Re: Yawn. More definitions of intelligence? [WAS Re: [agi] Ben's Definition of Intelligence]
Richard, I don't think Shane and Marcus's overview of definitions-of-intelligence is poor quality. I think it is just doing something different than what you think it should be doing. The overview is exactly that: A review of what researchers have said about the definition of intelligence. This is useful as a view into the cultural mind-space of the research community regarding the intelligence concept. As for their formal definition of intelligence, I think it is worthwhile as a precise formulation of one perspective on the multidimensional concept of intelligence. I don't agree with them that they have somehow captured the essence of the concept of intelligence in their formal definition though; I think they have just captured one aspect... -- Ben G On 1/14/08, Richard Loosemore [EMAIL PROTECTED] wrote: Pei Wang wrote: On Jan 13, 2008 7:40 PM, Richard Loosemore [EMAIL PROTECTED] wrote: And, as I indicated, my particular beef was with Shane Legg's paper, which I found singularly content-free. Shane Legg and Marcus Hutter have a recent publication on this topic, http://www.springerlink.com/content/jm81548387248180/ which is much richer in content. Unfortunately, this paper is not so much richer in content as containing a larger number of words and formulae. It adds nothing to the previous (poor quality) paper, falls into exactly the same pitfalls as before, and repeats the trick of pulling an arbitrary mathematical definition out of the hat without saying why this definition should correspond with the natural or commonsense definition. Any fool can mathematize a definition of a commonsense idea without actually saying anything new. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=85626959-5ab00c
[agi] Re: [singularity] The establishment line on AGI
Also, this would involve creating a close-knit community through conferences, journals, common terminologies/ontologies, email lists, articles, books, fellowships, collaborations, correspondence, research institutes, doctoral programs, and other such devices. (Popularization is not on the list of community-builders, although it may have its own value.) Ben has been involved in many efforts in these directions -- I wonder if he was thinking of Kuhn. Indeed, working toward the formation of such a community is one of the motivations underlying the AGI-08 conference. And also underlying the OpenCog AGI project I'm initiating together with the SIAI, see opencog.org My prior efforts in this direction, such as -- AGI email list -- 2006 AGI workshop -- two AGI edited volumes have been successful but smaller-scale. My feeling is that the time is ripe for the self-organization of a really viable AGI research community. In connection with AGI-08, we have put up a wiki page intended to gather proposals and suggestions regarding the formation of a more robust AGI community http://www.agi-08.org/proposals.php If any of y'all have relevant ideas, feel free to post them there. I don't actually have a lot of time for community-building activities, as my main focus is on Novamente LLC (and Novamente's work on AGI plus its narrow-AI consulting work that pays my bills). But, I try to make time for community-building, because I think it's very important and will benefit all of us working in the field. I did read Kuhn back in college, and was impressed with his insight, along with (even more so) that of Imre Lakatos, with his theory of scientific research programmes. In Lakatos's terms, what needs to be done is to build a community that can turn AGI into an overall progressive research program. I discuss these philosophy of science ideas a bit in the Hidden Pattern, and earlier in an essay http://www.goertzel.org/dynapsyc/2004/PhilosophyOfScience_v2.htm Further back, I remember when I was 5 years old, reading a draft of a book my dad was writing (a textbook of Marxist sociology), and encountering the word paradigm and not knowing what it meant. As I recall, I asked him and he tried to explain and I did not understand the explanation very well ;-p ... and truth be told, I still find it a fuzzy term, preferring Lakatos's characterization of research programmes. However, Kuhn had more insight than Lakatos into the sociological dynamics surrounding scientific research programmes... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=85618740-3a68d2
Re: Yawn. More definitions of intelligence? [WAS Re: [agi] Ben's Definition of Intelligence]
Your job is to be diplomatic. Mine is to call a spade a spade. ;-) Richard Loosemore I would rephrase it like this: Your job is to make me look diplomatic ;-p - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=85657842-0be0ab
Re: [agi] Ben's Definition of Intelligence
On definitions of intelligence, the canonical reference is http://www.vetta.org/shane/intelligence.html which lists 71 definitions. Apologies if someone already pointed out Shane's page in this thread, I didn't read every message carefully. An AGI definition of intelligence surely has, by definition! - to be general rather than complex and emphasize general problemsolving/learning. That seems to be what you actually mean. Mike: Obviously, my achieving complex goals in complex environments definition is intended to include generality. It could be rephrased as effectively achieving a wide variety of complex goals in various complex environments, with the general implicit in the wide. I also gave a math version of the definition in 1993, which is totally unambiguous due to being math rather than words. I have not bothered to look at the precise relations btw my older math definition and Shane Legg and Marcus Hutter's more recent math definition of intelligence. They are not identical but have a similar spirit. Intelligence has many dimensions. A crucial dimension of a true intelligence* is that it is general. It is a general problem-solver and general learner, able to solve, and learn how to solve, problems in many, and potentially infinite, domains - *without* being specially preprogrammed for any one of them. All computers to date have been specialists. The goal of Artificial General Intelligence is to create the first generalist. The problem with your above definition is that it uses terms that are themselves so extremely poorly-defined ;-) Arguably it rules out the brain, which is heavily preprogrammed by evolution in order to be good at certain things like vision, arm and hand movement, social interaction, language parsing, etc. And it does not rule out AIXItl type programs which achieve flexibility trivially, at the cost of utilizing unacceptably much computational resources... The reality is that achieving general intelligence given finite resources is probably always going to involve a combination of in-built biases and general learning ability. And where the line is drawn between in-built biases and preprogramming is something that current comp/cog-sci does not allow us to formally articulate in a really useful way. This is a subtle issue, as e.g. a program for carrying out a specific task, coupled with a general- purpose learner of the right level of capability, may in effect serve as a broader inductive bias helping with a wider variety of tasks. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=85169576-90c0ab
Re: [agi] Incremental Fluid Construction Grammar released
I'll be a lot more interested when people start creating NLP systems that are syntactically and semantically processing statements about words, sentences and other linguistic structures and adding syntactic and semantic rules based on those sentences. Depending on exactly what you mean by this, it's not a very far-off thing, and there probably are systems that do this in various ways. In a lexical grammar approach to NLP, most of the information about the grammar is in the lexicon. So all that's required for the system to learn new syntactic rules is to make the lexicon adaptive. For instance, in the link grammar framework, all that's required is for the AI to be able to edit the link grammar dictionary, which tells the syntactic link types associated with various words. This just requires a bit of abductive inference of the general form: 1) I have no way to interpret sentence S syntactically, yet pragmatically I know that sentence S is supposed to mean (set of logical relations) M 2) If word W (in sentence S) had syntactic link type L attached to it, then I could syntactically interpret sentence S to yield meaning M 3) Thus, I abductively infer that W should have L attached to it (with a certain level of probabilistic confidence) There is nothing conceptually difficult here, and nothing beyond the state of the art. The link grammar exists (among other frameworks), and multiple frameworks for abductive inference exist (including Novamente's PLN framework). The bottleneck is really the presence of data of type 1), i.e. of instances in which the system knows what a sentence is supposed to mean even though it can't syntactically parse it. One way to get a system this kind of data is via embodiment. But this is not the only way. It can also be done via pure conversation, for example. Suppose i'm talking to an AI, as follows: AI: What's your name Ben: I be Ben Goertzel AI: What?? Ben: I am Ben Goertzel AI: Thanks Now, the AI may not know the grammatical rule needed to parse I be Ben Goertzel But, after the conversation is done, it knows that the meaning is supposed to be equivalent to that of I am Ben Goertzel and thus it can edit it grammar (e.g. the link parser dictionary) appropriately, in this case to incorporate the Ebonic grammatical structure of be. Another way to provide training of type 1) would be if the system had a corpus of multiple different sentences all describing the same thing -- wherein it could parse some of the sentences and not others. In short, I feel that adapting grammar rules based on experience is not an extremely hard problem, though there are surely some moderate-level hidden gotchas. The bottlenecks in this regard appear to be -- getting the AI the experience -- boring old systems integration -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=84197075-424a2e
