence - is a reflection of the difficulties here - the unsolved problems of defining problems. However those difficulties are not that great or insuperable. Not much point in arguing further here - all I can say now is TRY it - try focussing your work the other way round - I'm confident you'll find it makes life vastly easier and more productive. Defining what it does is just as essential for the designer as for the consumer. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 5:57 PM Subject: Re: [agi] The role of incertainty P.S. This is a truly weird conversation. It's like you're saying..Hell it's a box, why should I have to tell you what my box does? Only insiders care what's inside the box. The rest of the world wants to know what it does - and that's the only way they'll buy it and pay attention to it - and the only reason they should. Life's short. Well, I am not trying to sell the Novamente Cognition Engine to the average Joe as ANYTHING, because it is not finished. When it is finished, I will still not try to sell it to the average Joe (or Mike ;-) as a purpose-specific product, because it is not one. What I will try to sell to people are purpose-specific products, such as virtual pets that they can train, or software systems they can use (if they're biologists) to find patterns in their data, etc. I understand that what people want to pay for, are purpose-specific products. However, what will enable the construction of a wide variety of purpose-specific products, is a general-purpose AGI engine... To use a rough analogy, suppose it was a long time ago and I was developing the world's first internal combustion engine. Then we could argue... Mike: What are you working on, Ben? Ben: I'm building an internal combustion engine Mike: What does it do? Ben: Well, it's a device in which rapid oxidation, of gas and air occurs in a confined space called a combustion chamber. This exothermic reaction of a fuel with an oxidizer creates gases of high temperature and pressure, which are permitted to expand. The defining feature of an internal combustion engine is that useful work is performed by the expanding hot gases acting directly to cause pressure, further causing movement of the piston inside the cylinder. Mike: What? Ben: Well, you burn stuff in a closed chamber and it makes pistons move up and down Mike: Oh. Well who the hell would want to buy something that does that? No one wants to watch pistons move up and down, at least not in my neck of the woods. Ben: Well you can use it for all sorts of different things Mike: Like what? Ben: Well, to power a car, or a locomotive, or an electrical generator ... or even a backpack helicopter. Maybe a robot. A lawnmower. Mike: Ok, so if you want to get your engine built, you need to set a specific goal. For instance, your goal could be to build a lawnmower. Ben: Well, that could be a good incremental goal -- to make a small version of my engine to power a lawnmower. But no particular goal is going to encapsulate all the applications of the engine. The main point is that I'm building an engine that lets you burn fuel and thus create mechanical work -- and this can be used for all sorts of different things. Mike: But, if you want people to buy it, you have to tell them what it will do for them. No one wants to buy a machine that sits in their livingroom and makes pistons bob up and down. Ben: Ok, look. This conversation is getting frustrating. I'm going to close the email window and get back to work. Mike: Darn, this conversation is getting frustrating. I don't want to buy a bunch of exothermic reactions, I want to buy something that does something specific for me. *** -- Ben You could ask them for the specific purpose of the generator, and they would say: Well
Re: [agi] The role of incertainty
Not much point in arguing further here - all I can say now is TRY it - try focussing your work the other way round - I'm confident you'll find it makes life vastly easier and more productive. Defining what it does is just as essential for the designer as for the consumer. Focusing on making systems that can achieve narrowly-defined tasks is EXACTLY what the AI field has been doing for the last couple decades. Unsurprisingly, they have had some modest success at making systems that can achieve narrowly-defined tasks, and no success at moving toward artificial general intelligence. -- 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/?member_id=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: The difficulty here is that the problems to be solved by an AI or AGI machine are NOT accepted, well-defined. We cannot just take Pei's NARS, say, or NOvaemnte, and say well obviously it will apply to all these different kinds of problems. No doubt it will apply to many. But you have to explain. You have to classify the problems. We indeed have done that. What you suggested in exactly what I called Capability-AI in http://nars.wang.googlepages.com/wang.AI_Definitions.pdf . I agree that it is closer to many people's intuitive understanding to intelligence --- after all, we judge other people's intelligence by what practical problems they can solve. However, this understanding has serious limitation, as analyzed in the paper, as well as shown by the history of AI, since your idea is quite close to mainstream AI. Again, I'm not really trying to convince you, but to show you that if some AGI researchers don't do what you consider as obvious, they may have some consideration which cannot be simply rejected as obviously wrong. Pei Indeed, you will at some point be able to (or can already) describe different AI architectures almost as engines - but it's bringing all those problems together - which is a mixture of a psychological and philosophical problem. Background here: the fact that psychologists are still arguing about whether g exists - general intelligence - is a reflection of the difficulties here - the unsolved problems of defining problems. However those difficulties are not that great or insuperable. Not much point in arguing further here - all I can say now is TRY it - try focussing your work the other way round - I'm confident you'll find it makes life vastly easier and more productive. Defining what it does is just as essential for the designer as for the consumer. