Re: RE: FW: [agi] A paper that actually does solve the problem of consciousness
We cannot ask Feynman, but I actually asked Deutsch. He does not only think QM is our most basic physical reality (he thinks math and computer science lie in quantum mechanics), but he even takes quite seriously his theory of parallel universes! and he is not alone. Speaking by myself, I would agree with you, but I think we would need to relativize the concept of agreement. I don't think QM is just another model of merely mathematical value to make finite predictions. I think physical models say something about our physical reality. If you deny QM as part of our physical reality then I guess you deny any other physical model. I wonder then what is left to you. You perhaps would embrace total skepticism, perhaps even solipsism. Current trends have moved from there to a more relativized positions, where models are considered so, models, but still with some value as part of our actual physical reality (just as Newtonian physics is not just completely wrong after General Relativity since it still describes a huge part of our physical reality). Well, I don't embrace solipsism, but that is really a philosophic and personal rather than scientific matter ... and, I'm not going talk here about what is, which IMO is not a matter for science ... but merely about what science can tell us. And, science cannot tell us whether QM or some empirically-equivalent, wholly randomness-free theory is the right one... ben g --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: RE: FW: [agi] A paper that actually does solve the problem of consciousness
2008/12/1 Ben Goertzel [EMAIL PROTECTED]: And, science cannot tell us whether QM or some empirically-equivalent, wholly randomness-free theory is the right one... If two theories give identical predictions under all circumstances about how the real world behaves, then they are not two separate theories, they are merely rewordings of the same theory. And choosing between them is arbitrary; you may prefer one to the other because human minds can visualise it more easily, or it's easier to calculate, or you have an aethetic preference for it. -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: RE: FW: [agi] A paper that actually does solve the problem of consciousness
If two theories give identical predictions under all circumstances about how the real world behaves, then they are not two separate theories, they are merely rewordings of the same theory. And choosing between them is arbitrary; you may prefer one to the other because human minds can visualise it more easily, or it's easier to calculate, or you have an aethetic preference for it. -- Philip Hunt, [EMAIL PROTECTED] However, the two theories may still have very different consequences **within the minds of the community of scientists** ... Even though T1 and T2 are empirically equivalent in their predictions, T1 might have a tendency to lead a certain community of scientists in better directions, in terms of creating new theories later on However, empirically validating this property of T1 is another question ... which leads one to the topic of scientific theories about the sociological consequences of scientific theories ;-) ben g --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] AIXI
--- On Sun, 11/30/08, Philip Hunt [EMAIL PROTECTED] wrote: Can someone explain AIXI to me? AIXI models an intelligent agent interacting with an environment as a pair of interacting Turing machines. At each step, the agent outputs a symbol to the environment, and the environment outputs a symbol and a numeric reward signal to the agent. The goal of the agent is to maximize the accumulated reward. Hutter proved that the optimal solution is for the agent to guess, at each step, that the environment is simulated by the shortest program that is consistent with the interaction observed so far. Hutter also proved that the optimal solution is not computable because the agent can't know which of its guesses are halting Turing machines. The best it can do is pick numbers L and T, try all 2^L programs up to length L for T steps each in order of increasing length, and guess the first one that is consistent. If there are no matches, then it needs to choose larger L and T and try again. That solution is called AIXI^TL. It's time complexity is O(T 2^L). In general, it may require L up to the length of the observed interaction (because there is a fast program that outputs the agent's observations from a list of length L). In a separate paper ( http://www.vetta.org/documents/ui_benelearn.pdf ), Legg and Hutter propose defining universal intelligence as the expected reward of an AIXI agent in random environments. The value of AIXI is not that it solves the general intelligence problem, but rather it explains why the problem is so hard. It also justifies a general principle that is already used in science and in practical machine learning algorithms: to choose the simplest hypothesis that fits the data. It formally defines simple as the length of the shortest program that outputs a description of the hypothesis. For example, to avoid overfitting in neural networks, you should use the smallest number of connections and the least amount of training needed to fit the training data, then stop. In this case, the complexity of your neural network is the length of the shortest program that outputs the configuration of your network and its weights. Even if you don't know what that program is, and haven't chosen a programming language, you may reasonably expect that fewer connections, smaller weights, and coarser weight quantization will result in a shorter program. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] AIXI
