[agi] Re: Accidental Genius
Ryan, Thanks for the clarifications and the links! Cheers, Brad - Original Message - From: Bryan Bishop [EMAIL PROTECTED] To: [EMAIL PROTECTED]; [EMAIL PROTECTED]; [EMAIL PROTECTED]; [EMAIL PROTECTED] Sent: Wednesday, May 07, 2008 9:46 PM Subject: Re: Accidental Genius On Wed, May 7, 2008 at 9:21 PM, Brad Paulsen [EMAIL PROTECTED] wrote: I happened to catch a program on National Geographic Channel today entitled Accidental Genius. It was quite interesting from an AGI standpoint. One of the researchers profiled has invented a device that, by sending electromagnetic pulses through a person's skull to the appropriate spot in the left hemisphere of that person's brain, can achieve behavior similar to that of an idiot savant in a non-brain-damaged person (in the session shown, this was a volunteer college student). That's Snyder's work.* http://www.wireheading.com/brainstim/savant.html http://heybryan.org/mediawiki/index.php/rTMS Re: savantism, http://heybryan.org/intense_world_syndrome.html DIY rTMS: http://transcenmentalism.org/OpenStim/tiki (there's a mailing list) http://heybryan.org/mailing_lists.html Before being zapped by the device, the student is taken through a series of exercises. One is to draw a horse from memory. The other is to read aloud a very familiar saying with a slight grammatical mistake in it (the word the is duplicated, i.e., the the, in the saying -- sorry I can't recall the saying used). Then the student is shown a computer screen full of dots for about 1 second and asked to record his best guess at how many dots there were. This exercise is repeated several times (with different numbers of dots each time). It's not just being zapped, it's being specifically stimulated in a certain region of the brain; think of it like actually targetting the visual cortex, or actually targetting the anterior cingulate, the left ventrolateral amygdala, etc. And that's why this is interesting. I wrote somewhat about this on my site once: http://heybryan.org/recursion.html Specifically, if this can be used to modify attention, then can we use it to modify attention re: paying attention to attention? Sounds like a direct path to the singularity to me. The student is then zapped by the electromagnetic pulse device for 15 minutes. It's kind of scary to watch the guy's face flinch uncontrollably as each pulse is delivered. But, while he reported feeling something, he claimed there was no pain or disorientation. His language facilities were unimpaired (they zap a very particular spot in the left hemisphere based on brain scans taken of idiot savants). Right. The DIY setups that I have heard of haven't been able to be all that high-powered due to safety concerns -- not safety re: the brain, but safety when considering working with superhigh voltages so close to one's head. ;-) You can watch the episode on-line here: http://channel.nationalgeographic.com/tv-schedule. It's not scheduled for repeat showing anytime soon. Awesome. Thanks for the link. That's not a direct link (I couldn't find one). When you get to that Web page, navigate to Wed, May 7 at 3PM and click the More button under the picture. Unfortunately, the full-motion video is the size of a large postage stamp. The full screen view uses stop motion (at least i did on my laptop using a DSL-based WiFi hotspot). The audio is the same in both versions. - Bryan * Damien Broderick had to correct me on this, once. :-) --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups OpenCog.org (Open Cognition Project) group. To post to this group, send email to [EMAIL PROTECTED] To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/opencog?hl=en -~--~~~~--~~--~--~--- --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Re: pattern definition
Boris: I define intelligence as an ability to predict/plan by discovering projecting patterns within an input flow. IOW a capacity to generalize. A general intelligence is something that generalizes from incoming info. about the world. Well, no it can't be just that. Look at what you write at the end of your blog: Hope this makes sense. And it doesn't literally make much sense because your blog has a lot of generalizations with no examples - no individualizations/particularisations of, for example, what individual/particular problems your algorithms might apply to. The making sense level of your brain - an AGI that works - is the level that seeks individual examples (and exceptions) for every generalization. A general intelligence doesn't just generalize, it individualizes. It can talk not just about the field of AGI but about Boris K, Ben G., Stephen Reed, etc etc. And it has to, otherwise those generalizations don't make sense. I'm stressing this because so many people's ideas about AGI like yours involve only, or almost only a generalizing intelligence with no individualizing, sensemaking level. Boris: Entities must not be multiplied unnecessarily. William of Okkam. A pattern is a set of matching inputs. A match is a partial identity of the comparands. The comparands for general intelligence must incrementally indefinitely scale in complexity. The scaling must start from the bottom: uncompressed single-integer comparands, the match here is the sum of bitwise AND. For more see my blog: http://scalable-intelligence.blogspot.com/ Boris. - Original Message - From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, May 08, 2008 1:17 PM Subject: [agi] Re: pattern definition [EMAIL PROTECTED] wrote: Hello I am writing a literature review on AGI and I am mentioning the definition of pattern as explained by Ben in his work. A pattern is a representation of an object on a