Re: [agi] Incremental Fluid Construction Grammar released
On Jan 10, 2008 10:26 AM, William Pearson [EMAIL PROTECTED] wrote: On 10/01/2008, Benjamin Goertzel [EMAIL PROTECTED] wrote: I'll be a lot more interested when people start creating NLP systems that are syntactically and semantically processing statements *about* words, sentences and other linguistic structures and adding syntactic and semantic rules based on those sentences. Note the new emphasis ;-) You example didn't have statements *about* words, but new rules were inferred from word usage. Well, here's the thing. Dictionary text and English-grammar-textbook text are highly ambiguous and complex English... so you'll need a very sophisticated NLP system to be able to grok them... OTOH, you could fairly easily define a limited, controlled syntax encompassing a variety of statements about words, sentences and other linguistic structures, and then make a system add syntactic and semantic rules based on these sentences. But I don't see what the point would be, because telling the system stuff in the controlled syntax would be basically as much work as explicitly encoding the rules... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=84320793-5fc1e6
Re: [agi] Incremental Fluid Construction Grammar released
Hi, Yes, the Texai implementation of Incremental Fluid Construction Grammar follows the phrase structure approach in which leaf lexical constituents are grouped into a structure (i.e. construction) hierarchy. Yet, because it is incremental and thus cognitively plausible, it should scale to longer sentences better than any non-incremental alternative. I agree that the incremental approach to parsing is the correct one, as opposed to the whole sentence at once approach taken in the link parser and most other parsers. However, this is really a quite separate issue from the choice of hierarchical phrase structure based grammar versus dependency grammar. For instance, Word Grammar is a dependency based approach that incorporates incremental parsing (but has not been turned into a viable computational system). -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=84321988-45541e
Re: [agi] Incremental Fluid Construction Grammar released
Do you plan to pay these non-experts, or recruit them as volunteers? ben On Jan 10, 2008 1:11 PM, Stephen Reed [EMAIL PROTECTED] wrote: Granted that from a logical viewpoint, using a controlled English syntax to acquire rules is as much work as explicitly encoding the rules. However, a suitable, engaging, bootstrap dialog system may permit a multitude of non-expert users to add the rules, thus dramatically reducing the amount of programmatic encoding, and the duration of the effort. That is my hypothesis and plan. -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: Benjamin Goertzel [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Thursday, January 10, 2008 11:06:45 AM Subject: Re: [agi] Incremental Fluid Construction Grammar released On Jan 10, 2008 10:26 AM, William Pearson [EMAIL PROTECTED] wrote: On 10/01/2008, Benjamin Goertzel [EMAIL PROTECTED] wrote: I'll be a lot more interested when people start creating NLP systems that are syntactically and semantically processing statements *about* words, sentences and other linguistic structures and adding syntactic and semantic rules based on those sentences. Note the new emphasis ;-) You example didn't have statements *about* words, but new rules were inferred from word usage. Well, here's the thing. Dictionary text and English-grammar-textbook text are highly ambiguous and complex English... so you'll need a very sophisticated NLP system to be able to grok them... OTOH, you could fairly easily define a limited, controlled syntax encompassing a variety of statements about words, sentences and other linguistic structures, and then make a system add syntactic and semantic rules based on these sentences. But I don't see what the point would be, because telling the system stuff in the controlled syntax would be basically as much work as explicitly encoding the rules... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; Never miss a thing. Make Yahoo your homepage. This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=84346888-b3c207
Re: [agi] Incremental Fluid Construction Grammar released
On Jan 10, 2008 10:03 PM, Matt Mahoney [EMAIL PROTECTED] wrote: All this discussion of building a grammar seems to ignore the obvious fact that in humans, language learning is a continuous process that does not require any explicit encoding of rules. I think either your model should learn this way, or you need to explain why your model would be more successful by taking a different route. Explicit encoding of grammars has a long history of failure, so your explanation should be good. At a minimum, the explanation should describe how humans actually learn language and why your method is better. Matt, If you read the paper at the top of this list http://www.novamente.net/papers/ you will see a brief summary of the reasoning behind the approach I am taking. It is only 8 pages long so it should be quick to read, though it obviously does not explain all details in that length. The abstract is as follows: * Abstract— Current work is described wherein simplified versions of the Novamente Cognition Engine (NCE) are being used to control virtual agents in virtual worlds such as game engines and Second Life. In this context, an IRC (imitation- reinforcement-correction) methodology is being used to teach the agents various behaviors, including simple tricks and communicative acts. Here we describe how this work may potentially be exploited and extended to yield a pathway toward giving the NCE robust, ultimately human-level natural language conversation capability. The pathway starts via using the current system to instruct NCE-controlled agents in semiosis and gestural communication; and then continues via integration of a particular sort of hybrid rule-based/statistical NLP system (which is currently partially complete) into the NCE-based virtual agent system, in such a way as to allow experiential adaptation of the rules underlying the NLP system, * I do not think that a viable design for an AGI needs to include a description of human learning (of language or anything else). No one understands exactly how the human brain works yet, but that doesn't mean we can't potentially have success with non-brain-emulating AGI approaches. My favorite theorists of human language are Richard Hudson (see his 2007 book Language Networks) and Tomassello (see his book Constructing a Language). I actually believe my approach to language in AGI is quite close to their ideas. But I don't have time/space to justify this statement in an email. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=84570254-afda8d
Re: [agi] Incremental Fluid Construction Grammar released
And how would a young child or foreigner interpret on the Washington Monument or shit list? Both are physical objects and a book *could* be resting on them. Sorry, my shit list is purely mental in nature ;-) ... at the moment, I maintain a task list but not a shit list... maybe I need to get better organized!!! Ben, your question is *very* disingenuous. Who, **me** ??? There is a tremendous amount of domain/real-world knowledge that is absolutely required to parse your sentences. Do you have any better way of approaching the problem? I've been putting a lot of thought and work into trying to build and maintain precedence of knowledge structures with respect to disambiguating (and overriding incorrect) parsing . . . . and don't believe that it's going to be possible without a severe amount of knwledge . . . . What do you think? OK... Let's assume one is working within the scope of an AI system that includes an NLP parser, a logical knowledge representation system, and needs some intelligent way to map the output of the latter into the former. Then, in this context, there are three approaches, which may be tried alone or in combination: 1) Hand-code rules to map the output of the parser into a much less ambiguous logical format 2) Use statistical learning across a huge corpus of text to somehow infer these rules [I did not ever flesh out this approach as it seemed implausible, but I have to recognize its theoretical possibility] 3) Use **embodied** learning, so that the system can statistically infer the rules from the combination of parse-trees with logical relationships that it observes to describe situations it sees [This is the best approach in principle, but may require years and years of embodied interaction for a system to learn.] Obviously, Cycorp has taken Approach 1, with only modest success. But I think part of the reason they have not been more successful is a combination of a bad choice of parser with a bad choice of knowledge representation. They use a phrase structure grammar parser and predicate logic, whereas I believe if one uses a dependency grammar parser and term