I really appreciate Matt's comments about this even though I am wary of the field. It is important to have some ideas about why the AI problem is so hard, and that insight is best told with some descriptive information like Matt's message. Of course, if no one is asking why then the poster has to wonder if he should explain it. However, I do not believe that the proposition that the shortest program that can produce the trial results would establish a solution to an AI problem is a sound philosophical basis for AGI. We need to be able to show that the program can learn about new things. Since this question has to be expressed as open ended statement using some vague general form, it is impossible or at least very hard to define a definitive test basis that could be used to establish the shortest program that can achieve the goal. Instead we use techniques that seem to do be adaptable and then try to figure out how to systematically deal with all of the errors that these methods tend to produce. Jim Bromer On Mon, Dec 1, 2008 at 12:04 PM, Matt Mahoney [EMAIL PROTECTED] wrote: --- On Sun, 11/30/08, Philip Hunt [EMAIL PROTECTED] wrote: Can someone explain AIXI to me? AIXI models an intelligent agent interacting with an environment as a pair of interacting Turing machines. At each step, the agent outputs a symbol to the environment, and the environment outputs a symbol and a numeric reward signal to the agent. The goal of the agent is to maximize the accumulated reward. Hutter proved that the optimal solution is for the agent to guess, at each step, that the environment is simulated by the shortest program that is consistent with the interaction observed so far. Hutter also proved that the optimal solution is not computable because the agent can't know which of its guesses are halting Turing machines. The best it can do is pick numbers L and T, try all 2^L programs up to length L for T steps each in order of increasing length, and guess the first one that is consistent. If there are no matches, then it needs to choose larger L and T and try again. That solution is called AIXI^TL. It's time complexity is O(T 2^L). In general, it may require L up to the length of the observed interaction (because there is a fast program that outputs the agent's observations from a list of length L). In a separate paper ( http://www.vetta.org/documents/ui_benelearn.pdf ), Legg and Hutter propose defining universal intelligence as the expected reward of an AIXI agent in random environments. The value of AIXI is not that it solves the general intelligence problem, but rather it explains why the problem is so hard. It also justifies a general principle that is already used in science and in practical machine learning algorithms: to choose the simplest hypothesis that fits the data. It formally defines simple as the length of the shortest program that outputs a description of the hypothesis. For example, to avoid overfitting in neural networks, you should use the smallest number of connections and the least amount of training needed to fit the training data, then stop. In this case, the complexity of your neural network is the length of the shortest program that outputs the configuration of your network and its weights. Even if you don't know what that program is, and haven't chosen a programming language, you may reasonably expect that fewer connections, smaller weights, and coarser weight quantization will result in a shorter program. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] AIXI
That was helpful. Thanks. 2008/12/1 Matt Mahoney [EMAIL PROTECTED]: --- On Sun, 11/30/08, Philip Hunt [EMAIL PROTECTED] wrote: Can someone explain AIXI to me? AIXI models an intelligent agent interacting with an environment as a pair of interacting Turing machines. At each step, the agent outputs a symbol to the environment, and the environment outputs a symbol and a numeric reward signal to the agent. The goal of the agent is to maximize the accumulated reward. Hutter proved that the optimal solution is for the agent to guess, at each step, that the environment is simulated by the shortest program that is consistent with the interaction observed so far. Hutter also proved that the optimal solution is not computable because the agent can't know which of its guesses are halting Turing machines. The best it can do is pick numbers L and T, try all 2^L programs up to length L for T steps each in order of increasing length, and guess the first one that is