simpler scale. For example, a pattern in a drawing of a mathematical curve could be a program that can compute the curve from a formula (Looks et al. 2004). My supervisor told me that she doesn?t see how this can be simpler than the actual drawing. Any other definition I could use in the same context to explain to a non-technical audience? thanks xav Xav, [I am copying this to the AGI mailing list because it is more appropriate there than on Singularity] A more general definition of pattern would include the idea that there is a collection of mechanisms that take in a source of information (e.g. an image consisting of a grid of pixels) and respond in such a way that each mechanism 'triggers' in some way when a particular arrangement of signal values appears in the information source. Note that the triggering of each mechanism is the 'recognition' of a pattern, and the mechanism in question is a 'recognizer' of a pattern. (In this way of looking at things, there are many mechanisms, one for each pattern). The 'particular arrangement of signal values' is the pattern itself. Most importantly note that a mechanism does not have to trigger for some exact, deterministic set of signal values. For example, a mechanism could respond in a stochastic, noisy way to a whole bunch of different arrangements of signal values. It is allowed to be slightly inconsistent, and not always respond in the same way to the same input (although it would be a particularly bad pattern recognizer if it did not behave in a reasonably consistent way!). The amount of the 'triggering' reaction does not have to be all-or-nothing, either: the mechanism can give a graded response. What the above paragraph means is that the thing that we call a 'pattern' is actually 'whatever makes a mechanism trigger', and we have to be extremely tolerant of the fact that a wide range of different signal arrangements will give rise to triggering ... so a pattern is something much more amorphous and hard to define than simply *one* arrangement of signals. Finally, there is one more twist to this definition, which is very important. Everything said above was about arrangements of signals in the primary information source ... but we also allow that some mechanisms are designed to trigger on an arrangement of other *mechanisms*, not just primary input signals. In other words, this pattern finding system is hierarchical, and there can be abstract patterns. This definition of pattern is the most general one that I know of. I use it in my own work, but I do not know if it has been explicitly published and named by anyone else. In this conception, patterns are defined by the mechanisms that trigger, and further deponent sayeth not what they are, in any more fundamental way. And one last thing: as far as I can seem this does not easily map onto the concept of Kolmogorov complexity. At least, the mapping is very awkward and uninformative, if it exists. If a
Re: Symbol Grounding [WAS Re: [agi] AGI-08 videos]
- Original Message From: Mike Tintner [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, May 8, 2008 9:05:22 PM Subject: Re: Symbol Grounding [WAS Re: [agi] AGI-08 videos] I just want to make the point that I think categorical grounding is necessary for AGI, but I believe that it could be done through symbol processing. The reason is that old AI had some kind of learning facility. If symbols were associated with generalized forms that represented possible arrangements of data, they could then be used to designate categorical analysis that was implicitly derived from the IO data environment as well. So while I agree that categorical grounding is necessary, it is just that I also believe that could have been accomplished with AI programs that used symbol processing before Harnad's publication. I just believe that they lacked another facility which I call conceptual integration. My idea of conceptual integration includes blending but it is not limited to it. (And the computers in that era were too wimpy.) Jim Bromer Hi Jim, It's simply I think - and I stand to be corrected - that he has never pushed those levels v. hard at all. They are definitely there in his writing, but not elaborated. So the only enduring impression he has left, IMO, is the idea of symbol grounding - which people have interpreted in various ways. As you can imagine, I personally would have liked to see a lot more re those intermediate levels. And if he had pushed them, someone would presumably have brought him up in conection with Jeff Hawkins' work. Your Subscription Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
On Fri, May 9, 2008 at 4:29 AM, Matt Mahoney [EMAIL PROTECTED] wrote: I claim there is no P such that P(P,y) = P(y) for all y. (I assume you mean something like P((P,y))=P(y)). If P(s)=0 (one answer to all questions), then P((P,y))=0 and P(y)=0 for all y. -- Vladimir Nesov [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Accidental Genius
Right on. Everything I've read esp. Grandin, suggests strongly autism is crucially hypersensitivity rather than an emotional disorder. If every time the normal person touched someone, they got the equivalent of an electric shock, they'd stay away from people too. [Thanks for your previous links too]. Bryan: We discuss how excessive neuronal processing may render the world painfully intense when the neocortex is affected and even aversive when the amygdala is affected, leading to social and environmental withdrawal. --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