logic, the process becomes a lot easier. So far as I can tell, in texai you are replicating Cyc's choices in this regard (phrase structure grammar + predicate logic). In Novamente, we are aiming at a combination of the 3 approaches. We are encoding a bunch of rules, but we don't ever expect to get anywhere near complete coverage with them, and we have mechanisms (some designed, some already in place) that can generalize the rule base to learn new, probabilistic rules, based on statistical corpus analysis and based on embodied experience. In our rule encoding approach, we will need about 5000 mapping rules to map syntactic parses of commonsense sentences into term logic relationships. Our inference engine will then generalize these into hundreds of thousands or millions of specialized rules. This is current work, research in progress. We have about 1000 rules in place now and will soon stop coding them and start experimenting with using inference to generalize and apply them. If this goes well, then we'll put in the work to encode the rest of the rules (which is not very fun work, as you might imagine). Emotionally and philosophically, I am more drawn to approach 3 (embodied learning), but pragmatically, I have reluctantly concluded that the hybrid approach we're currently taking has the greatest odds of rapid success. In the longer term, we intend to throw out the standalone grammar parser we're using and have syntax parsing done via our core AI processing -- but we're now using a standalone grammar parser as a sort of scaffolding. I note that this is not the main NM RD thrust right now -- it is at the moment somewhat separate from our work on embodied imitative/reinforcement/corrective learning of virtual agents. However, the two streams of work are intended to come together, as I've outlined in my paper for WCCI 2008, http://www.goertzel.org/new_research/WCCI_AGI.pdf -- Ben -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=83967477-a9e1c4
Re: [agi] Incremental Fluid Construction Grammar released
A perhaps nicer example is Get me the ball for which RelEx outputs definite(ball) singular(ball) imperative(get) singular(me) definite(me) _obj(get, me) _obj2(get, ball) and RelExToFrame outputs Bringing:Theme(get,me) Bringing:Beneficiary(get,me) Bringing:Theme(get,ball) Bringing:Agent(get,you) Note that the RelEx output is already abstracted and semantified compared to what comes out of a grammar parser. -- Ben On Jan 9, 2008 5:59 PM, Benjamin Goertzel [EMAIL PROTECTED] wrote: Can you give about ten examples of rules? (That would answer a lot of my questions above) That would just lead to really long list of questions that I don't have time to answer right now In a month or two, we'll write a paper on the rule-encoding approach we're using, and I'll post it to the list, which will make this approach clearer. Where did you get the rules? Did you hand-code them or get them from somewhere? As you know we have a system called RelEx that transforms the output of the link parser into higher-level semantic relationships. We then have a system of rules that map RelEx output into a set of frame-element relationships constructed mostly based on FrameNet. For the sentence Ben kills chickens RelEx outputs _obj(kill, chicken) present(kill) plural(chicken) uncountable(Ben) _subj(kill, Ben) and the RelExToFrame rules output Killing:Killer(kill,Ben) Killing:Victim(kill,chicken) Temporal_colocation:Event(present,kill) But I really don't have time to explain all the syntax and notation in detail... if it's not transparent... And I want to stress that I consider this kind of system pretty useless on its own, it's only potentially valuable if coupled with other components like we have in Novamente, such as an uncertain inference engine and an embodied learning system... Such rules IMO are mainly valuable to give a starting-point to a learning system, not as the sole or primary cognitive material of an AI system. And using them as a starting-point requires very careful design... The 5000 rules figure is roughly rooted in the 825 frames in FrameNet; each frame corresponds to a number of rules, most of which are related to specific verb/preposition combinations. Another way to look at it is that each rule corresponds roughly to a Lojban word/argument combination... pretty much, FrameNet and the Lojban dictionary are doing the same thing, which is to precisely specify commonsense subcategorization frames. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=83993607-803936
Re: [agi] Incremental Fluid Construction Grammar released
Can you give about ten examples of rules? (That would answer a lot of my questions above) That would just lead to really long list of questions that I don't have time to answer right now In a month or two, we'll write a paper on the rule-encoding approach we're using, and I'll post it to the list, which will make this approach clearer. Where did you get the rules? Did you hand-code them or get them from somewhere? As you know we have a system called RelEx that transforms the output of the link parser into higher-level semantic relationships. We then have a system of rules that map RelEx output into a set of frame-element relationships constructed mostly based on FrameNet. For the sentence Ben kills chickens RelEx outputs _obj(kill, chicken) present(kill) plural(chicken) uncountable(Ben) _subj(kill, Ben) and the RelExToFrame rules output Killing:Killer(kill,Ben) Killing:Victim(kill,chicken) Temporal_colocation:Event(present,kill) But I really don't have time to explain all the syntax and notation in detail... if it's not transparent... And I want to stress that I consider this kind of system pretty useless on its own, it's only potentially valuable if coupled with other components like we have in Novamente, such as an uncertain inference engine and an embodied learning system... Such rules IMO are mainly valuable to give a starting-point to a learning system, not as the sole or primary cognitive material of an AI system. And using them as a starting-point requires very careful design... The 5000 rules figure is roughly rooted in the 825 frames in FrameNet; each frame corresponds to a number of rules, most of which are related to specific verb/preposition combinations. Another way to look at it is that each rule corresponds roughly to a Lojban word/argument combination... pretty much, FrameNet and the Lojban dictionary are doing the same thing, which is to precisely specify commonsense subcategorization frames. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=83995036-45a6ce
Re: [agi] Incremental Fluid Construction Grammar released
Processing a dictionary in a useful way requires quite sophisticated language understanding ability, though. Once you can do that, the hard part of the problem is already solved ;-) Ben On Jan 9, 2008 7:22 PM, William Pearson [EMAIL PROTECTED] wrote: On 09/01/2008, Benjamin Goertzel [EMAIL PROTECTED] wrote: Let's assume one is working within the scope of an AI system that includes an NLP parser, a logical knowledge representation system, and needs some intelligent way to map the output of the latter into the former. Then, in this context, there are three approaches, which may be tried alone or in combination: 1) Hand-code rules to map the output of the parser into a much less ambiguous logical format 2) Use statistical learning across a huge corpus of text to somehow infer these rules [I did not ever flesh out this approach as it seemed implausible, but I have to recognize its theoretical possibility] 3) Use **embodied** learning, so that the system can statistically infer the rules from the combination of parse-trees with logical relationships that it observes to describe situations it sees [This is the best approach in principle, but may require years and years of embodied interaction for a system to learn.] Isn't there a 4th potential one? I would define the 4th as being something like 4) Use a language that can describe itself to bootstrap quickly new phrase usage. These can be seen in humans when processing dictionary/thesaurus like statements or learning a new language. The following paragraphs can be seen as examples of sentances that would need this kind of system to deal with and make use of the information in them: The word, on, can be used in many different situations. One of these is to imply one thing is above another and supported by it. The prefix dis can mean apart or break apart. Enchant can mean to take control by magical means. What might disenchant mean? * ---End examples It requires the system to be able to process this statement then add the appropriate rules. It may be tentative in keeping or using the rules, gathering information on how useful it finds it while processing text. It is different from handcoding, because it should enable anyone to add rules after a minimal set of language description language has been added. It should be combined with 3 however, so that rules don't always need to be given explicitly. I think this type of learning/instruction has the ability to be a lot quicker than any system that mainly relies on inference. I don't know of systems that are using this sort of thing. And it is a bit above the level I am working at, at the moment. Anyone know of systems that parse and then use sentances in this fashion? Will Pearson * I'm unsure how much work people are doing on the use of prefixes and suffixes to infer the meaning/usage of new words. I certainly use it a lot myself. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=84074230-e1fae9
Re: [agi] A Simple Mathematical Test of Cog Sci.