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 5:57 PM Subject: Re: [agi] The role of incertainty P.S. This is a truly weird conversation. It's like you're saying..Hell it's a box, why should I have to tell you what my box does? Only insiders care what's inside the box. The rest of the world wants to know what it does - and that's the only way they'll buy it and pay attention to it - and the only reason they should. Life's short. Well, I am not trying to sell the Novamente Cognition Engine to the average Joe as ANYTHING, because it is not finished. When it is finished, I will still not try to sell it to the average Joe (or Mike ;-) as a purpose-specific product, because it is not one. What I will try to sell to people are purpose-specific products, such as virtual pets that they can train, or software systems they can use (if they're biologists) to find patterns in their data, etc. I understand that what people want to pay for, are purpose-specific products. However, what will enable the construction of a wide variety of purpose-specific products, is a general-purpose AGI engine... To use a rough analogy, suppose it was a long time ago and I was developing the world's first internal combustion engine. Then we could argue... Mike: What are you working on, Ben? Ben: I'm building an internal combustion engine Mike: What does it do? Ben: Well, it's a device in which rapid oxidation, of gas and air occurs in a confined space called a combustion chamber. This exothermic reaction of a fuel with an oxidizer creates gases of high temperature and pressure, which are permitted to expand. The defining feature of an internal combustion engine is that useful work is performed by the expanding hot gases acting directly to cause pressure, further causing movement of the piston inside the cylinder. Mike: What? Ben: Well, you burn stuff in a closed chamber and it makes pistons move up and down Mike: Oh. Well who the hell would want to buy something that does that? No one wants to watch pistons move up and down, at least not in my neck of the woods. Ben: Well you can use it for all sorts of different things Mike: Like what? Ben: Well, to power a car, or a locomotive, or an electrical generator ... or even a backpack helicopter. Maybe a robot. A lawnmower. Mike: Ok, so if you want to get your engine built, you need to set a specific goal. For instance, your goal could be to build a lawnmower. Ben: Well, that could be a good incremental goal -- to make a small version of my engine to power a lawnmower. But no particular goal is going to encapsulate all the applications of the engine. The main point is that I'm building an engine that lets you burn fuel and thus create mechanical work -- and this can be used for all sorts of different things. Mike: But, if you want people to buy it, you have to tell them what it will do for them. No one wants to buy a machine that sits in their livingroom and makes pistons bob up and down. Ben: Ok, look. This conversation is getting frustrating. I'm going to close
Re: [agi] The role of incertainty
No, I keep saying - I'm not asking for the odd narrowly-defined task - but rather defining CLASSES of specific problems that your/an AGI will be able to tackle. Part of the definition task should be to explain how if you can solve one kind of problem, then you will be able to solve other distinct kinds. It's interesting - I'm not being in any way critical - that this isn't getting through. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 7:04 PM Subject: Re: [agi] The role of incertainty Not much point in arguing further here - all I can say now is TRY it - try focussing your work the other way round - I'm confident you'll find it makes life vastly easier and more productive. Defining what it does is just as essential for the designer as for the consumer. Focusing on making systems that can achieve narrowly-defined tasks is EXACTLY what the AI field has been doing for the last couple decades. Unsurprisingly, they have had some modest success at making systems that can achieve narrowly-defined tasks, and no success at moving toward artificial general intelligence. -- 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/?; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.467 / Virus Database: 269.6.2/782 - Release Date: 01/05/2007 02:10 - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: No, I keep saying - I'm not asking for the odd narrowly-defined task - but rather defining CLASSES of specific problems that your/an AGI will be able to tackle. Well, we have thought a lot about -- virtual agent control in simulation worlds (both pets and humanlike avatars) -- natural language question answering -- recognition of patterns in large bodies of scientific data Part of the definition task should be to explain how if you can solve one kind of problem, then you will be able to solve other distinct kinds. We can certainly explain that re Novamente, but IMO it is not the best way to get across how the system works to others with a technical interest in AGI. It may well be a useful mode of description for marketing purposes, however. 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/?member_id=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On 5/1/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: No, I keep saying - I'm not asking for the odd narrowly-defined task - but rather defining CLASSES of specific problems that your/an AGI will be able to tackle. Well, we have thought a lot about -- virtual agent control in simulation worlds (both pets and humanlike avatars) -- natural language question answering -- recognition of patterns in large bodies of scientific data and, math theorem proving... - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