- Original Message From: Matt Mahoney [EMAIL PROTECTED] --- Jim Bromer [EMAIL PROTECTED] wrote: I don't want to get into a quibble fest, but understanding is not necessarily constrained to prediction. What would be a good test for understanding an algorithm? -- Matt Mahoney, [EMAIL PROTECTED] -- I don't have a ready answer for this. First of all, (maybe I do have a ready answer to this), understanding has to be understood in the context of partial understanding. I can understand something about a subject without being an expert in the subject, and I am Skeptical of anyone who claims that total understanding is feasible, (except for a bounded discussion of a bounded concept in which case I would only be skeptical with a small s.) So to start with, I could say that understanding an algorithm could be defined by various kinds of partial knowledge of it. What kinds of input does it react to, and what kinds of internal actions does it take? What kind of output does it produce. Can generalizations of the input it takes, its internal actions and its output be made. What was it designed to do? Can relations between specific examples or derived generalizations of its input, its internal actions and its output be made. While some of this kind of knowledge would require some kind of intelligence, others could be expressed in simpler data-concepts. Harnad's categorical grounding comes to mind. An experimental AI program would be capable of deriving data from the operation of an algorithm if its program was created around this paradigm of examining an algorithm. It could then create its own kind of analyses of the algorithm, and even though it might not be the same as an analysis that we might create, it still might be usable to produce something that would border on understanding. The capacity of prediction is significant in the kind of derived generalizations and categorical exemplars that I am thinking of, but the concept of understanding must go beyond simple prediction. Jim Bromer Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
RE: [agi] Re: pattern definition
So many overloads - pattern, complexity, atoms - can't we come up with new terms like schfinkledorfs? - but a very interesting question is - given an image of W x H pixels of 1 bit depth (on or off), one frame, how many patterns exist within this grid? When you think about it, it becomes an extremely difficult question to answer because within a static image you can have dupes, different sizes, dimensions, distortions, compressions, expansions, combo's... it's crazy. BUT, there is a pattern to the patterns - there's a mathematical description of them. John -Original Message- From: Richard Loosemore [mailto:[EMAIL PROTECTED] Sent: Thursday, May 08, 2008 11:18 AM To: agi@v2.listbox.com Subject: [agi] Re: pattern definition [EMAIL PROTECTED] wrote: Hello I am writing a literature review on AGI and I am mentioning the definition of pattern as explained by Ben in his work. A pattern is a representation of an object on a simpler scale. For example, a pattern in a drawing of a mathematical curve could be a program that can compute the curve from a formula (Looks et al. 2004). My supervisor told me that she doesn?t see how this can be simpler than the actual drawing. Any other definition I could use in the same context to explain to a non-technical audience? thanks xav Xav, [I am copying this to the AGI mailing list because it is more appropriate there than on Singularity] A more general definition of pattern would include the idea that there is a collection of mechanisms that take in a source of information (e.g. an image consisting of a grid of pixels) and respond in such a way that each mechanism 'triggers' in some way when a particular arrangement of signal values appears in the information source. Note that the triggering of each mechanism is the 'recognition' of a pattern, and the mechanism in question is a 'recognizer' of a pattern. (In this way of looking at things, there are many mechanisms, one for each pattern). The 'particular arrangement of signal values' is the pattern itself. Most importantly note that a mechanism does not have to trigger for some exact, deterministic set of signal values. For example, a mechanism could respond in a stochastic, noisy way to a whole bunch of different arrangements of signal values. It is allowed to be slightly inconsistent, and not always respond in the same way to the same input (although it would be a particularly bad pattern recognizer if it did not behave in a reasonably consistent way!). The amount of the 'triggering' reaction does not have to be all-or-nothing, either: the mechanism can give a graded response. What the above paragraph means is that the thing that we call a 'pattern' is actually 'whatever makes a mechanism trigger', and we have to be extremely tolerant of the fact that a wide range of different signal arrangements will give rise to triggering ... so a pattern is something much more amorphous and hard to define than simply *one* arrangement of signals. Finally, there is one more twist to this definition, which is very important. Everything said above was about arrangements of signals in the primary information source ... but we also allow that some mechanisms are designed to trigger on an arrangement of other *mechanisms*, not just primary input signals. In other words, this pattern finding system is hierarchical, and there can be abstract patterns. This definition of pattern is the most general one that I know of. I use it in my own work, but I do not know if it has been explicitly published and named by anyone else. In this conception, patterns are defined by the mechanisms that trigger, and further deponent sayeth not what they are, in any more fundamental way. And one last thing: as far as I can seem this does not easily map onto the concept of Kolmogorov complexity. At least, the mapping is very awkward and uninformative, if it exists. If a mechanism triggers on a possibly stochastic, nondeterminstic set of features, it can hardly be realised by a feasible algorithm, so talking about a pattern as an algorithm that can generate the source seems, to me at least, to be unworkable. Hope that is useful. Richard Loosemore P.S. Nice to see some Welsh in the boilerplate stuff at the bottom of your message. I used to work at Bangor in the early 90s, so it brought back fond memories to see Prifysgol Bangor! Are you in the Psychology department? --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; 2bb036 Powered by Listbox: http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your