consistent. If there are no matches, then it needs to choose larger L and T and try again. That solution is called AIXI^TL. It's time complexity is O(T 2^L). In general, it may require L up to the length of the observed interaction (because there is a fast program that outputs the agent's observations from a list of length L). In a separate paper ( http://www.vetta.org/documents/ui_benelearn.pdf ), Legg and Hutter propose defining universal intelligence as the expected reward of an AIXI agent in random environments. The value of AIXI is not that it solves the general intelligence problem, but rather it explains why the problem is so hard. It also justifies a general principle that is already used in science and in practical machine learning algorithms: to choose the simplest hypothesis that fits the data. It formally defines simple as the length of the shortest program that outputs a description of the hypothesis. For example, to avoid overfitting in neural networks, you should use the smallest number of connections and the least amount of training needed to fit the training data, then stop. In this case, the complexity of your neural network is the length of the shortest program that outputs the configuration of your network and its weights. Even if you don't know what that program is, and haven't chosen a programming language, you may reasonably expect that fewer connections, smaller weights, and coarser weight quantization will result in a shorter program. -- Matt Mahoney, [EMAIL PROTECTED] --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?; Powered by Listbox: http://www.listbox.com -- Philip Hunt, [EMAIL PROTECTED] Please avoid sending me Word or PowerPoint attachments. See http://www.gnu.org/philosophy/no-word-attachments.html --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] AIXI
On Mon, Dec 1, 2008 at 8:04 PM, Matt Mahoney [EMAIL PROTECTED] wrote: The value of AIXI is not that it solves the general intelligence problem, but rather it explains why the problem is so hard. It doesn't explain why it's hard (is impossible hard?). That you can't solve a problem exactly, doesn't mean that there is no simple satisfactory solution. It also justifies a general principle that is already used in science and in practical machine learning algorithms: to choose the simplest hypothesis that fits the data. It formally defines simple as the length of the shortest program that outputs a description of the hypothesis. It's Solomonoff's universal induction, a much earlier result. Hutter generalized Solomonoff's induction to decision-making and proved some new results, but the idea of simple hypotheses prior and proof that it does good at learning are Solomonoff's. See ( http://www.scholarpedia.org/article/Algorithmic_probability ) for introduction. -- Vladimir Nesov [EMAIL PROTECTED] http://causalityrelay.wordpress.com/ --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Seeking CYC critiques
Steve, The KRAKEN paper was quite interesting, and has a LOT in common with my own Dr. Eliza. However, I saw no mention of Dr. Eliza's secret sauce, that boosts it from answering questions to solving problems given symptoms. The secret sauce has two primary ingredients: 1. The syntax of differential symptom statements - how people state a symptom that separates it from similar symptoms of other conditions. 2. Questions, the answers to which will probably carry #1 above recognizable differential symptom statements. Both of the above seem to require domain *experienced* people to code, as book learning doesn't seem to convey what people typically say, or what you have to say to them to get them to state their symptom in a differential way. Also, I suspect that knowledge coded today wouldn't work well in 50 years, when common speech has shifted. I finally gave up on having Dr. Eliza answer questions, because the round trip error rate seemed to be inescapably high. This is the product of: 1. The user's flaws in their world model. 2. The user's flaws in formulating their question. 3. The computer's errors in parsing the question. 4. The computer's errors in formulating an answer. 5. The user's errors in understanding the answer. 6. The user's errors from filing the answer into a flawed world model. Between each of these is: x.5 English's shortcomings in providing a platform to accurately state the knowledge, question, or answer. While each of these could be kept to 5%, it seemed completely hopeless to reduce the overall error rate to low enough to actually make it good for anything useful. Of course, everyone on this forum concentrates on #3 above, when in the real world, this is often/usually swamped by the others. Hence, I am VERY curious. Has KRAKEN found a worthwhile/paying niche in the world with itsw question answering, where people actually use it to their benefit? If so, then how did they deal with the round trip error rate? KRAKEN contains lots of good ideas, several of which were already on my wish list for Dr. Eliza sometime in the future. I suspect that a merger of technologies might be a world-beater. I wonder if the folks at Cycorp would