On Jan 7, 2008 9:12 AM, Mike Tintner [EMAIL PROTECTED] wrote: Robert, Look, the basic reality is that computers have NOT yet been creative in any significant way, and have NOT yet achieved AGI - general intelligence, - or indeed any significant rulebreaking adaptivity; (If you disagree, please provide examples. Ben keeps claiming/implying he's solved them or made significant advances, but when pressed never provides any indication of how). We all agree that AGI is not yet achieved. Space travel to Proxima Centauri is also not yet achieved, nor is human cloning ... there is a big difference in science between -- not yet achieved, but seems possible based on available knowledge and -- doesn't seem possible based on available knowledge If you are truly serious about solving these problems, I suggest, you should prepared to be hurt - you should be ready to consider truly radical ideas - for the ground on which you stand to be questioned - and be seriously shaken up. You should WELCOME any and all of your assumptions being questioned. Even if, let's say, what I or someone else suggests is in the end nutty, drastic ideas are good for you to contemplate at least for a while. Most of us on this list are already aware of the possibility that it is not possible to achieve high levels of intelligence using digital computer programs, given realistic space and time constraints. It is scientifically possible that Penrose is right, and to achieve human-like levels of intelligence in a machine, one needs to use a machine making use of weird, as yet poorly understood quantum gravity effects. However, at present, that Penrose-ean hypothesis does not seem that likely to most of us on this list; and given the current state of science, it's not a hypothesis that we really can explore in detail. Quantum gravity is in a confused state and quantum computing (let alone quantum gravity computing) is in its infancy. There is also always the possibility that the whole modern scientific world-view is deeply flawed in a way that is relevant to AGI. Maybe digital computers are unable to lead to human-level AI, for some reason totally unrelated to computability theory and quantum gravity and all that. There is plenty in the world that we don't understand -- I recommend Damien Broderick's recent and excellent book Outside the Gates of Science for anyone who doesn't agree But, this list is devoted to exploring the hypothesis that AGI **can** be achieved via creating intelligent machines -- and mainly, at the moment, to the hypothesis that it can be achieved via creating intelligent digital computer programs. We realize this hypothesis may be wrong, but it seems likely enough to us to merit a lot of attention and effort aimed at validation. Your supposed arguments against the hypothesis are nowhere near as original as you seem to think, and nearly everyone on this list has heard them before and not found them convincing. I read What Computers Can't Do by Hubert Dreyfus as a child in the 1970's and your diatribes don't seem to add anything to what he said there. If you think the whole digital-computer-AGI pursuit is a wrong direction and a waste of time, that's fine. But why do you feel the need to keep repeatedly informing us of this fact? For instance, I think string theory is probably wrong. But I don't see any point in spending my time trolling on string theory email lists and harping on this point repeatedly and confusingly. Let them explore their hypothesis... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82594941-c3bbc7
Re: [agi] A Simple Mathematical Test of Cog Sci.
On Jan 7, 2008 12:08 PM, David Butler [EMAIL PROTECTED] wrote: Would two AGI's with the same initial learning program, same hardware in a controlled environment (same access to a specific learning base-something like an encyclopedia) learn at different rates and excel in different tasks? Yes ... Even in the extreme case of identical external stimuli, two AGI systems could evolve slightly differently due to consequences of rounding error. However, if the AGI systems were built carefully enough (so as not to be susceptible to rounding error or other related phenomena) it could be made so that with totally identical environments they were totally identical in behavior, so long as no hardware failures occurred. (I note though that minor hardware failures like small defects in RAM or disk could always intervene and play the same role as roundoff error, potentially setting the two AGIs with identical code and identical environmental stimuli on different courses.) -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82647670-987d16
Re: [agi] Re: AGI-08 - Call for Participation
Nothing of that nature is planned at present ... as we the conference organizers are rather busy with other stuff, we've been pretty much fully whelmed with the organization of the First Life conference... It might be fun to do an in-world AGI meet-up a couple weeks after AGI-08, with an aim of discussing the AGI-08 papers... not sure how many paper authors would show up, though... -- Ben On Jan 7, 2008 2:07 PM, Bob Mottram [EMAIL PROTECTED] wrote: Will there be any AGI08 activities in Second Life? On 07/01/2008, Bruce Klein [EMAIL PROTECTED] wrote: quick agi-08 update... all 49 papers are now online and reg is open: www.agi-08.org/papers www.agi-08.org/register promo video: www.agi-08.org/video On Dec 11, 2007 1:51 PM, Bruce Klein [EMAIL PROTECTED] wrote: The First Conference on Artificial General Intelligence (AGI-08) March 1-3, 2008 at Memphis, Tennessee, USA Early Registration Deadline: January 31, 2008 Conference Website: http://www.agi-08.org Artificial General Intelligence (AGI) research focuses on the original and ultimate goal of AI --- to create intelligence as a whole. AGI seeks to create software or hardware systems that are generally intelligent in roughly the same sense that humans are, rather than being specialized problem-solvers such as most of the systems currently studied in the AI field. Current research in the AGI field is vigorous and diverse, exploring a wide range of possible paths, including theoretical and experimental computer science, cognitive science, neuroscience, and innovative interdisciplinary methodologies. AGI-08 is the very first international conference in this emerging field of science and engineering. The conference is organized with the cooperation of AAAI, and welcomes researchers and students in all relevant disciplines. Different from conventional conferences, AGI-08 is planned to be intensively discussion oriented. All the research papers accepted for publication in the Proceedings (49 papers total) will be available in advance online, so that attendees may arrive prepared to discuss the relevant issues with the authors and each other. The sessions of the conference will be organized to facilitate open and informed intellectual exchange on themes of common interest. Besides the technical discussions, time will also be scheduled at AGI-08 for an exploratory discussion of possible ways to work toward the formation of a more cohesive AGI research community -- including future conferences, publications, organizations, etc. After the two-and-half day conference, there will be a half day workshop on the broader implications of AGI technology, including ethical, sociological and futurological considerations. Yours, Organizing Committee, AGI-08 http://www.agi-08.org/organizing.php This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82707146-f12793
Re: [agi] Re: AGI-08 - Call for Participation
I'll forward this request to those who will be handling such things... thx ben On Jan 7, 2008 3:35 PM, Vladimir Nesov [EMAIL PROTECTED] wrote: Ben, I'm certainly not in position to ask for it, but if it's possible, can some kind of microphones be used during presentations on agi-08 (if someone is going to film it)? Audio was very poor in videos from previous events. -- Vladimir Nesovmailto:[EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82759215-d609f4
Re: [agi] A Simple Mathematical Test of Cog Sci.
