Re: [agi] Lossy ** lossless compressi
However, I think that a lossless model can reasonably derive this information by observing that p(x, x') is approximately equal to p(x) or p(x'). In other words, knowing both x and x' does not tell you any more than x or x' alone, or CDM(x, x') ~ 0.5. I think this is a reasonable way to model lossy behavior in humans. How does a lossless model observe that "Jim is extremely fat" and "James continues to be morbidly obese"are approximately equal? I would assume that it would have to be via the same world model that a lossy model would -- which is wy above the bitstream level. Also, I think that going at this via a probability model is not the way to go. knowing both x and x' does not tell you any more than x or x' alone Can't you rephrase this with the following approximately equal phrases: You need to discard either x or x' to reach a canonical form, or Discarding either x or x' is not a lossy operation? Mark - Original Message - From: "Matt Mahoney" [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Sunday, August 27, 2006 10:32 PM Subject: Re: [agi] Lossy ** lossless compressi In showing that compression implies AI, I first make the simplifying assumption that everyone shares the same language model. Then I relax that assumption and argue that this makes it easier for a machine to pass the Turing test. But I see your point. I argued that a lossless model knows everything that a lossy model does, plus more, because the lossless model knows p(x) and p(x'), while a lossy model only knows p(x) + p(x'). However I missed that the lossy model knows that x and x' are equivalent, while the lossless model does not. However, I think that a lossless model can reasonably derive this information by observing that p(x, x') is approximately equal to p(x) or p(x'). In other words, knowing both x and x' does not tell you any more than x or x' alone, or CDM(x, x') ~ 0.5. I think this is a reasonable way to model lossy behavior in humans. -- Matt Mahoney, [EMAIL PROTECTED] - Original Message From: Philip Goetz [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Sunday, August 27, 2006 9:23:25 PM Subject: Re: [agi] Lossy ** lossless compressi On 8/25/06, Matt Mahoney [EMAIL PROTECTED] wrote: As I stated earlier, the fact that there is normal variation in human language models makes it easier for a machine to pass the Turing test. However, a machine with a lossless model will still outperform one with a lossy model because the lossless model has more knowledge. That would be true only if there were one correct language model, AND you knew what it was. Besides which, every human has a lossy model. It seems to me that by your argument, a machine with a lossless model would "out-perform" a human, and thus /fail/ the Turing test. - Phil --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]--- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Lossy ** lossless compressi
On 8/28/06, Mark Waser [EMAIL PROTECTED] wrote: How does a lossless model observe that Jim is extremely fat and James continues to be morbidly obese are approximately equal? Actually I think I just may have invented one possible way to do that using a lossless probabilistic model in my previous email to this list. Did you read it? Anyway, in case you have a hard time figuring it out all by yourself, the idea behind it can be pretty straigthforwardly generalized to be used with phrases and thus I think phrases can be observed to be approximately equal if they can occur in pretty much identical contexts and affect the distribution of following words and phrases approximately equivalently. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, Stephen Reed [EMAIL PROTECTED] wrote: An assumption of mine that can be debated perhaps in aseparate message thread, is that there should beeffectively only one AGI, allowing for a federation ofAGI's contrived to prevent war between them. I've explained my opinion of the various AI conquer the world memes before, this probably isn't the thread for a repeat :) However, I think there might be something to be said for this idea for a completely different reason. OpenOffice works well under the GPL because of what it does: everyone puts their own little OpenOffice installation on their own PC and uses it for individual-scale tasks, and the GPL lets them do this. Google wouldn't work at all well under the GPL. Why? Because if everyone had their own little Google, it would be quite useless [1]. The system's usefulness comes from the fact that there is only one Google, and it is _big_, in terms of both knowledge and the computing resources to use that knowledge. A serious AGI will have to end up making Google look like those '10 PRINT HELLO: GOTO 10' programs we used to write on our childhood 8-bit computers. If everyone just downloads their own copy and tweaks it separately from everyone else's, the sum total of value generated will be effectively zero. Now, I don't think I'd have the license say you must donate CPU cycles in payment for using this both because I don't see any way to enforce it and because it would justifiably annoy people who want to e.g. play with it on a laptop without an Internet connection. What I would consider doing (haven't thought very much about this so far, might be flaws in the idea, but I think it's at least worth a look) if I were going to do open source AGI, is take an idea from GPL and say: You may do anything you like with this on your own PC, but you may not _distribute_ an incompatible version. Any modified version that gets distributed, must seamlessly hook into the network of other copies. [1] I know there are some companies that have licensed Google to catalog their intranet stuff, but this is small change by comparison, and even this doesn't apply to AGI because the latter's knowledge acquisition will be far less regular and at least for the foreseeable future far less fully automatable than Google's. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