be interested in such an effort? BTW, http://www.DrEliza.com is up and down these days, with plans for a new and more reliable version to be installed next weekend. Any thoughts? Steve Richfield == On 11/29/08, Stephen Reed [EMAIL PROTECTED] wrote: Hi Robin, There are no Cyc critiques that I know of in the last few years. I was employed seven years at Cycorp until August 2006 and my non-compete agreement expired a year later. An interesting competition was held by Project Halohttp://www.projecthalo.com/halotempl.asp?cid=30in which Cycorp participated along with two other research groups to demonstrate human-level competency answering chemistry questions. Results are herehttp://www.projecthalo.com/content/docs/ontologies_in_chemistry_ISWC2.pdf. Although Cycorp performed principled deductive inference giving detailed justifications, it was judged to have performed inferior due to the complexity of its justifications and due to its long running times. The other competitors used special purpose problem solving modules whereas Cycorp used its general purpose inference engine, extended for chemistry equations as needed. My own interest is in natural language dialog systems for rapid knowledge formation. I was Cycorp's first project manager for its participation in the the DARPA Rapid Knowledge Formation project where it performed to DARPA's satisfaction, but subsequently its RKF tools never lived up to Cycorp's expectations that subject matter experts could rapidly extend the Cyc KB without Cycorp ontological engineers having to intervene. A Cycorp paper describing its KRAKEN system is herehttp://www.google.com/url?sa=tsource=webct=rescd=1url=http%3A%2F%2Fwww.cyc.com%2Fdoc%2Fwhite_papers%2Fiaai.pdfei=IDgySdKoIJzENMzqpJcLusg=AFQjCNG1VlgQxAKERyiHj4CmPohVeZxRywsig2=o50LFe4D6TRC3VwC7ZNPxw . I would be glad to answer questions about Cycorp and Cyc technology to the best of my knowledge, which is growing somewhat stale at this point. 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 -- *From:* Robin Hanson [EMAIL PROTECTED] *To:* agi@v2.listbox.com *Sent:* Saturday, November 29, 2008 9:46:09 PM *Subject:* [agi] Seeking CYC critiques What are the best available critiques of CYC as it exists now (vs. soon after project started)? Robin Hanson [EMAIL PROTECTED] http://hanson.gmu.edu Research Associate, Future of Humanity Institute at Oxford University Associate Professor of Economics, George Mason University MSN 1D3, Carow Hall, Fairfax VA 22030- 703-993-2326 FAX: 703-993-2323 -- *agi* |
Re: [agi] Seeking CYC critiques
Mike, On 12/1/08, Mike Tintner [EMAIL PROTECTED] wrote: I wonder whether you'd like to outline an additional list of English/language's shortcomings here. I've just been reading Gary Marcus' Kluge - he has a whole chapter on language's shortcomings, and it would be v. interesting to compare and analyse. The real world is a wonderful limitless-dimensioned continuum of interrelated happenings. We have but a limited window to this, and have an even more limited assortment of words that have very specific meanings. Languages like Arabic vary pronunciation or spelling to convey additional shades of meaning, and languages like Chinese convey meaning via joined concepts. These may help, but they do not remove the underlying problem. This is like throwing pebbles onto a map and ONLY being able to communicate which pebble is closest to the intended location. Further, many words have multiple meanings, which is like only being able to specify certain disjoint multiples of pebbles, leaving it to AI to take a WAG (Wild Ass Guess) which one was intended. This becomes glaring obvious in language translation. I learned this stuff from people on the Russian national language translator project. Words in these two languages have very different shades of meaning, so that in general, a sentence in one language can NOT be translated to the other language with perfect accuracy, simply because the other language lacks words with the same shading. This is complicated by the fact that the original author may NOT have intended all of the shades of meaning, but was stuck with the words in the dictionary. For example, a man saying sit down in Russian to a woman, is conveying something like an order (and not a request) to sit down, shut up, and don't move. To remove that overloading, he might say please sit down in Russian. Then, it all comes down to just how he pronounces the please as to what he REALLY means, but of course, this is all lost in print. So, just how do you translate please sit down so as not to miss the entire meaning? One of my favorite pronunciation examples is excuse me. In Russian, it is approximately eezveneetsya minya and is typically spoken with flourish to emphasize apology. In Arabic, it is approximately afwan without emphasis on either syllable, and is typically spoken curtly, as if to say yea, I know I'm an idiot. It is really hard to pronounce these two syllables without emphases, but with flourish. There is much societal casting of meaning to common concepts. The underlying issue here is the very concept of translation, be it into