I think if you look at the history of most industries, you'll find that it often takes a long time for them to move from becoming producer-centric to consumer-centric. [There are some established terms for this, wh. I've forgotten]. When making things people are often first preoccupied with the tools and the machinery, rather than the ultimate function. And producers are often extremely resistant to looking at things from the other POV. As I said to Ben, the crucial cultural background here is that intelligence and creativity have not been properly defined in any sphere. There is no consensus about types of problems, about the difference between AI and AGI, or, more crucially, between divergent and convergent intelligence, etc. etc. So I don't agree that you can assume that a given AI architecture or system will be able to solve a whole set of problems. And a large part of my point is that the question what does it do? SHOULD be obvious, but isn't because there's clearly a whole producer-centric culture within AI/AGI of ignoring it. P.S. I think I see that one problem I'm having communicating with both you Ben, is that you're both working within a fading dichotomy of AI -specific well-defined problems vs AGI - general problem-solving which supposedly doesn't have to be defined (and keep pushing me into the AI camp). I'm saying you do have to define what your AGI will do - but define it as a tree - 1) a general class of problems - supported by 2) examples of specific types of problem within that class. I'm calling for something different to the traditional alternatives here. I doubt that anyone is doing much thinking about general CLASSES of problems. I've been trying to do it in my posts . - Original Message - From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 7:08 PM Subject: Re: [agi] The role of incertainty On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: The difficulty here is that the problems to be solved by an AI or AGI machine are NOT accepted, well-defined. We cannot just take Pei's NARS, say, or NOvaemnte, and say well obviously it will apply to all these different kinds of problems. No doubt it will apply to many. But you have to explain. You have to classify the problems. We indeed have done that. What you suggested in exactly what I called Capability-AI in http://nars.wang.googlepages.com/wang.AI_Definitions.pdf . I agree that it is closer to many people's intuitive understanding to intelligence --- after all, we judge other people's intelligence by what practical problems they can solve. However, this understanding has serious limitation, as analyzed in the paper, as well as shown by the history of AI, since your idea is quite close to mainstream AI. Again, I'm not really trying to convince you, but to show you that if some AGI researchers don't do what you consider as obvious, they may have some consideration which cannot be simply rejected as obviously wrong. Pei Indeed, you will at some point be able to (or can already) describe different AI architectures almost as engines - but it's bringing all those problems together - which is a mixture of a psychological and philosophical problem. Background here: the fact that psychologists are still arguing about whether g exists - general intelligence - is a reflection of the difficulties here - the unsolved problems of defining problems. However those difficulties are not that great or insuperable. Not much point in arguing further here - all I can say now is TRY it - try focussing your work the other way round - I'm confident you'll find it makes life vastly easier and more productive. Defining what it does is just as essential for the designer as for the consumer. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 5:57 PM Subject: Re: [agi] The role of incertainty P.S. This is a truly weird conversation. It's like you're saying..Hell it's a box, why should I have to tell you what my box does? Only insiders care what's inside the box. The rest of the world wants to know what it does - and that's the only way they'll buy it and pay attention to it - and the only reason they should. Life's short. Well, I am not trying to sell the Novamente Cognition Engine to the average Joe as ANYTHING, because it is not finished. When it is finished, I will still not try to sell it to the average Joe (or Mike ;-) as a purpose-specific product, because it is not one. What I will try to sell to people are purpose-specific products, such as virtual pets that they can train, or software systems they can use (if they're biologists) to find patterns in their data, etc. I understand that what people want to pay for, are purpose-specific products. However, what will enable the construction of a wide variety of purpose-specific products, is a general-purpose AGI engine... To use