--- Russell Wallace [EMAIL PROTECTED] wrote: A serious AGI will have to end up making Google look like those '10 PRINT HELLO: GOTO 10' programs we used to write on our childhood 8-bit computers. Agreed. If everyone just downloads their own copy and tweaks it separately from everyone else's, the sum total of value generated will be effectively zero. Yes, but suppose the government of China decides to download an open source AGI and install it on one or more of their Top 500 supercomputer facilities? Certain indivuals may have vast compute resources at their disposal. Now, I don't think I'd have the license say you must donate CPU cycles in payment for using this both because I don't see any way to enforce it and because it would justifiably annoy people who want to e.g. play with it on a laptop without an Internet connection. Let me elaborate: I am considering a primary deployment for my open source project using Jabber, an open source chat protocal adopted by Google Chat among others. Any user could just chat with the AGI. If they want the AGI to perform tasks for them on their own computer, then they would download some components and be subject to the constraints of the license. From public research soliciations I know that the US CIA is interested in AI to assist their intelligence analysts. If they were to download and intall an open source AGI they would permit policy control from the unclassified side, e.g. not violate US laws, but no information could come out of a classified software component back to the central, completely open AGI. What I would consider doing (haven't thought very much about this so far, might be flaws in the idea, but I think it's at least worth a look) if I were going to do open source AGI, is take an idea from GPL and say: You may do anything you like with this on your own PC, but you may not _distribute_ an incompatible version. Any modified version that gets distributed, must seamlessly hook into the network of other copies. Exactly, how to detect evil modifications is a safety issue. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, Stephen Reed [EMAIL PROTECTED] wrote: Yes, but suppose the government of China decides todownload an open source AGI and install it on one ormore of their Top 500 supercomputer facilities? Suppose the government of China decide to get hold of CAD, simulation software etc and install it on their computers and use it for designing bombs and missiles? Well then they can do that, and indeed doubtless they have. Same answer. How do we regulate the use of computers for nefarious deeds today? Well mostly we don't, and when we try (e.g. the American government with cryptography), it doesn't work and the attempt does far more harm than good. What we do, by and large, is not bother - instead, we just outlaw the nefarious deeds themselves, regardless of the tools that were used. I think this will remain true in the future (or if it doesn't, we'll be in big trouble). That having been said, if you're serious about preventing the abuse of your software, I think the only answer is, don't distribute it. Follow the path of Novamente and indeed Google themselves (albeit for different reasons) and keep the software on your own machines and sell the services it provides. That might be a better route with regard to resources. It's not clear to me whether an open source AGI project relying on donated manpower and computing power could obtain enough of those. Then again, maybe it could; I don't really know either way. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, Stephen Reed [EMAIL PROTECTED] wrote: I assume that you fully understand the benefits andbusiness case of an open source project, and that yourpoint is made even with the former fully considered. Yes. For that matter, my answer would be the same if you proposed a closed source project that sold a binary distribution like Microsoft Office. I would respond to the proprietary AGI alternativewith the observation that one may suppose, as do I, that only one AGI is safer than many, possiblyopposing, AGIs.With the proprietary model, therewill be a market for others to enter.On the otherhand an established open source project precludescompetition, e.g. only one Wikipedia. I think safety is maximized by maximizing the probability of successful development of AGI within whatever time we have available rather than trying to minimize the probability that one or more AGIs will be abused, but that is a different question. If minimizing the probability that an AGI will be abused is your priority, the best approach might be to try to get there first and remain so far ahead of the competition as to have a near monopoly, as e.g. IBM did in the mainframe market in its heyday. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 28/08/06, Russell Wallace [EMAIL PROTECTED] wrote: On 8/28/06, Stephen Reed [EMAIL PROTECTED] wrote: Google wouldn't work at all well under the GPL. Why? Because if everyone had their own little Google, it would be quite useless [1]. The system's usefulness comes from the fact that there is only one Google, and it is _big_, in terms of both knowledge and the computing resources to use that knowledge. But google gets its knowledge from lots of little actors (web page makers). I suspect the thing that will replace google will get its information from lots of little AIs each attached to a person/government or other organisation. While AGI will likely be a google replacer, it will also be an outlook replacer as well. The micro scale and the macro. If the macro AGI can't translate between differences in language or representation that the micro AGIs have acquired from being open source, then we probably haven't done our job properly. Will Pearson --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 28/08/06, Russell Wallace [EMAIL PROTECTED] wrote: On 8/28/06, William Pearson [EMAIL PROTECTED] wrote: If the macro AGI can't translate between differences in language or representation that the micro AGIs have acquired from being open source, then we probably haven't done our job properly. But I don't think that will. I think that job is impossible to do, or rather that doing it would require a complete, fully-educated AGI - which is precisely what we are trying to achieve, so we can't rely on its existence while we are trying to build it. I was thinking more long term than you. I agree in the first phase we can't rely on it being to translate different information from different AGI. But to start with I wouldn't attempt the google killer, merely the outlook killer. We may well not have enough computing resources available to do it on the cheap using local resources. But that is the approach I am inclined to take, I'll just wait until we do. The open source distibuted google killer will have the problem of who decides what goals the system has/starts with (depending upon your philosophy), and how to upgrade the collective if the goals were incorrect to start with. It is also not as amenable to experiment as the micro level systems are. Will --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, William Pearson [EMAIL PROTECTED] wrote: I was thinking more long term than you. I agree in the first phase wecan't rely on it being to translate different information fromdifferent AGI. But to start with I wouldn't attempt the google killer,merely the outlook killer. Okay, but... We may well not have enough computing resources available to do it onthe cheap using local resources. But that is the approach I am inclined to take, I'll just wait until we do. Computing power isn't the only issue, and probably not the most important one; what do you think an Outlook killer could do that Outlook doesn't already do, and how would it know how to do it?The open source distibuted google killer will have the problem of who decides whatgoals the system has/starts with (depending upon your philosophy) Do what the users want you to do. andhow to upgrade the collective if the goals were incorrect to startwith. In the case of an open source AGI project, there would be no requirement that all users form a collective as far as their goals are concerned, only that they agree on running, maintaining and enhancing the software to serve their separate goals, just as is the case with e.g. the Internet today. It is also not as amenable to experiment as the micro levelsystems are. True. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
Hi Stephen, As a small operation independent of Cyc, distributing your AGI system as open source is likely to be a good strategy. As a small university PI developing visualization software, distributing my systems as open source turned out to be very good for my project. Our collaborators and customers came to us out of the blue, without any need for a marketing department we couldn't afford. There were very few negatives (we did have a developer in China who wanted me to pay them to get access to some enhanccements they had made, but I simply declined). In my first major system, Vis5D, there was a problem with divergent versions. We were able to work with developers to unify the most important versions, but it was hard work and there were still numerous divergent versions. For another major system, VisAD, I specifically designed it a high level of abstraction and with classes designed to be extended to allow developers to make low level changes, and so far there have not been divergent versions. I am a bit skeptical whether legal wording in the license will restrain developers from making divergent versions - as a small operation, are you really prepared to take violators to court? But if your design makes divergence less necesary, most developers will see the advantage of a unified version that permits sharing. By open source distribution you are expressing optimism about human nature, and your developer community will mostly justify that optimism. The best approach for the few who disappoint you is to simply ignore them. Good luck, Bill On Mon, 28 Aug 2006, Stephen Reed wrote: I would appreciate comments regarding additional constraints, if any, that should be applied to a traditional open source license to achieve a free but safe widespread distribution of software that may lead to AGI. As background, I was recently layed off by Cycorp, the creators of the Cyc knowledge base, and I am taking this opportunity to pursue my own AGI ideas full time. Although I am a ResearchCyc licensee I am considering a roadmap leading to a completely open source AGI. An assumption that some may challenge is that AGI software should be free in the first place. I think that this approach has proved useful for both software (e.g. MySQL database) and knowledge (Wikipedia). Could additional terms and conditions for an AGI open source license retain these benefits yet be safe? The leading GNU Public License forbids any further constraints beyond its own terms so lets think about the Apache Software License (ASL). http://www.apache.org/licenses/LICENSE-2.0.html Here is a key clause: Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare Derivative Works of, publicly display, publicly perform, sublicense, and distribute the Work and such Derivative Works in Source or Object form. An assumption of mine that can be debated perhaps in a separate message thread, is that there should be effectively only one AGI, allowing for a federation of AGI's contrived to prevent war between them. A second assumption is that the existing legal structure, in particular license enforcement throughout the world, can handle an open source AGI. What about an AGI open source license, similar to the above ASL in which the user must, to comply with the license, federate their downloaded AGI with the existing AGI system and thus subordinate it to ethic, legal and safety controls previously established? Governance of a open source distributed AGI, with users who could be citizens of enemy countries, is an issue that might be addressed by license terms and conditions - any thoughts? Cheers. -Steve __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, Bill Hibbard [EMAIL PROTECTED] wrote: By open source distribution you are expressing optimismabout human nature, and your developer community willmostly justify that optimism. The best approach for thefew who disappoint you is to simply ignore them. I agree. When I suggested a no incompatible versions clause in the license, I wasn't thinking in terms of then you can fight and win lots of court cases!; Linus Torvalds, Guido van Rossum et al haven't had to do that after all. I think that, as in the case of the GPL, most people would respect the terms of the license without having to be coerced; and I agree that the first line of defense against incompatible forking is to design the architecture such that incompatible forks aren't needed. (This is different from the question of what if [insert favorite bad guys] use it for nefarious purposes. I still think the only way to guarantee that doesn't happen is to never let any copies of the code out of your grasp.) To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
--- Russell Wallace [EMAIL PROTECTED] wrote: On 8/28/06, Bill Hibbard [EMAIL PROTECTED] wrote: By open source distribution you are expressing optimism about human nature, and your developer community will mostly justify that optimism. The best approach for the few who disappoint you is to simply ignore them. I agree. When I suggested a no incompatible versions clause in the license, I wasn't thinking in terms of then you can fight and win lots of court cases!; Linus Torvalds, Guido van Rossum et al haven't had to do that after all. I think that, as in the case of the GPL, most people would respect the terms of the license without having to be coerced; and I agree that the first line of defense against incompatible forking is to design the architecture such that incompatible forks aren't needed. (This is different from the question of what if [insert favorite bad guys] use it for nefarious purposes. I still think the only way to guarantee that doesn't happen is to never let any copies of the code out of your grasp.) Thanks for the clarification. Now I see how to integrate your thinking with my own. -Steve __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 28/08/06, Russell Wallace [EMAIL PROTECTED] wrote: On 8/28/06, William Pearson [EMAIL PROTECTED] wrote: We may well not have enough computing resources available to do it on the cheap using local resources. But that is the approach I am inclined to take, I'll just wait until we do. Computing power isn't the only issue, and probably not the most important one; what do you think an Outlook killer could do that Outlook doesn't already do, and how would it know how to do it? Things like hooking it up to low quality sound video feeds and have it judge by posture/expression/time of day what the most useful piece of information in the RSS feeds/email etc to provide to the user is. We would have to program a large amounts of the behaviour to start with, but also by the dynamics and mechanism we create it would get more of an information about what the individual user wanted. The open source distibuted google killer will have the problem of who decides what goals the system has/starts with (depending upon your philosophy) Do what the users want you to do. Hmm. Possibly what we are talking about is not so different. and how to upgrade the collective if the goals were incorrect to start with. In the case of an open source AGI project, there would be no requirement that all users form a collective as far as their goals are concerned, only that they agree on running, maintaining and enhancing the software to serve their separate goals, just as is the case with e.g. the Internet today. Wouldn't interoperability be maintained by the same sort of pressures that mean that everyones tweaked version of Open Office shares the same file formats? The fact that the first mover that is incompatible loses then benefits from remaining compatible? Will --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, William Pearson [EMAIL PROTECTED] wrote: Things like hooking it up to low quality sound video feeds and have itjudge by posture/_expression_/time of day what the most useful piece ofinformation in the RSS feeds/email etc to provide to the user is. Wewould have to program a large amounts of the behaviour to start with, but also by the dynamics and mechanism we create it would get more ofan information about what the individual user wanted. Hmm... okay... it's not obvious to me that would be useful, but maybe it would. The nice thing about being a pessimist, one's surprises are more likely to be pleasant ones. Surprise me ^.^ Wouldn't interoperability be maintained by the same sort of pressuresthat mean that everyones tweaked version of Open Office shares the same file formats? The fact that the first mover that is incompatibleloses then benefits from remaining compatible? Yes, I would rely primarily on such incentives to maintain compatibility. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: Sampo [agi] Lossy ** lossless compressi