a human language, or a table form in an AI engine.. Really good translations have more footnotes than translation, where these shades of meaning are explained, yet modern translation programs produce no footnotes, which pretty much consigns them to the trash translation pile, even with perfect disambiguation, which of course is impossible. Even the AI engines, that can carry these subtle overloadings, are unable to determine what nearby meaning the author actually intended. Hence, no finite language can convey specific meanings from within a limitlessly-dimensional continuum of potential meanings. English does better than most other languages, but it is still apparently not good enough even for automated question answering, which was my original point. Everywhere semantic meaning is touched upon, both within the wetware and within software, additional errors are introduced. This makes many answers worthless and all answers suspect, even before they are formed in the mind of the machine. Have I answered your question? Steve Richfield --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Seeking CYC critiques
Steve, Thanks. I was just looking for a systematic, v basic analysis of the problems language poses for any program, which I guess mainly come down to multiplicity - multiple -word meanings -word pronunciations -word spellings -word endings -word fonts -word/letter layout/design -languages [mixed discourse] -accents -dialects -sentence constructions to include new and novel -words -pronunciations -spellings -endings -layout/design -languages -accents -dialects -sentence constructions -all of which are *advantages* for a GI as opposed to a narrow AI. The latter wants the right meaning, the former wants many meanings - enables flexibility and creativity of explanation and association. Have I left anything out? Steve: MT:: I wonder whether you'd like to outline an additional list of English/language's shortcomings here. I've just been reading Gary Marcus' Kluge - he has a whole chapter on language's shortcomings, and it would be v. interesting to compare and analyse. The real world is a wonderful limitless-dimensioned continuum of interrelated happenings. We have but a limited window to this, and have an even more limited assortment of words that have very specific meanings. Languages like Arabic vary pronunciation or spelling to convey additional shades of meaning, and languages like Chinese convey meaning via joined concepts. These may help, but they do not remove the underlying problem. This is like throwing pebbles onto a map and ONLY being able to communicate which pebble is closest to the intended location. Further, many words have multiple meanings, which is like only being able to specify certain disjoint multiples of pebbles, leaving it to AI to take a WAG (Wild Ass Guess) which one was intended. This becomes glaring obvious in language translation. I learned this stuff from people on the Russian national language translator project. Words in these two languages have very different shades of meaning, so that in general, a sentence in one language can NOT be translated to the other language with perfect accuracy, simply because the other language lacks words with the same shading. This is complicated by the fact that the original author may NOT have intended all of the shades of meaning, but was stuck with the words in the dictionary. For example, a man saying sit down in Russian to a woman, is conveying something like an order (and not a request) to sit down, shut up, and don't move. To remove that overloading, he might say please sit down in Russian. Then, it all comes down to just how he pronounces the please as to what he REALLY means, but of course, this is all lost in print. So, just how do you translate please sit down so as not to miss the entire meaning? One of my favorite pronunciation examples is excuse me. In Russian, it is approximately eezveneetsya minya and is typically spoken with flourish to emphasize apology. In Arabic, it is approximately afwan without emphasis on either syllable, and is typically spoken curtly, as if to say yea, I know I'm an idiot. It is really hard to pronounce these two syllables without emphases, but with flourish. There is much societal casting of meaning to common concepts. The underlying issue here is the very concept of translation, be it into a human language, or a table form in an AI engine.. Really good translations have more footnotes than translation, where these shades of meaning are explained, yet modern translation programs produce no footnotes, which pretty much consigns them to the trash translation pile, even with perfect disambiguation, which of course is impossible. Even the AI engines, that can carry these subtle overloadings, are unable to determine what nearby meaning the author actually intended. Hence, no finite language can convey specific meanings from within a limitlessly-dimensional continuum of potential meanings. English does better than most other languages, but it is still apparently not good enough even for automated question answering, which was my original point. Everywhere semantic meaning is touched upon, both within the wetware and within software, additional errors are introduced. This makes many answers worthless and all answers suspect, even before they are formed in the mind of the machine. Have I answered your question? Steve Richfield -- agi | Archives | Modify Your Subscription --- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244id_secret=120640061-aded06 Powered by Listbox: http://www.listbox.com