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: No, I keep saying - I'm not asking for the odd narrowly-defined task - but rather defining CLASSES of specific problems that your/an AGI will be able to tackle. Part of the definition task should be to explain how if you can solve one kind of problem, then you will be able to solve other distinct kinds. Did nature have a specific task in mind when our brains evolved? Much like an AGI, we as humans are capable of doing MANY things. To sum it up, AGI could be described as a machine that is capable of using pattern recognition, classification, and analysis to produce better pattern recognition, classification and analysis systems for itself. The results of this apply to every problem that could ever be asked to solve. The traditional approach to AI is to do exactly what you're asking: solve individual problems and build them up until we have something that, on every observable level, is equivalent to a thinking person. For the last 50 years, this hasn't produced any promising results in terms of cognition. It's interesting - I'm not being in any way critical - that this isn't getting through. -- Josh Treadwell [EMAIL PROTECTED] 480-206-3776 - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: As I said to Ben, the crucial cultural background here is that intelligence and creativity have not been properly defined in any sphere. There is no consensus about types of problems, about the difference between AI and AGI, or, more crucially, between divergent and convergent intelligence, etc. etc. So I don't agree that you can assume that a given AI architecture or system will be able to solve a whole set of problems. I don't assume that, and that is exactly why I listed five different working definitions in the paper I mentioned, and argued that no one can replace the other completely. And a large part of my point is that the question what does it do? SHOULD be obvious, but isn't because there's clearly a whole producer-centric culture within AI/AGI of ignoring it. As this debate shows, what is considered as obvious by different people are obviously different. ;-) P.S. I think I see that one problem I'm having communicating with both you Ben, is that you're both working within a fading dichotomy of AI -specific well-defined problems vs AGI - general problem-solving which supposedly doesn't have to be defined (and keep pushing me into the AI camp). I'm saying you do have to define what your AGI will do - but define it as a tree - 1) a general class of problems - supported by 2) examples of specific types of problem within that class. I'm calling for something different to the traditional alternatives here. I doubt that anyone is doing much thinking about general CLASSES of problems. I've been trying to do it in my posts . I have made it very clear in my papers why I don't want to define intelligence by the set of practical problems it can solve. It is fine for you to disagree, but at least you should try to see why I take this position before claiming it to be wrong for obvious reasons. Pei - Original Message - From: Pei Wang [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 7:08 PM Subject: Re: [agi] The role of incertainty On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: The difficulty here is that the problems to be solved by an AI or AGI machine are NOT accepted, well-defined. We cannot just take Pei's NARS, say, or NOvaemnte, and say well obviously it will apply to all these different kinds of problems. No doubt it will apply to many. But you have to explain. You have to classify the problems. We indeed have done that. What you suggested in exactly what I called Capability-AI in http://nars.wang.googlepages.com/wang.AI_Definitions.pdf . I agree that it is closer to many people's intuitive understanding to intelligence --- after all, we judge other people's intelligence by what practical problems they can solve. However, this understanding has serious limitation, as analyzed in the paper, as well as shown by the history of AI, since your idea is quite close to mainstream AI. Again, I'm not really trying to convince you, but to show you that if some AGI researchers don't do what you consider as obvious, they may have some consideration which cannot be simply rejected as obviously wrong. Pei Indeed, you will at some point be able to (or can already) describe different AI architectures almost as engines - but it's bringing all those problems together - which is a mixture of a psychological and philosophical problem. Background here: the fact that psychologists are still arguing about whether g exists - general intelligence - is a reflection of the difficulties here - the unsolved problems of defining problems. However those difficulties are not that great or insuperable. Not much point in arguing further here - all I can say now is TRY it - try focussing your work the other way round - I'm confident you'll find it makes life vastly easier and more productive. Defining what it does is just as essential for the designer as for the consumer. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 5:57 PM Subject: Re: [agi] The role of incertainty P.S. This is a truly weird conversation. It's like you're saying..Hell it's a box, why should I have to tell you what my box does? Only insiders care what's inside the box. The rest of the world wants to know what it does - and that's the only way they'll buy it and pay attention to it - and the only reason they should. Life's short. Well, I am not trying to sell the Novamente Cognition Engine to the average Joe as ANYTHING, because it is not finished. When it is finished, I will still not try to sell it to the average Joe (or Mike ;-) as a purpose-specific product, because it is not one. What I will try to sell to people are purpose-specific products, such as virtual pets that they can train, or software systems they can use (if they're biologists) to find patterns in their data, etc. I understand that what people want
Re: [agi] The role of incertainty