Actually I think I just may have invented one possible way to do that using a lossless probabilistic model in my previous email to this list. Did you read it? :-) I read it. I think that you have to be in a perfect world situation for what you propose to be feasible (i.e. it requires seeing the phrases in pretty much identical contexts -- which you have to recognize as identical -- and then being able to tell that that they affect the distribution of following words and phrases approximately equivalently -- which is an equally large problem). I'm afraid that it *really* does not look like a *feasible* solution at all (to me -- as I said, I think that going at all this via a probability model is not the way to go since it entails getting and analyzing those probabilities from far less data than I think is necessary for those operations). Mark - Original Message - From: Sampo Etelavuori [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, August 28, 2006 8:56 AM Subject: **SPAM** Re: [agi] Lossy ** lossless compressi On 8/28/06, Mark Waser [EMAIL PROTECTED] wrote: How does a lossless model observe that Jim is extremely fat and James continues to be morbidly obese are approximately equal? Actually I think I just may have invented one possible way to do that using a lossless probabilistic model in my previous email to this list. Did you read it? Anyway, in case you have a hard time figuring it out all by yourself, the idea behind it can be pretty straigthforwardly generalized to be used with phrases and thus I think phrases can be observed to be approximately equal if they can occur in pretty much identical contexts and affect the distribution of following words and phrases approximately equivalently. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 28/08/06, Russell Wallace [EMAIL PROTECTED] wrote: On 8/28/06, William Pearson [EMAIL PROTECTED] wrote: Things like hooking it up to low quality sound video feeds and have it judge by posture/expression/time of day what the most useful piece of information in the RSS feeds/email etc to provide to the user is. We would have to program a large amounts of the behaviour to start with, but also by the dynamics and mechanism we create it would get more of an information about what the individual user wanted. Hmm... okay... it's not obvious to me that would be useful, but maybe it would. The nice thing about being a pessimist, one's surprises are more likely to be pleasant ones. Surprise me ^.^ Possibly I am not explaining things clearly enough. One of my motivations for developing AI, apart from the challenge, is to enable me to get the information I need, when I need it. As a lot of the power I have in this world is through what I buy, I need to have this information available when I might buy something, which may be when I am in social situations etc. I can be a lot better ethical consumer with the the details I need at the right time given to me. As such I am interested in wearable and ubiquitous computing. Due to the constraints wearable computer place upon the designer, you really want the correct information given to you and nothing else that may distract the user unnecessarily. Knowing what the correct information is will entail knowing about the user and the uses current environment. Whether they rate energy efficiency or CO2 emissions as a priority, for example. It will also entail the google like system you are focused upon. I also think that a system designed to understand our body language/gestures/moods will also be able to be more easily and naturally trained as it has more information coming in about what we want and we will not have to be so explicit in our instructions. I'm also a pessimist in that I don't think an era of light will entail just because AI is invented, but I hope it will allow the few people that care to close the information gap that exists between producers and consumers. Or the government and the populace for that matter. And provide an economy marginally closer to what is promised by free market theory. You have hinted at the normative value of AI, I'm curious what you find it to be? Is it simply to speed up technological development so that we can escape the gravity well? Will Will --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
On 8/28/06, William Pearson [EMAIL PROTECTED] wrote: Possibly I am not explaining things clearly enough. One of mymotivations for developing AI, apart from the challenge, is to enableme to get the information I need, when I need it.As a lot of the power I have in this world is through what I buy, I need to have this information available when I might buy something,which may be when I am in social situations etc. I can be a lot betterethical consumer with the the details I need at the right time given to me. As such I am interested in wearable and ubiquitous computing.Due to the constraints wearable computer place upon the designer, youreally want the correct information given to you and nothing else thatmay distract the user unnecessarily. Ah, so you see this on a wearable... okay... that makes a bit more sense, and also of what you said earlier about computing