On Jan 5, 2008 10:52 PM, Mike Tintner [EMAIL PROTECTED] wrote: I think I've found a simple test of cog. sci. I take the basic premise of cog. sci. to be that the human mind - and therefore its every activity, or sequence of action - is programmed. No. This is one perspective taken by some cognitive scientists. It does not characterize the field. (This has huge implications for AGI - you guys believe that an AGI must be programmed for its activities, I contend that free composition instead is essential for truly adaptive, general intelligence and is the basis of all animal and human activities). Spontaneous, creative self-organized activity is a key aspect of Novamente and many other AGI designs. So how to test cog sci? I contend that the proper, *ideal* test is to record humans' actual streams of thought about any problem - like, say, writing an essay - and even just a minute's worth will show that, actually, humans have major difficulties following anything like a joined-up, rational train of thought - or any stream that looks remotely like it could be programmed overall. A) While introspection is certainly a valid and important tool for inspiring work in AI and cog sci, it is not a test of anything. There is much empirical evidence showing that humans' introspections of their own cognitive processes are highly partial and inaccurate. For instance, if we were following the arithmetic algorithms that we think we are, there is no way the timing of our responses when solving arithmetic problems would come out the way they actually do. (I don't have the references for this work at hand, but I saw it years ago in the Journal of Math Psych I believe.) B) Whether something looks like it's following a simple set of rules doesn't mean much. Chaotic underlying dynamics can give rise to high-level orderly behavior; and simple systems of rules can give rise to apparently disorderly, incomprehensibly complex behaviors. Cf the whole field of complex-systems dynamics. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82365583-966081
Re: [agi] A Simple Mathematical Test of Cog Sci.
I don't really understand what you mean by programmed ... nor by creative You say that, according to your definitions, a GA is programmed and ergo cannot be creative... How about, for instance, a computer simulation of a human brain? That would be operated via program code, hence it would be programmed -- so would you consider it intrinsically noncreative? Could you please define your terms more clearly? thx ben On Jan 6, 2008 1:21 PM, Mike Tintner [EMAIL PROTECTED] wrote: MT: This has huge implications for AGI - you guys believe that an AGI must be programmed for its activities, I contend that free composition instead is essential for truly adaptive, general intelligence and is the basis of all animal and human activities). Ben: Spontaneous, creative self-organized activity is a key aspect of Novamente and many other AGI designs. Ben, You are saying that your pet presumably works at times in a non-programmed way - spontaneously and creatively? Can you explain briefly the computational principle(s) behind this, and give an example of where it's applied, (exploration of an environment, say)? This strikes me as an extremely significant, even revolutionary claim to make, and it would be a pity if, as with your analogy claim, you simply throw it out again without any explanation. And I'm wondering whether you are perhaps confused about this, (or I have confused you) - in the way you definitely are below. Genetic algorithms, for example, and suchlike classify as programmed and neither truly spontaneous nor creative. Note that Baum asked me a while back what test I could provide that humans engage in free thinking. He, quite rightly, thought it a scientifically significant claim to make, that demanded scientific substantiation. My test is not a test, I stress though, of free will. But have you changed your mind about this? It's hard though not a complete contradiction to believe in a mind being spontaneously creative and yet not having freedom of decision. MT: I contend that the proper, *ideal* test is to record humans' actual streams of thought about any problem Ben: While introspection is certainly a valid and important tool for inspiring work in AI and cog sci, it is not a test of anything. Ben, This is a really major - and very widespread - confusion. A recording of streams of thought is what it says - a direct or recreated recording of a person's actual thoughts. So, if I remember right, some form of that NASA recording of subvocalisation when someone is immediately thinking about a problem, would classify as a record of their thoughts. Introspection is very different - it is a report of thoughts, remembered at a later, often much later time. A record(ing) might be me saying I want to kill you, you bastard in an internal daydream. Introspection might be me reporting later: I got very angry with him in my mind/ daydream. Huge difference. An awful lot of scientists think, quite mistakenly, that the latter is the best science can possibly hope to do. Verbal protocols - getting people to think aloud about problems - are a sort of halfway house (or better). - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82398434-a3e5d5
Re: [agi] A Simple Mathematical Test of Cog Sci.
On Jan 6, 2008 4:00 PM, Mike Tintner [EMAIL PROTECTED] wrote: Ben, Sounds like you may have missed the whole point of the test - though I mean no negative comment by that - it's all a question of communication. A *program* is a prior series or set of instructions that shapes and determines an agent's sequence of actions. A precise itinerary for a journey. Even if the programmer doesn't have a full but only a very partial vision of that eventual sequence or itinerary. (The agent of course can be either the human mind or a computer). OK, then any AI that is implemented in computer software is by your definition a programmed AI. Whether it is based on GA's, neural nets, logical theorem-proving or whatever. So, is your argument that digital computer programs can never be creative, since you have asserted that programmed AI's can never be creative? -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82448475-4978a0
Re: [agi] A Simple Mathematical Test of Cog Sci.
Mike, The short answer is that I don't believe that computer *programs* can be creative in the hard sense, because they presuppose a line of enquiry, a predetermined approach to a problem - ... But I see no reason why computers couldn't be briefed rather than programmed, and freely associate across domains rather than working along predetermined lines. But the computer that is being briefed is still running some software program, hence is still programmed -- and its responses are still determined by that program (in conjunction w/ the environment, which however it perceives only thru a digital bit stream) I don't however believe that purely *digital* computers are capable of all the literally imaginative powers (as already discussed elsewhere) that are also necessary for true creativity and general intelligence. I don't know how you define a literally imaginative power. So, it seems like you are saying -- digital computer software can never truly be creative or possess general intelligence Is this your assertion? It is not an original one of course: Penrose, Dreyfus and many others have argued the same point. The latter paragraph of yours I've quoted could be straight out of The Emeperor's New Mind by Penrose. Penrose then notes that quantum computers can compute only the same stuff that digital computers can; so he posits that general intelligence is possible only for quantum gravity computers, which is what he posits the brain is. I think Penrose is most probably wrong, but at least I understand what he is saying... I'm just trying to understand what your perspective actually is... thx Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82464788-e73a96
Re: [agi] A Simple Mathematical Test of Cog Sci.