Re: [agi] Re: pattern definition
And it doesn't literally make much sense because your blog has a lot of generalizations with no examples - no individualizations/particularisations of, for example, what individual/particular problems your algorithms might apply to. The making sense level of your brain - an AGI that works - is the level that seeks individual examples (and exceptions) for every generalization. If you need examples you're in the wrong field. A general intelligence doesn't just generalize, it individualizes. It can talk not just about the field of AGI but about Boris K, Ben G., Stephen Reed, etc etc. And it has to, otherwise those generalizations don't make sense. I'm stressing this because so many people's ideas about AGI like yours involve only, or almost only a generalizing intelligence with no individualizing, sensemaking level. Boris: Entities must not be multiplied unnecessarily. William of Okkam. A pattern is a set of matching inputs. A match is a partial identity of the comparands. The comparands for general intelligence must incrementally indefinitely scale in complexity. The scaling must start from the bottom: uncompressed single-integer comparands, the match here is the sum of bitwise AND. For more see my blog: http://scalable-intelligence.blogspot.com/ Boris. - Original Message - From: Richard Loosemore [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, May 08, 2008 1:17 PM Subject: [agi] Re: pattern definition [EMAIL PROTECTED] wrote: Hello I am writing a literature review on AGI and I am mentioning the definition of pattern as explained by Ben in his work. A pattern is a representation of an object on a simpler scale. For example, a pattern in a drawing of a mathematical curve could be a program that can compute the curve from a formula (Looks et al. 2004). My supervisor told me that she doesn?t see how this can be simpler than the actual drawing. Any other definition I could use in the same context to explain to a non-technical audience? thanks xav Xav, [I am copying this to the AGI mailing list because it is more appropriate there than on Singularity] A more general definition of pattern would include the idea that there is a collection of mechanisms that take in a source of information (e.g. an image consisting of a grid of pixels) and respond in such a way that each mechanism 'triggers' in some way when a particular arrangement of signal values appears in the information source. Note that the triggering of each mechanism is the 'recognition' of a pattern, and the mechanism in question is a 'recognizer' of a pattern. (In this way of looking at things, there are many mechanisms, one for each pattern). The 'particular arrangement of signal values' is the pattern itself. Most importantly note that a mechanism does not have to trigger for some exact, deterministic set of signal values. For example, a mechanism could respond in a stochastic, noisy way to a whole bunch of different arrangements of signal values. It is allowed to be slightly inconsistent, and not always respond in the same way to the same input (although it would be a particularly bad pattern recognizer if it did not behave in a reasonably consistent way!). The amount of the 'triggering' reaction does not have to be all-or-nothing, either: the mechanism can give a graded response. What the above paragraph means is that the thing that we call a 'pattern' is actually 'whatever makes a mechanism trigger', and we have to be extremely tolerant of the fact that a wide range of different signal arrangements will give rise to triggering ... so a pattern is something much more amorphous and hard to define than simply *one* arrangement of signals. Finally, there is one more twist to this definition, which is very important. Everything said above was about arrangements of signals in the primary information source ... but we also allow that some mechanisms are designed to trigger on an arrangement of other *mechanisms*, not just primary input signals. In other words, this pattern finding system is hierarchical, and there can be abstract patterns. This definition of pattern is the most general one that I know of. I use it in my own work, but I do not know if it has been explicitly published and named by anyone else. In this conception, patterns are defined by the mechanisms that trigger, and further deponent sayeth not what they are, in any more fundamental way. And one last thing: as far as I can seem this does not easily map onto the concept of Kolmogorov complexity. At least, the mapping is very awkward and uninformative, if it exists. If a mechanism triggers on a possibly stochastic, nondeterminstic set of features, it can hardly be realised by a feasible algorithm, so talking about a pattern as an algorithm that can generate the source seems, to me at least, to be unworkable. Hope that is useful. Richard Loosemore P.S.