power, since wearables are much more constrained in that regard than desktops. Knowing what the correct information is will entail knowing about theuser and the uses current environment. Whether they rate energy efficiency or CO2 emissions as a priority, for example. It will alsoentail the google like system you are focused upon. I should clarify: I think competing with Google in the search market is a losing proposition, that's already wrapped up; I'd look for new markets that nobody is serving well today. I use it only as an example of a software system that needs a lot of knowledge and computing power and is therefore run on a central rather than local basis. You have hinted at the normative value of AI, I'm curious what youfind it to be? Is it simply to speed up technological development so that we can escape the gravity well? Break the boundaries of space and time that currently apply to human life. Specifically: 1) Escape the gravity well. Or more precisely, we can already do that, but we can't live anywhere other than Earth, because the number of tasks that need to be carried out to keep a person alive for a year vastly exceeds the number of things a person can do in a year. Cracking that complexity barrier needs qualitative technological advances. 2) Stop or at least slow down the loss of fifty-plus million lives per year. That's a matter both of developing the hardware tools to work proficiently at the molecular level (i.e. some form of nanotechnology) and again the software tools to handle the complexity. To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Lossy ** lossless compressi
On 8/28/06, Mark Waser wrote: How does a lossless model observe that Jim is extremely fat and James continues to be morbidly obese are approximately equal? I realize this is far beyond the capabilities of current data compression programs, which typically predict the next byte in the context of the last few bytes using learned statistics. Of course we must do better. The model has to either know, or be able to learn, the relationships between Jim and James, is and continues to be, fat and obese, etc. I think a 1 GB corpus is big enough to learn most of this knowledge using statistical methods. C:\res\data\wikigrep -c . enwik9 File enwik9: 10920493 lines match enwik9: grep: input lines truncated - result questionable C:\res\data\wikigrep -i -c fat enwik9 File enwik9: 1312 lines match enwik9: grep: input lines truncated - result questionable C:\res\data\wikigrep -i -c obese enwik9 File enwik9: 111 lines match enwik9: grep: input lines truncated - result questionable C:\res\data\wikigrep -i obese enwik9 |grep -c fat File STDIN: 14 lines match So we know that obese occurs in about 0.001% of all paragraphs, but in 1% of paragraphs containing fat. This is an example of a distant bigram model, which has been shown to improve word perplexity in offline models [1]. We can improve on this method using e.g. latent semantic analysis [2] to exploit the transitive property of semantics: if A appears near (means) B and B appears near C, then A predicts C. Likewise, syntax is learnable. For example, if you encounter the X is you know that X is a noun, so you can predict a X was or Xs rather than he X or Xed. This type of knowledge can be exploited using similarity modeling [3] to improve word preplexity. (Thanks to Rob Freeman for pointing me to this). Let me give one more example using the same learning mechanism by which syntax is learned: All men are mortal. Socrates is a man. Therefore Socrates is mortal. All insects have 6 legs. Ants are insects. Therefore ants have 6 legs. Now predict: All frogs are green. Kermit is a frog. Therefore... [1] Rosenfeld, Ronald, A Maximum Entropy Approach to Adaptive Statistical Language Modeling, Computer, Speech and Language, 10, 1996. [2] Bellegarda, Jerome R., Speech recognition experiments using multi-span statistical language models, IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing, 717-720, 1999. [3] Ido Dagan, Lillian Lee, Fernando C. N. Pereira, Similarity-Based Models of Word Cooccurrence Probabilities, Machine Learning, 1999. http://citeseer.ist.psu.edu/dagan99similaritybased.html -- Matt Mahoney, [EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] Lossy ** lossless compressi
I think a 1 GB corpus is big enough to learn most of this knowledge using statistical methods. So we know that obese occurs in about 0.001% of all paragraphs, but in 1% of paragraphs containing fat. OK. Now try obese and morbidly or obese and clinically. I suspect that you are far more likely to statistically end up with obese being some form a disease (that being the context where is normally used) than it is to end up as fat. Statistical methods get absolutely trashed when you start switching contexts unless they can tell (or more likely, are told) that you've switched contexts. They are great at pulling context-specific clusters out of specific contexts but unless you get cross-context explanatory data (that you'll probably interpret with other than statistical methods -- see next section), I don't believe that statistical methods will recognize obese and fat as synonyms. Likewise, syntax is learnable. For example, if you encounter the X is you know that X is a noun, so you can predict a X was or Xs rather than he X or Xed. This type of knowledge can be exploited using similarity modeling [3] to improve word preplexity. Let me give one more example using the same learning mechanism by which syntax is learned: All men are mortal. Socrates is a man. Therefore Socrates is mortal. All insects have 6 legs. Ants are insects. Therefore ants have 6 legs. Now predict: All frogs