If you believe in principle that no digital computer program can ever be creative, then there's no point in me or anyone else rambling on at length about their own particular approach to digital-computer-program creativity... One question I have is whether you would be convinced that digital programs ARE capable of true creativity, by any possible actual achievements of digital computer programs... If a digital computer program made a great painting, wrote a great novel, proved a great theorem, patented dozens of innovative inventions, etc. -- would you be willing to admit it's creative, or would you argue that due to its digital nature, it must have achieved these things in a noncreative way? Ben On Jan 6, 2008 6:58 PM, Mike Tintner [EMAIL PROTECTED] wrote: Well we (Penrose co) are all headed in roughly the same direction, but we're taking different routes. If you really want the discussion to continue, I think you have to put out something of your own approach here to spontaneous creativity (your terms) as requested. Yes, I still see the mind as following instructions a la briefing, but only odd ones, not a whole rigid set of them., a la programs. And the instructions are open-ended and non-deterministically open to interpretation, just as my briefing/instruction to you - Ben go and get me something nice for supper - is. Oh, and the instructions that drive us, i.e. emotions, are always conflicting, e.g [Ben:] I might like to.. but do I really want to get that bastard anything for supper? Or have the time to, when I am on the very verge of creating my stupendous AGI? Listen, I can go on and on - the big initial deal is the claim that the mind isn't - no successful AGI can be - driven by a program, or thoroughgoing SERIES/SET of instructions - if it is to solve even minimal general adaptive, let alone hard creative problems. No structured approach will work for an ill-structured problem. You must give some indication of how you think a program CAN be generally adaptive/ creative - or, I would argue, squares (programs are so square, man) can be circled :). Mike, The short answer is that I don't believe that computer *programs* can be creative in the hard sense, because they presuppose a line of enquiry, a predetermined approach to a problem - ... But I see no reason why computers couldn't be briefed rather than programmed, and freely associate across domains rather than working along predetermined lines. But the computer that is being briefed is still running some software program, hence is still programmed -- and its responses are still determined by that program (in conjunction w/ the environment, which however it perceives only thru a digital bit stream) I don't however believe that purely *digital* computers are capable of all the literally imaginative powers (as already discussed elsewhere) that are also necessary for true creativity and general intelligence. I don't know how you define a literally imaginative power. So, it seems like you are saying -- digital computer software can never truly be creative or possess general intelligence Is this your assertion? It is not an original one of course: Penrose, Dreyfus and many others have argued the same point. The latter paragraph of yours I've quoted could be straight out of The Emeperor's New Mind by Penrose. Penrose then notes that quantum computers can compute only the same stuff that digital computers can; so he posits that general intelligence is possible only for quantum gravity computers, which is what he posits the brain is. I think Penrose is most probably wrong, but at least I understand what he is saying... I'm just trying to understand what your perspective actually is... - Release Date: 1/5/2008 11:46 AM - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=82484935-6a7f84
Re: [agi] NL interface
Matt, I agree w/ your question... I actually think KB's can be useful in principle, but I think they need to be developed in a pragmatic way, i.e. where each item of knowledge added can be validated via how useful it is for helping a functional intelligent agent to achieve some interesting goals... ben g What would you do with the knowledge base after you build it? I know this sounds like a dumb question, but Cyc has built a huge base of common sense knowledge in a structured format, but it isn't useful for anything. Of course that is not the result they anticipated. How will you avoid the same type of (very expensive) failure? What type of knowledge will it contain? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=80324636-31670d
Re: [agi] OpenCog
On Dec 28, 2007 5:59 AM, YKY (Yan King Yin) [EMAIL PROTECTED] wrote: OpenCog is definitely a positive thing to happen in the AGI scene. It's been all vaporware so far. Yes, it's all vaporware so far ;-) On the other hand, the code we hope to release as part of OpenCog actually exists, but it's not yet ready for opening-up as some of it needs to be extracted from the overall Novamente code base, and other parts of it need to be cleaned-up in various ways... Much of the reason for yakking about it months in advance of releasing it, was a desire to assess the level of enthusiasm for it. There are a number of enthusiastic potential OpenCog developers on the OpenCog mail list, so in that regard, I feel the response has been enough to merit proceeding with the project... I wonder what would be the level of participation? Time will tell! -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=79870666-e314ea
Re: [agi] OpenCog
On Dec 28, 2007 8:28 AM, Richard Loosemore [EMAIL PROTECTED] wrote: Benjamin Goertzel wrote: I wish you much luck with your own approach And, I would imagine that if you create a software framework supporting your own approach in a convenient way, my own currently favored AI approaches will not be conveniently explorable within it. That's the nature of framework-building. Actually, that would be a serious miusunderstanding of the framework and development environment that I am building. Your system would be just as easy to build as any other. My purpose is to create a description language that allows us to talk about different types of AGI system, and then construct design variations autonmatically. I don't believe it is possible to create a framework that both a) is unbiased regarding design type b) makes it easy to construct AGI designs Just as different programming languages are biased toward different types of apps, so with different AGI frameworks... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=79885135-d592af
Re: [agi] OpenCog
Loosemore wrote: I am sorry, but I have reservations about the OpenCog project. The problem of building an open-source AI needs a framework-level tool that is specifically designed to allow a wide variety of architectures to be described and expressed. OpenCog, as far as I can see, does not do this, but instead takes a particular assortment of mechanisms as its core, then suggests that people add modules onto this core. This is not a framework-level approach, but a particular-system approach that locks all future work into the limitations of the initial core. For example, I have many, many AGI designs that I need to explore, but as far as I can see, none of them can be implemented at all within the OpenCog system. I would have to rewrite OpenCog completely to get it to meet my needs. Hi Richard, To be sure, OpenCog is not intended to be equally useful for all possible AGI approaches. To provide something equally useful for all AGI approaches, one would need to make something extremely broad -- basically, one would need to make a highly general-purpose operating-system and/or programming-language, rather than a specific software framework. OpenCog is designed to support a certain family of AGI designs, but is not designed to conveniently support all possible AGI designs. Definitely, there is room in the world for more than one AGI framework. As an example the CCortex platform seems like it may be a good framework within which to build biologically realistic NN based AGI systems (note, this is based on their literature only, I've never tried their system). I wish you much luck with your own approach And, I would imagine that if you create a software framework supporting your own approach in a convenient way, my own currently favored AI approaches will not be conveniently explorable within it. That's the nature of framework-building. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=79828215-b4b8b5
[agi] OpenCog