Re: [agi] Seeking CYC critiques
Mike, More than multiplicity is the issue of discrete-point semantics vs. continuous real-world possibilities. Multiplicity could potentially be addressed by requiring users to put (clarifications) following unclear words (e.g. in response to diagnostic messages to clarify input). Dr. Eliza already does some of this, e.g. when it encounters If ... then ... it complains that it just wants to know the facts, and NOT how you think the world works. However, such approaches are unable to address the discrete vs. continuous issue, because every clarifying word has its own fuzziness, you don't know what the user's world model (and hence its discrete points) is, etc. Somewhat of an Islamic scholar (needed for escape after being sold into servitude in 1994), I am sometimes asked to clarify really simple-sounding concepts like agent of Satan. The problem is that many people from our culture simply have no place in their mental filing system for this information, without which, it is simply not possible to understand things like the present Middle East situation. Here, the discrete points that are addressable by their world-model are VERY far apart. For those of you who do understand agent of Satan, this very mental incapacity MAKES them agents of Satan. This is related to a passage in the Qur'an that states that most of the evil done in the world is done by people who think that they are doing good. Sounds like George Bush, doesn't it? In short, not only is this definition, but also this reality is circular. Here is one of those rare cases where common shortcomings in world models actually have common expressions referring to them. Too bad that these expressions come from other cultures, as we could sure use a few of them. Anyway, I would dismiss the multiplicity viewpoint, not because it is wrong, but because it guides people into disambiguation, which is ultimately unworkable. Once you understand that the world is a continuous domain, but that language is NOT continuous, you will realize the hopelessness of such efforts, as every question and every answer is in ERROR, unless by some wild stroke of luck, it is possible to say EXACTLY what is meant. As an interesting aside Bayesian programs tend (89%) to state their confidence, which overcomes some (13%) of such problems. Steve Richfield = On 12/1/08, Mike Tintner [EMAIL PROTECTED] wrote: Steve, Thanks. I was just looking for a systematic, v basic analysis of the problems language poses for any program, which I guess mainly come down to multiplicity - multiple -word meanings -word pronunciations -word spellings -word endings -word fonts -word/letter layout/design -languages [mixed discourse] -accents -dialects -sentence constructions to include new and novel -words -pronunciations -spellings -endings -layout/design -languages -accents -dialects -sentence constructions -all of which are *advantages* for a GI as opposed to a narrow AI. The latter wants the right meaning, the former wants many meanings - enables flexibility and creativity of explanation and association. Have I left anything out? Steve: MT:: I wonder whether you'd like to outline an additional list of English/language's shortcomings here. I've just been reading Gary Marcus' Kluge - he has a whole chapter on language's shortcomings, and it would be v. interesting to compare and analyse. The real world is a wonderful limitless-dimensioned continuum of interrelated happenings. We have but a limited window to this, and have an even more limited assortment of words that have very specific meanings. Languages like Arabic vary pronunciation or spelling to convey additional shades of meaning, and languages like Chinese convey meaning via joined concepts. These may help, but they do not remove the underlying problem. This is like throwing pebbles onto a map and ONLY being able to communicate which pebble is closest to the intended location. Further, many words have multiple meanings, which is like only being able to specify certain disjoint multiples of pebbles, leaving it to AI to take a WAG (Wild Ass Guess) which one was intended. This becomes glaring obvious in language translation. I learned this stuff from people on the Russian national language translator project. Words in these two languages have very different shades of meaning, so that in general, a sentence in one language can NOT be translated to the other language with perfect accuracy, simply because the other language lacks words with the same shading. This is complicated by the fact that the original author may NOT have intended all of the shades of meaning, but was stuck with the words in the dictionary. For example, a man saying sit down in Russian to a woman, is conveying something like an order (and not a request) to sit down, shut up, and don't move. To remove that overloading, he might say please sit down in Russian. Then, it all comes down to just