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
Hi Matt, You asked: What would be a good test for understanding an algorithm? As I mentioned before, I want to create a system capable of being taught - specifically capable of being taught skills. And I strongly share your interest in answers to this question. A student should be able to demonstrate to its mentor that it has learned, or is making progress learning, the given subject matter. A small use case might be the following: Skill: Trimming the whitespace off both ends of a character string. Domain knowledge: * a character string is a sequence of characters * characters can be partitioned into whitespace and non-whitespace charactersAssumed existing algorithm skills: * a sequence of objects can be inspected by index position * a sequence of objects can be mutated by the remove operation Tests to be passed by Texai following the skill acquisition: * what are the preconditions for applying this skill? * what are the postconditions for applying this skill? * can this skill be applied to the sequence [1, 2, 3, 4, 5] ? * can this skill be applied to the string abc? * trim the whitespace off an empty string * trim the whitespace off the stringabc * trim the whitespace off the string ab cCheers. -Steve Stephen L. Reed Artificial Intelligence Researcher http://texai.org/blog http://texai.org 3008 Oak Crest Ave. Austin, Texas, USA 78704 512.791.7860 - Original Message From: Matt Mahoney [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, May 8, 2008 11:02:33 PM Subject: Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers) --- Jim Bromer [EMAIL PROTECTED] wrote: I don't want to get into a quibble fest, but understanding is not necessarily constrained to prediction. What would be a good test for understanding an algorithm? -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
Matt, On 5/9/08, Stephen Reed [EMAIL PROTECTED] wrote: Skill: Trimming the whitespace off both ends of a character string. One of the many annoyances of writing real-world AI programs is having to write this function; to replace the broken system functions that are supposed to do this, but which don't work properly with control characters, UNICODE characters that should be considered to be whitespace (e.g. long spaces), etc. Clearly, some intelligent humans haven't yet mastered this algorithm, and ordinary software testing methods have failed to disclose the remaining bugs. Steve Richfield --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
On 5/9/08, Stephen Reed [EMAIL PROTECTED] wrote: Skill: Trimming the whitespace off both ends of a character string. One of the many annoyances of writing real-world AI programs is having to write this function; to replace the broken system functions that are supposed to do this, but which don't work properly with control characters, UNICODE characters that should be considered to be whitespace (e.g. long spaces), etc. Clearly, some intelligent humans haven't yet mastered this algorithm, and ordinary software testing methods have failed to disclose the remaining bugs. Steve Richfield --- I agree. And you have to find the instructions before you can read them. (Seriously.) Jim Bromer Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Re: pattern definition
Jim, I doubt that your specification equals my individualization. If I want to be able to recognize the individuals, Curtis/Brian/Carl/ and Billi Bromer,only images will do it: http://www.dunningmotorsales.com/IMAGES/people/Curtis%20Bromer.jpg http://www.newyorksocialdiary.com/socialdiary/2006/02_27_06/images/BRomer-Sir-PThomas.jpg http://www.stellarsales.com/images/carl3.jpg http://www.dec-sped.org/images/executiveboard/Billi_Bromer.jpg If you'd like to try a logical, verbal, mathematical description of any oneof those individuals, so that someone will be sure to recognize them, be my guest :). That's why they put photos on your passport and not program printouts or verbal descriptions. All the words in the world won't tell you what a bucket looks like. McLuhan Jim/MT:, The making sense level of your brain - an AGI that works - is the level that seeks individual examples (and exceptions) for every generalization. A general intelligence doesn't just generalize, it individualizes. It can talk not just about the field of AGI but about Boris K, Ben G., Stephen Reed, etc etc. And it has to, otherwise those generalizations don't make sense. I'm stressing this because so many people's ideas about AGI ... involve only, or almost only a generalizing intelligence with no individualizing, sensemaking level. -- I agree with what Mike was saying in the part of his message I quoted here, except that the ability to understand involves the ability to make generalizations. But, a generalization can be seen as a specific relative to another level of generalization. I also think most people who have been seriously involved in AI and who think of AI in terms of generalization realize that specification must play an important role in understanding. Jim Bromer -- Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. -- agi | Archives | Modify Your Subscription --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Re: pattern definition
Mike, what is your stance on vector images? --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
[agi] organising parallel processes, try2
I'll try to explain it more.. Suppose you have a lot of processes, all containing some production rule(s). They communicate with messages. They all should get cpu time somehow. Some processes just do low-level responses, some monitor other processes, etc. Some are involved in looking at the world, some involved in planning, etc. I'm thinking of a system like SOAR, but in parallel. Are there any systems that work like this, and have some way to organise the processes (assign cpu time, guide the communication, group according to some criteria..) I'd like to look at a bunch of those and compare the pros cons thanks Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] organising parallel processes, try2