Re the recent discussion of OpenCog -- this recent post I made to the OpenCog mailing list may perhaps help clarify the intentions underlying the project further. -- Ben -- Forwarded message -- From: Benjamin Goertzel [EMAIL PROTECTED] Date: Dec 27, 2007 11:07 AM Subject: Re: Project Questions To: [EMAIL PROTECTED] On Dec 27, 2007 9:27 AM, Pat [EMAIL PROTECTED] wrote: Ben, Thank you for the link to the paper. My thinking has been that in order to develop a machine capable of general intelligence, a specification would need to be developed which outlines the functionality of a thinking machine, separate from any implementation issues. I totally agree -- and, I have tried to do that in a 350-page manuscript, which I plan to release online sometime in the first half of 2008. HOWEVER, even though I am a big fan of my AI design, it's obvious that there are going to be many lessons learned during the course of working out more detailed designs of subcomponents and experimenting with implementations. This is a useful conversation, because I'm seeing that it in talking about OpenCog it will be valuable to distinguish -- OpenCog core -- Specific AGI designs that can be built on the OpenCog core, generally in a modular fashion (each AGI system comprising a certain set of MindAgents and a certain set of functional units) The OpenCog core may be used for a load of different AGI designs My own AGI design is one particular design that can be built on the OpenCog core. The AGI design that I will advocate for building on top of the OpenCog core is a variant of the Novamente design, I'm not sure what to call it, but it will get a name before the OpenCog launch... However, I also want to explicitly encourage the creation of other AGI designs on top of OpenCog. Hopefully there can be crosspollination of different approaches. I can see where your approach of taking diverse contributions of software and integrating them around a framework could be instrumental in the exploration and discovery of the specification (among other benefits). As I hope I've clarified above, -- I do have a fairly precise specification I'm interested in using OpenCog to explore, which is closely related (but not identical) to my Novamente specification -- However, I don't intend OpenCog to be restricted to the implementation. exploration of this specification of mine Thanks for your questions, they are certainly good ones... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=79828670-a37b2c
Re: [agi] BMI/BCI Growing Fast
I think that at first sight this goes to support my position in the original argument with Ben- namely that there are all kinds of ways to get at or read minds, and there is now an increasing momentum to do that. Being able to read the stream of subvocalizations coming out from a person's mind, is a very very very long way from being able to read the internal cognitive dynamics of a person's thoughts. And, the technology used for the former does not seem capable of being incrementally extended to do the latter. I agree that mind-reading will happen, and that the pace of growth of mind-reading related technologies is exponential. But still you seem to be a bit overoptimistic about the value of the exponent. (Although, I don't discount the possibility of some wild outlier innovation coming along...) -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=79286730-5a7369
[agi] AGI, NLP, embodiment and gesture
Hi all, Here you'll find a paper http://goertzel.org/new_research/WCCI_AGI.pdf that I've submitted to the WCCI 2008 Special Session on Human-level AI. It tries to summarize the big picture about how advanced AI can be achieved via synthesizing NLP and virtual embodiment... The paper refers to another paper An Integrative Methodology for Teaching Embodied Non-Linguistic Agents which is linked to from http://www.agi-08.org/schedule.php -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=79438803-5dc039
[agi] Mizar translated to TPTP !
For those interested in automated theorem-proving, I'm pleased to announce a major advance in tools has occurred... The Mizar library of formalized math has finally been translated into a sensible format, usable for training automated theorem-proving systems ;-) Josef Urban informed me that * I am now writing an extended version of a paper describing export of Mizar proofs to TPTP (see http://www.springerlink.com/content/t88848500815t188/ for the conference version), and have just remembered your request for having Mizar proofs exported to KIF or TPTP. So yes, I have translated it all to TPTP derivations, which can be ATP-verified by tools like GDV (see the paper). The page which presents it all is http://www.cs.miami.edu/~tptp/MizarTPTP/ , the TSTP icons there give you the TPTP derivations corresponding to each Mizar proof (see http://www.activemath.org/workshops/MathUI/07/proceedings/Urban-etal-MizarIDV-MathUI07.pdf for explanation of the site). If you want tar.gz of all the proofs, some version is at http://lipa.ms.mff.cuni.cz/~urban/nd_problems1.tar.gz (watch out, it is huge). There are still some completness bugs, the biggest is lack of explicit arithmetic evaluations done by mizar (inferences like 4+5=9 that dumb FOL ATP system will not prove), but otherwise it mostly works (see the paper). * TPTP, unlike Mizar syntax, is a small and straightforward BNF grammar which should be translatable into the internal KR of any AI system with a formal-logic-based aspect, with only moderate effort. So we can now import all of undergrad math and some grad-level math into AI systems, which provides the possibility of doing inductive reasoning on a large corpus of proofs to teach AI systems how to prove theorems... -- Ben G On Mon, 19 Mar 2007, Josef Urban wrote: Ben, the MPTP export keeps at this moment quite a lot of the original Mizar proof structure in the TPTP 'useful info' slot, but not all of it. Depending on what kind of learning you want to do, it might or might not suffice. The biggest ommission is probably the lack of further description of unproved propositions (precisely in TPTP syntax: fofs with role 'unknown' and mptp_info(_,_,proposition,_,_)). Most of them are probably natural deduction (ND) assumptions, but some are not (e.g. propositions about constants introduced by the 'consider' keyword - they can be rather thought of as definitions). It should not be difficult to add the missing info there - the TPTP format is produced directly from the XML, which contains all the proof structure. There have been several reasons for postponing this so far - my focus on getting the reproving of the simple steps right (which is sort of a precondition for ATP cross-verification of Mizar, which in turn is a precondition for productive use of deductive tools like ATP as part of larger knowledge-based systems tailored for Mizar), and also a focus on getting the reproving of theorems from their external references right, which in a sense gives you a large-scale proof structure (which is not only usable for learning, but also - unlike the ND internal proofs - understandable to ATPs, and thus allowing things like the MPTP challenge). Another reason was that I did not want to decide about the details of MPTP ND annotations, until I decided about the export of ND proof structure to TPTP. The latter accidentally happened last week (not only as a next step for the cross-verification, but also as a megalomanic plan to build the detailed MML DAG with milions of nodes :-). So some missing annotations will appear in the next few weeks (maybe even days), and more importantly, the Mizar ND proofs will become TPTP proofs (if needed without any ND - though there is a good chance that assumptions will become acknowledged and processed parts of TPTP proofs). To sum up: - you can have 'raw Mizar' loaded by using the XML - that gives you access to the formulas and Mizar proof structure; the disadvantage is that for any deductive tool which you might want to use, you'll have to define the translation to its logic - there is the 'raw MPTP', with formulas in extended TPTP syntax containing the mptp_info annotations; this is sort of a middle way between the XML and standard TPTP; the annotations are likely to get a bit better in near future, what they annotate is still the Mizar ND structure - there are pre-generated 'standard TPTP' problem sets in the MPTP distro and the MPTPChallenge distro; these you can feed to ATP systems, and also learning systems (the symbols and proposition names are stable there - always the same semantics); for quite a lot of them an ATP proof can be found (and used for learning) - there will (hopefully soon) be a full TPTP (i.e. mostly/completely non-ND) proof structure export, compatible with the proofs produced by ATP systems like EP. Josef On Sun, 18 Mar 2007, Ben Goertzel wrote: Josef, Thanks for your reply! However, I'm not
Re: [agi] BMI/BCI Growing Fast