Re: FW: [agi] A paper that actually does solve the problem of consciousness
Ed, they used to combine ritalin with lsd for psychotherapy. It assists in absorbing insights achieved from psycholitic doses, which is a term for doses that are not fully psychedelic. Those are edifying on their own but are less organized. I don't know if you can get this in a clinical setting today. But these molecules are gradually being apprehended as tools On 11/30/08, Ben Goertzel [EMAIL PROTECTED] wrote: Ed, Unfortunately to reply to your message in detail would absorb a lot of time, because there are two issues mixed up 1) you don't know much about computability theory, and educating you on it would take a lot of time (and is not best done on an email list) 2) I may not have expressed some of my weird philosophical ideas about computability and mind and reality clearly ... though Abram, at least, seemed to get them ;) [but he has a lot of background in the area] Just to clarify some simple things though: Pi is a computable number, because there's a program that would generate it if allowed to run long enough Also, pi has been proved irrational; and, quantum theory really has nothing directly to do with uncomputability... About How can several pounds of matter that is the human brain model the true complexity of an infinity of infinitely complexity things? it is certainly thinkable that the brain is infinite not finite in its information content, or that it's a sort of antenna that receives information from some infinite-information-content source. I'm not saying I believe this, just saying it's a logical possibility, and not really ruled out by available data... Your reply seems to assume that the brain is a finite computational system and that other alternatives don't make sense. I think this is an OK working assumption for AGI engineers but it's not proved by any means. My main point in that post was, simply, that science and language seem intrinsically unable to distinguish computable from uncomputable realities. That doesn't necessarily mean the latter don't exist but it means they're not really scientifically useful entities. But, my detailed argument in favor of this point requires some basic understanding of computability math to appreciate, and I can't review those basics in an email, it's too much... ben g On Sun, Nov 30, 2008 at 4:20 PM, Ed Porter [EMAIL PROTECTED] wrote: Ben, On November 19, 2008 5:39 you wrote the following under the above titled thread: -- Ed, I'd be curious for your reaction to http://multiverseaccordingtoben.blogspot.com/2008/10/are-uncomputable-entities-useless-forhtml which explores the limits of scientific and linguistic explanation, in a different but possibly related way to Richard's argument. -- In the below email I asked you some questions about your article, which capture my major problem in understanding it, and I don't think I ever receive a reply The questions were at the bottom of such a long post you may well never have even seen them. I know you are busy, but if you have time I would be interested in hearing your answers to the following questions about the following five quoted parts (shown in red if you are seeing this in rich text) from you article. If you are too busy to respond just say so, either on or off list. - (1) In the simplest case, A2 may represent U directly in the language, using a single expression How, can U be directly represented in the language if it is uncomputable? I assume you consider any irational number, such as pi to be uncomputable (although, at least pi has a forumula that with enough computation can approach it as a limit –I assume that for most real numbers if there is such a formula, we do not know it.) (By the way, do we know for a fact that pi is irational, and if so how do we know other than that we have caluclated it to millions of places and not yet found an exact solution?) Merely communicating the symbol pi only represents the number if the agent receiving the communication has a more detailed definition, but any definition, such as a formula for iteratively approaching pi, which presumably is what you mean by R_U would only be an approximation. So U could never by fully represented unless one had infinite time --- and I generally consider it a waste of time to think about infinate time unless there is something valuable about such considerations that has a use in much more human-sized chunks of time. In fact, it seems the major message of quantum mechanics is that even physical reality doesn't have the time or machinery to compute uncomputable things, like a space constructed of dimensions each correspond to all the real numbers within some astronomical range . So the real number line is not really real. It is at best a construct of the human mind that can at best only be approximated in part.