Hi, The Texai system, as I envision its deployment, will have the following characteristics: * a lot of processes * a lot of hosts * message passing between processes, that are arranged in a hierarchical control system * higher level processes will be deliberative, executing compiled production rules (e.g. acquired skills) * lower level processes will be reactive, even so far as not to contain any state whatsoever, if the sensed world itself will suffice * some higher level processes on each host will be agents of the Host Resource Allocation Agency and will have the learned skills sufficient to optimally allocate host resources (e.g. CPU cores, RAM, KB cache) on behalf of other processes (i.e. agents) * I have not yet thought much about how these resources should be allocated except to initially adopt the scheduling algorithms used by the Linux OS for its processes (e.g. each process has a priority, schedule the processes to achieve maximum use of the resources, allow real-time response for processes that must have it, do not allow low priority processes to starve, etc.) Cheers. -Steve Stephen L. Reed Artificial Intelligence Researcher http://texai.org/blog http://texai.org 3008 Oak Crest Ave. Austin, Texas, USA 78704 512.791.7860 - Original Message From: rooftop8000 [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Friday, May 9, 2008 3:24:14 PM Subject: [agi] organising parallel processes, try2 I'll try to explain it more.. Suppose you have a lot of processes, all containing some production rule(s). They communicate with messages. They all should get cpu time somehow. Some processes just do low-level responses, some monitor other processes, etc. Some are involved in looking at the world, some involved in planning, etc. I'm thinking of a system like SOAR, but in parallel. Are there any systems that work like this, and have some way to organise the processes (assign cpu time, guide the communication, group according to some criteria..) I'd like to look at a bunch of those and compare the pros cons thanks Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
--- Steve Richfield [EMAIL PROTECTED] wrote: Matt, On 5/8/08, Matt Mahoney [EMAIL PROTECTED] wrote: --- Steve Richfield [EMAIL PROTECTED] wrote: On 5/7/08, Vladimir Nesov [EMAIL PROTECTED] wrote: See http://www.overcomingbias.com/2008/01/newcombs-proble.html After many postings on this subject, I still assert that ANY rational AGI would be religious. Not necessarily. You execute a program P that inputs the conditions of the game and outputs 1 box or 2 boxes. Omega executes a program W as follows: if P outputs 1 box then put $1 million in box B else leave box B empty. No matter what P is, it cannot call W because it would be infinite recursion. QED this is NOT the program that Omega executes. No, it is given that Omega never makes a mistake. Please try again. A rational agent only has to know that there are some things it cannot compute. In particular, it cannot understand its own algorithm. There is a LOT wrapped up in your only. It is one thing to know that you can't presently compute certain things that you have identified, and quite another to believe that an unseen power changes things that you have NOT identified as being beyond your present (flawed) computational abilities. No matter how extensive your observations, you can NEVER be absolutely sure that you understand anything, and you will in fact fail to understand key details of some things without realizing it. With a good workable explanation of the variances between predicted and actual events (God), of course you will continue to look for less divine explanations, but at exactly what point do you broadly dismiss ALL divine explanations, in the absence of alternative explanations? Intelligent agents cannot recognize higher levels of intelligence in other agents. We invoke divine explanation (godlike AI) because people have trouble accepting mathematical proofs of this statement. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
--- Vladimir Nesov [EMAIL PROTECTED] wrote: On Fri, May 9, 2008 at 4:29 AM, Matt Mahoney [EMAIL PROTECTED] wrote: I claim there is no P such that P(P,y) = P(y) for all y. (I assume you mean something like P((P,y))=P(y)). If P(s)=0 (one answer to all questions), then P((P,y))=0 and P(y)=0 for all y. You're right. But we wouldn't say that the trivial language P = {0,1}* understands anything. That is a problem with my formal definition of understanding. I teach a C++ class. To test my student's understanding of the language, I give them exam questions with code and ask them what it outputs. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
[agi] Self-maintaining Architecture first for AI
After getting completely on the wrong foot last time I posted something, and not having had time to read the papers I should have. I have decided to try and start afresh and outline where I am coming from. I'll get around to do a proper paper later. There are two possible modes for designing a computer system. I shall characterise them as the active and the passive. The active approach attempts to solve the problem directly where as the passive approach gives a framework under which the problem and related other ones can be more easily solved. The passive approach is generally less efficient but much more reconfigurable. The passive approach is used when there is large number of related possible problems, with a large variety of solutions. Examples of the passive approach are mainly architectures, programming languages and operating systems, with a variety of different goals. They are not always completely passive, for example automatic garbage collection impacts the system somewhat. One illuminating example is the variety of security systems that have been built along this structure. Security in this sense means that the computer system is composed of domains, where not all of them are equally trusted or allowed resources. Now it is possible to set up a passive system designed with security in mind insecurely, by allowing all domains to access every file on the hard disk. Passive systems do not guarantee the solution they are aiming to aid, the most they can do is allow as many possible things