I'm saying you do have to define what your AGI will do - but define it as a tree - 1) a general class of problems - supported by 2) examples of specific types of problem within that class. I'm calling for something different to the traditional alternatives here. I doubt that anyone is doing much thinking about general CLASSES of problems. I've been trying to do it in my posts . I understand the approach you're advocating, and I certainly **could** take it in regard to Novamente, I just don't really see any great value in taking such an approach. It wouldn't cause us to do our work any differently. Maybe it would be useful for better communicating our work to certain people, such as you, though ;-) -- 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Well, you see I think only the virtual agent problems are truly generalisable. The others it strikes me, haven't got a hope of producing AGI, and are actually narrow. But as I said, the first can probably be generalised in terms of agents seeking goals within problematic environments - and you can see how, in principle, at any rate, an AGI that mastered some such problems, might go on to master related but nevertheless very different problems re very different environments. So, I repeat, - thinking in terms of classes of problems will help you the producer and not just the consumer - it will help you, I would argue, focus your efforts on where they are most likely to be rewarding. It also involves a different kind of thinking, in my impression, than you have actually been employing - and if that's true, play with it a lot before rejecting it. But no need to take this further - although if you do want to explore actual classes of problems further as such, I'm still open to that. Been good talking to you. - Original Message - From: Benjamin Goertzel To: agi@v2.listbox.com Sent: Tuesday, May 01, 2007 7:32 PM Subject: Re: [agi] The role of incertainty On 5/1/07, Benjamin Goertzel [EMAIL PROTECTED] wrote: On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: No, I keep saying - I'm not asking for the odd narrowly-defined task - but rather defining CLASSES of specific problems that your/an AGI will be able to tackle. Well, we have thought a lot about -- virtual agent control in simulation worlds (both pets and humanlike avatars) -- natural language question answering -- recognition of patterns in large bodies of scientific data and, math theorem proving... -- 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/?; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.467 / Virus Database: 269.6.2/782 - Release Date: 01/05/2007 02:10 - 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
On 5/1/07, Mike Tintner [EMAIL PROTECTED] wrote: Well, you see I think only the virtual agent problems are truly generalisable. The others it strikes me, haven't got a hope of producing AGI, and are actually narrow. I think they are all generalizable in principle, but the virtual agents one is the easiest one to do in a generalizable way... But as I said, the first can probably be generalised in terms of agents seeking goals within problematic environments - and you can see how, in principle, at any rate, an AGI that mastered some such problems, might go on to master related but nevertheless very different problems re very different environments. Agreed 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=231415user_secret=fabd7936
Re: [agi] The role of incertainty
Yes, you are very right. And my point is that there are absolutely major philosophical issues here - both the general philosophy of mind and epistemology, and the more specific philosophy of AI. In fact, I think my characterisation of the issue as one of monism [general - behavioural as well as of substance] vs pluralism [again general - not just cultural] is probably the best one. So do post further thoughts, esp. re AI./AGI - this is well worth pursuing and elaborating. - Original Message - From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, April 30, 2007 3:31 PM Subject: [agi] The role of incertainty The discussion of uncertainty reminds me of a story about Piaget that struck a chord with me. Apparently, when Piaget was but a pup, he had the job of scoring tests given to kids. His job was to count the correct answers, but he started getting interested in the wrong answers. When he mentioned to his bosses that the wrong answers looked really interesting in their wrongness, they got made at him and pointed out that wrong was just wrong, and all they were interested in was how to make the kids get more right answers. At that point, P had a revelation: looking at right answers told him nothing about the children, whereas all the information about what they were really thinking was buried in the wrong answers. So he dumped his dead-end job and became Jean Piaget, Famous Psychologist instead. When I read the story I had a similar feeling of Aha! Thinking isn't about a lot of Right Thinking sprinkled with the occasional annoying Mistake. Thinking is actually a seething cauldron of Mistakes, some of which get less egregious over time and become Not-Quite-So-Bad Mistakes, which we call rational thinking. I think this attitude to how the mind works, though it is painted in bright colors, is more healthy than the attitude that thinking is about reasoning modulated by uncertainty. (Perhaps this is what irritates me so much about the people who call themselves Bayesians: people so desperate to believe that they are perfect that they have made a religion out of telling each other that they think perfectly, when in fact they are just as irrational as any other religious fanatic). ;-) 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/?; -- No virus found in this incoming message. Checked by AVG Free Edition. Version: 7.5.467 / Virus Database: 269.6.2/780 - Release Date: 29/04/2007 06:30 - 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=231415user_secret=fabd7936