are green. Kermit is a frog. Therefore... This isn't a statistical method (see other than statistical methods above :-). = = = = = So -- No, I *don't* believe that the 1GB corpus is big enough to learn most of this knowledge *USING STATISTICAL METHODS*. I *do* believe that it is large enough for other methods though. - Original Message - From: Matt Mahoney [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, August 28, 2006 3:37 PM Subject: Re: [agi] Lossy ** lossless compressi On 8/28/06, Mark Waser wrote: How does a lossless model observe that Jim is extremely fat and James continues to be morbidly obese are approximately equal? I realize this is far beyond the capabilities of current data compression programs, which typically predict the next byte in the context of the last few bytes using learned statistics. Of course we must do better. The model has to either know, or be able to learn, the relationships between Jim and James, is and continues to be, fat and obese, etc. I think a 1 GB corpus is big enough to learn most of this knowledge using statistical methods. C:\res\data\wikigrep -c . enwik9 File enwik9: 10920493 lines match enwik9: grep: input lines truncated - result questionable C:\res\data\wikigrep -i -c fat enwik9 File enwik9: 1312 lines match enwik9: grep: input lines truncated - result questionable C:\res\data\wikigrep -i -c obese enwik9 File enwik9: 111 lines match enwik9: grep: input lines truncated - result questionable C:\res\data\wikigrep -i obese enwik9 |grep -c fat File STDIN: 14 lines match So we know that obese occurs in about 0.001% of all paragraphs, but in 1% of paragraphs containing fat. This is an example of a distant bigram model, which has been shown to improve word perplexity in offline models [1]. We can improve on this method using e.g. latent semantic analysis [2] to exploit the transitive property of semantics: if A appears near (means) B and B appears near C, then A predicts C. Likewise, syntax is learnable. For example, if you encounter the X is you know that X is a noun, so you can predict a X was or Xs rather than he X or Xed. This type of knowledge can be exploited using similarity modeling [3] to improve word preplexity. (Thanks to Rob Freeman for pointing me to this). Let me give one more example using the same learning mechanism by which syntax is learned: All men are mortal. Socrates is a man. Therefore Socrates is mortal. All insects have 6 legs. Ants are insects. Therefore ants have 6 legs. Now predict: All frogs are green. Kermit is a frog. Therefore... [1] Rosenfeld, Ronald, A Maximum Entropy Approach to Adaptive Statistical Language Modeling, Computer, Speech and Language, 10, 1996. [2] Bellegarda, Jerome R., Speech recognition experiments using multi-span statistical language models, IEEE Intl. Conf. on Acoustics, Speech, and Signal Processing, 717-720, 1999. [3] Ido Dagan, Lillian Lee, Fernando C. N. Pereira, Similarity-Based Models of Word Cooccurrence Probabilities, Machine Learning, 1999. http://citeseer.ist.psu.edu/dagan99similaritybased.html -- Matt Mahoney, [EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
I would like to hear from others with this same point of view, and otherwise from anyone who has a idea that an open source AGI could be somehow made safe. While I also don't believe that you can protect your open source AGI from what if [insert favorite bad guys] use it for nefarious purposes, I'm of the opinion that the delta of how much they would benefit from it is far less than the delta that the good guys would benefit from it. Or, in other words, I suspect that it is far more likely that the [favorite bad guys] are going to have the funding and fortitude to get to AGI first while working alone than it is that the good guys can get to it first working alone. Also, in either case, open source will hasten AGI -- but less detrimentally if the bad guys get it first (vs. very beneficially if the good guys get it) -- since I also believe that earlier is also likely to be less detrimental even if the bad guys DO get it (for a whole slew of reasons including the beliefs that we are more likely to survive an early unfriendly AI than a later unfriendly AI, that there are threats that a biased read not friendly but not totally unfriendly AI will stop that could obliterate us otherwise, and that any sufficiently advanced AGI will go friendly given enough time). On the whole, I'm in favor of open-sourcing AGI with some precautions despite full knowledge that *no* precautions (including making something proprietary) will stop a sufficiently determined and reasonably knowledgeable bad guy (look at all the zero-day exploits coming out right on the heels of proprietary vendor patches). Mark - Original Message - From: Stephen Reed [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Monday, August 28, 2006 11:20 AM Subject: Re: [agi] AGI open source license --- Russell Wallace [EMAIL PROTECTED] wrote: That having been said, if you're serious about preventing the abuse of your software, I think the only answer is, don't distribute it. Follow the path of Novamente and indeed Google themselves (albeit for different reasons) and keep the software on your own machines and sell the services it provides. I assume that you fully understand the benefits and business case of an open source project, and that your point is made even with the former fully considered. I would like to hear from others with this same point of view, and otherwise from anyone who has a idea that an open source AGI could be somehow made safe. I would respond to the proprietary AGI alternative with the observation that one may suppose, as do I, that only one AGI is safer than many, possibly opposing, AGIs. With the proprietary model, there will be a market for others to enter. On the other hand an established open source project precludes competition, e.g. only one Wikipedia. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