to be represented and permit the prevention of certain things. A passive security system allows the prevention of a domain lowering the security of a part of another domain. The set of problems that I intend to help solve is the set of self-maintainanceing computer systems. Self-maintainance is basically reconfiguring the computer to be suited to the environment it finds itself in. The reason why I think it needs to be solved first before AI is attempted is 1) humans self-maintenance, 2) otherwise the very complex computer systems we build for AI will have to be maintained by ourselves which may become increasingly difficult as they approach human level. It is worth noting that I am using AI in the pure sense of being able to solve problems. It is entirely possible to get very high complexity problem solvers (including potentially passing the turing test) that cannot self-maintaince. There a large variety of possible AIs (different bodies/environments/computational resources/goals) as can be seen from the variety of humans and (proto?) intelligences of animals, so a passive approach is not unreasonable. In the case of self-maintaining system, what is that we wish the architecture to prevent? About the only thing we can prevent is a useless program being able to degrade of the system from the current level of operation by taking control of resources. However we also want to enable useful programs to be able to control more resources. To do this we must protect the resources and make sure the correct programs can somehow get the correct resources, the internal programs should do the rest. So it is a resource management problem. Any active force for better levels of operation has to come from the internal programs of the architecture, and once the higher level of operation has been reached the architecture should act as a ratchet to prevent it from slipping down again. Protecting resources amounts to the security problem which we have a fair amount of literature on and the only passive form of resource allocation we know of is a economic system. ... to be continued I might go into further detail about what I mean by resource but that will have to wait for a further post. Will Pearson --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
On Sat, May 10, 2008 at 2:09 AM, Matt Mahoney [EMAIL PROTECTED] wrote: --- Vladimir Nesov [EMAIL PROTECTED] wrote: (I assume you mean something like P((P,y))=P(y)). If P(s)=0 (one answer to all questions), then P((P,y))=0 and P(y)=0 for all y. You're right. But we wouldn't say that the trivial language P = {0,1}* understands anything. That is a problem with my formal definition of understanding. Then make a definition that fits your claim. As currently stated, it looks erroneous to me, and I can't see how it's possible to fix that without explicating your assertion mathematically. -- Vladimir Nesov [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
Matt, You asked What would be a good test for understanding an algorithm? Thanks for posing this question. It has been a good exercise. Assuming that the key word here is understanding rather than algorithm, I submit - A test of understanding is if one can give a correct *explanation* for any and all of the possible outputs that it (the thing to understand) produces. I see this as having merit in explaining understanding because it shows that one grasps the transformations that occur in the system. If one is given a set of inputs and states, then one could state what the output would be by stepping through the transformations that take place. This differs slightly from prediction because prediction demands that you be able to instrument every state and input for a given moment. This creates a distinction between it being hard or extremely hard or in practice impossible to predict vs. not so hard to understand what is going on. Stan Jim Bromer wrote: - Original Message From: Matt Mahoney [EMAIL PROTECTED] --- Jim Bromer [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: I don't want to get into a quibble fest, but understanding is not necessarily constrained to prediction. What would be a good test for understanding an algorithm? -- Matt Mahoney, [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] -- I don't have a ready answer for this. First of all, (maybe I do have a ready answer to this), understanding has to be understood in the context of partial understanding. I can understand something about a subject without being an expert in the subject, and I am Skeptical of anyone who claims that total understanding is feasible, (except for a bounded discussion of a bounded concept in which case I would only be skeptical with a small s.) So to start with, I could say that understanding an algorithm could be defined by various kinds of partial knowledge of it. What kinds of input does it react to, and what kinds of internal actions does it take? What kind of output does it produce. Can generalizations of the input it takes, its internal actions and its output be made. What was it designed to do? Can relations between specific examples or derived generalizations of its input, its internal actions and its output be made. While some of this kind of knowledge would require some kind of intelligence, others could be expressed in simpler data-concepts. Harnad's categorical grounding comes to mind. An experimental AI program would be capable of deriving data from the operation of an algorithm if its program was created around this paradigm of examining an algorithm. It could then create its own kind of analyses of the algorithm, and even though it might not be the same as an analysis that we might create, it still might be usable to produce something that would border on understanding. The capacity of prediction is significant in the kind of derived generalizations and categorical exemplars that I am thinking of, but the concept of understanding must go beyond simple prediction. Jim Bromer Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://us.rd.yahoo.com/evt=51733/*http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ *agi* | Archives http://www.listbox.com/member/archive/303/=now http://www.listbox.com/member/archive/rss/303/ | Modify http://www.listbox.com/member/?; Your Subscription [Powered by Listbox] http://www.listbox.com --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: Newcomb's Paradox (was Re: [agi] Goal Driven Systems and AI Dangers)
--- Vladimir Nesov [EMAIL PROTECTED] wrote: On Sat, May 10, 2008 at 2:09 AM, Matt Mahoney [EMAIL PROTECTED] wrote: --- Vladimir Nesov [EMAIL PROTECTED] wrote: (I assume you mean something like P((P,y))=P(y)). If P(s)=0 (one answer to all questions), then P((P,y))=0 and P(y)=0 for all y. You're right. But we wouldn't say that the trivial language P = {0,1}* understands anything. That is a problem with my formal definition of understanding. Then make a definition that fits your claim. As currently stated, it looks erroneous to me, and I can't see how it's possible to fix that without explicating your assertion mathematically. OK, let me make more clear the distinction between running a program and simulating it. Say that a program P simulates a program Q if for all y, P((Q,y)) = the output is +Q(y) where + means string concatenation. In other words, given Q and y, P prints not Q(y) but a statement describing what the output Q(y) would be. Then I claim there is no finite state machine P that can simulate itself (including the trivial case). P needs as many states as Q to simulate it, plus additional states to print the output is . I also claim that P understands Q can be reasonably interpreted as P can simulate Q, but I can't prove it :-) -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com
Re: [agi] Self-maintaining Architecture first for AI
William Pearson wrote: After getting completely on the wrong foot last time I posted something, and not having had time to read the papers I should have. I have decided to try and start afresh and outline where I am coming from. I'll get around to do a proper paper later. There are two possible modes for designing a computer system. I shall characterise them as the active and the passive. The active approach attempts to solve the problem directly where as the passive approach gives a framework under which the problem and related other ones can be more easily solved. The passive approach is generally less efficient but much more reconfigurable. The passive approach is used when there is large number of related possible problems, with a large variety of solutions. Examples of the passive approach are mainly architectures, programming languages and operating systems, with a variety of different goals. They are not always completely passive, for example automatic garbage collection impacts the system somewhat. One illuminating example is the variety of security systems that have been built along this structure. Security in this sense means that the computer system is composed of domains, where not all of them are equally trusted or allowed resources. Now it is possible to set up a passive system designed with security in mind insecurely, by allowing all domains to access every file on the hard disk. Passive systems do not guarantee the solution they are aiming to aid, the most they can do is allow as many possible things to be represented and permit the prevention of certain things. A passive security system allows the prevention of a domain lowering the security of a part of another domain. The set of problems that I intend to help solve is the set of self-maintainanceing computer systems. Self-maintainance is basically reconfiguring the computer to be suited to the environment it finds itself in. The reason why I think it needs to be solved first before AI is attempted is 1) humans self-maintenance, 2) otherwise the very complex computer systems we build for AI will have to be maintained by ourselves which may become increasingly difficult as they approach human level. It is worth noting that I am using AI in the pure sense of being able to solve problems. It is entirely possible to get very high complexity problem solvers (including potentially passing the turing test) that cannot self-maintaince. There a large variety of possible AIs (different bodies/environments/computational resources/goals) as can be seen from the variety of humans and (proto?) intelligences of animals, so a passive approach is not unreasonable. In the case of self-maintaining system, what is that we wish the architecture to prevent? About the only thing we can prevent is a useless program being able to degrade of the system from the current level of operation by taking control of resources. However we also want to enable useful programs to be able to control more resources. To do this we must protect the resources and make sure the correct programs can somehow get the correct resources, the internal programs should do the rest. So it is a resource management problem. Any active force for better levels of operation has to come from the internal programs of the architecture, and once the higher level of operation has been reached the architecture should act as a ratchet to prevent it from slipping down again. Protecting resources amounts to the security problem which we have a fair amount of literature on and the only passive form of resource allocation we know of is a economic system. ... to be continued I might go into further detail about what I mean by resource but that will have to wait for a further post. This is still quite ambiguous on a number of levels, so would it be possible for you to give us a road map of where the argument is going? At the moment I am not sure what the theme is. For example, your distinction between active and passive could mean that you think we should be building a general learning mechanism, or it could mean that you think we should be taking a Generative Programming approach to the construction of an AI, or ... probably several other meanings. Richard Loosemore --- agi Archives: http://www.listbox.com/member/archive/303/=now RSS Feed: http://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: http://www.listbox.com/member/?member_id=8660244id_secret=101455710-f059c4 Powered by Listbox: http://www.listbox.com