Stephen Reed wrote: I would appreciate comments regarding additional constraints, if any, that should be applied to a traditional open source license to achieve a free but safe widespread distribution of software that may lead to AGI. ... My personal opinion is that the best license is the GPL. Either version 2 or 3...currently I can't choose between them (partially because version 3 is still being written). Note that may large GPL projects are quite successful. Consider, e.g., gcc. The claim that because anyone CAN change the code, anyone WILL change the code is probably fallacious. Most of those who try find that their changes are less than good. Usually those who decide to create a fork find themselves being left behind by the pace of development. So generally everyone sticks with the main tree...and perhaps submits changes that they think desirable into the project. Occasionally a fork will be successful. (X Window is no longer being developed from the XFree86 tree, e.g.) But since the license is GPL, this doesn't make any difference. How do you keep the bad guys from using it? You keep on developing. Those who fork tend to fall behind, unless they get community support. Now I'll admit that this is an idealized picture of the development process, but the outline is correct. Keeping a project going takes a good manager...one who can herd cats. It requires inspiring a degree of faith and trust in people who will be working without being paid. This means you've got to inspire them as well as get them to trust you. And you've got to articulate a vision of where the project should be headed next, roadmap is the common term, without stifling creativity. P.S.: Note that gcc has several chunks. Each language has a largely separate implementation, but each needs to generate the same kind of intermediate representation. This allows several essentially independent teams to each work separately. As to just *how* independent... consider the gdc compiler ( http://sourceforge.net/projects/dgcc ). This project is prevented by licensing constraints from having ANY direct connection to the rest of gcc. Yet it can still be integrated into gcc by an end user. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
Re: [agi] AGI open source license
--- Charles D Hixson [EMAIL PROTECTED] wrote: Stephen Reed wrote: I would appreciate comments regarding additional constraints, if any, that should be applied to a traditional open source license to achieve a free but safe widespread distribution of software that may lead to AGI. ... My personal opinion is that the best license is the GPL. Either version 2 or 3...currently I can't choose between them (partially because version 3 is still being written). [snip] How do you keep the bad guys from using it? You keep on developing. Those who fork tend to fall behind, unless they get community support. As a long time gcc tool chain user I agree with the comments below but let me draw a distinction between an AGI and all software that precedes it. I think that an AGI might be dangerous, especially an uncontrolled AGI. I wish to impose well-founded ethical and law-abiding policies on the various downloaded AI softwares. There are already laws and enforcement world wide that constrains most bad guys, but there is little precedent outside of robotics, numerical control and embedded, e.g. medical software for building in safety as part and parcel of the software. That puts me at odds with the GPL which forbids any additional constraints in derivations of its license. The Apache Software License can be further constrained in derived works, e.g. adding a must-be-federated clause. A let's cross that bridge when we come to it attitude might be to open source the pre-AI software, developing action justification, Friendship goal structure, proscribed behaviors, policy hierarchies, self-improvement monitoring and so forth without usage constrictions. As you point out, this will ease adoption and the great majority of downloaders will choose to operate in federated mode, out of self-interest. Should evidence arise that an AGI could indeed evolve, then alarmists and skeptics will arise, a hoopla will ensue, and behavior constraints will be externally imposed. Now I'll admit that this is an idealized picture of the development process, but the outline is correct. Keeping a project going takes a good manager...one who can herd cats. It requires inspiring a degree of faith and trust in people who will be working without being paid. This means you've got to inspire them as well as get them to trust you. And you've got to articulate a vision of where the project should be headed next, roadmap is the common term, without stifling creativity. P.S.: Note that gcc has several chunks. Each language has a largely separate implementation, but each needs to generate the same kind of intermediate representation. This allows several essentially independent teams to each work separately. As to just *how* independent... consider the gdc compiler ( http://sourceforge.net/projects/dgcc ). This project is prevented by licensing constraints from having ANY direct connection to the rest of gcc. Yet it can still be integrated into gcc by an end user. --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED] __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com --- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