I would add that the Chinese universities are extremely eager to recruit Western professors to lead research labs in AI and other areas. Hugo DeGaris relocated there a year or so ago, and is quite relieved to be supplied with a bunch of excellent research assistants and loads of computational firepower for his work on neural nets and FPGA's ... he'd had a rough ride for a while, what with the bankruptcy of his Belgian employer Starlab, and a 6-year stint at Utah State University where he was unable to get significant US gov't research funding... When I talked to various university administrators in Wuhan (where Hugo is), it was quite clear that if I wanted to relocate there, I would have access to an essentially unlimited number of research programmers to help with my AI projects. Without needing to constantly write grant applications and compete for funds. Novamente LLC is in an exciting phase right now though; and my personal situation would make it difficult for me to relocate to China ... but it's interesting to know that backup plan is there... -- Ben G On Dec 15, 2007 5:47 AM, Gary Miller [EMAIL PROTECTED] wrote: Ben said That is sarcasm ... however, it's also my serious hypothesis as to why the Chinese gov't doesn't mind losing their best brightest... It may also be that China understands too that as more Chinese become Americans, China will have a greater exposure and political lobby within the United States. Look at how much political influence Israel now exerts within the United States government and corporations. Also as with other minorities, the more exposure that Americans have to them in their everyday life the less fear and distrust that will be experienced. As the Chinese people which I know have entered into higher end professional roles in the United States they are eager to form business alliances with company's back home in China. China is also still feeling great population pressure. I just returned from meeting my fiancé there and in the cities where I stayed it still felt very overpopulated by my standards. Even though they possess excellent mass transit, people are packed in buses like sardines and more people move from the countryside to the city everyday to find work. I was only there for ten days so I did not gain a lot of understanding of how they manage to keep everything running. But in just that short time I saw that they have the same drug, homelessness, and poverty problems that we have here. The vast majority of people I met there were very friendly towards Americans and even though I know there have to be a lot of us there, because I was not in the tourist areas, I could go two or three days without seeing another American. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76427523-f0fb03
Re: [agi] BMI/BCI Growing Fast
Mike, My comment is that this is GREAT research and development, but, for the near and probably medium future is very likely to be about perception and action rather than cognition. I.e., we are sort of on the verge of understanding how to hook up new sensors to the brain, and hook the brain up to new actuators. We're not there yet except in some pretty simple cases, but we're getting there. And that's exciting! But, we're still quite clueless about how to, say, hook the brain up to a calculator or to Google in a useful way... due to having a vastly insufficiently detailed knowledge of how the brain carries out cognitive operations... -- Ben G On Dec 14, 2007 5:22 AM, Mike Tintner [EMAIL PROTECTED] wrote: [Comments?] Brain-computer link systems on the brink of breakthrough, study finds Systems that directly connect silicon circuits with brains are under intensive development all over the world, and are nearing commercial application in many areas, according to a study just placed online. Neurobiologist Theodore W. Berger of the University of Southern California chaired the eight-member committee which compiled the International Assessment of Research and Development in Brain-Computer Interfaces, published in October by the World Technology Evaluation Center, Inc., of Baltimore MD Berger, who holds the David Packard Chair at the USC Viterbi School of Engineering and is Director of the USC Center for Neural Engineering contributed the introduction and two chapters of the report, which encompassed dozens of research institutes in Europe and Asia. The other committee members (and chapter authors) included John K. Chapin (SUNY Downstate Medical Center); Greg A. Gerhardt (University of Kentucky); Dennis J. McFarland (Wadsworth Center); José C. Principe (University of Florida); Dawn M. Taylor (Case Western Reserve); and Patrick A. Tresco (University of Utah). The report contains three overall findings on Brain-Computer Interface (BCI) work worldwide: -- BCI research is extensive and rapidly growing, as is growth in the interfaces between multiple key scientific areas, including biomedical engineering, neuroscience, computer science, electrical and computer engineering, materials science and nanotechnology, and neurology and neurosurgery. -- BCI research is rapidly approaching first-generation medical practice-clinical trials of invasive BCI technologies and significant home use of noninvasive, electroencephalography (EEG-based) BCIs. The panel predicts that BCIs soon will markedly influence the medical device industry, and additionally BCI research will rapidly accelerate in non-medical arenas of commerce as well, particularly in the gaming, automotive, and robotics industries. -- The focus of BCI research throughout the world was decidedly uneven, with invasive BCIs almost exclusively centered in North America, noninvasive BCI systems evolving primarily from European and Asian efforts. BCI research in Asia, and particularly China, is accelerating, with advanced algorithm development for EEG-based systems currently a hallmark of China's BCI program. Future BCI research in China is clearly developing toward invasive BCI systems, so BCI researchers in the US will soon have a strong competitor. The chapters of the report offer detailed discussion of specific work from around the world, work on Sensor Technology, Biotic-Abiotic Interfaces, BMI/BCI Modeling and Signal Processing, Hardware Implementation, Functional Electrical Stimulation and Rehabilitation Applications of BCIs, Noninvasive Communication Systems, Cognitive and Emotional Neuroprostheses, and BCI issues arising out of research organization-funding, translation-commercialization, and education and training. With respect to translation and commercialization, the Committee found that BCI research in Europe and Japan was much more tightly tied to industry compared to what is seen in the U.S., with multiple high-level mechanisms for jointly funding academic and industrial partnerships dedicated to BCIs, and mechanisms for translational research that increased the probability of academic prototypes reaching industrial paths for commercialization. The report is now downloadable online at the WTEC website, at http://www.wtec.org/bci/BCI-finalreport-10Oct2007-lowres.pdf http://www.wtec.org/bci/BCI-finalreport-10Oct2007-lowres.pdf Source: University of Southern California http://www.physorg.com/news116764966.html http://www.physorg.com/news116764966.html - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76030919-edd895
Re: [agi] BMI/BCI Growing Fast
Mike wrote: Personally, my guess is that serious mindreading machines will be a reality in the not too distant future - before AGI and seriously autonomous mobile robots. No way. Tell that to the neuroscientists in your local university neuro lab, and they'll get a good laugh ;-) The future course of AGI is contentious and uncertain (though I have my own strong opinions), but it seems very clear that mobile robotics tech is more advanced than mind-reading-type brain-imaging, and apt to progress far faster... We have no f***-ing idea how to read thoughts from the brain right now -- whereas we do have a pretty well worked out theory of probabilistic robotics, and are in a phase of optimizing and tuning and generalizing and getting the details right... What is the empirical grounds for your optimism? Just to be clear: I am sure that mindreading technology is coming, it's your relative timing estimate that perplexes me... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76052095-6b5f73
Re: [agi] BMI/BCI Growing Fast
Hi, From Bob Mottram on the AGI list: However, I'm not expecting to see the widespread cyborgisation of human society any time soon. As the article suggests the first generation implants are all devices to fulfill some well defined medical need, and will have to go through all the usual lengthy testing procedures before they're generally accepted. Only after the initial medical phase which could last several decades will brain implants be sufficiently inexpensive and be considered sufficiently safe that people start to think about using these things as a lifestyle, work or leisure enhancement rather as cosmetic surgery is today. Hmmm... it's interesting to speculate, though... If it were possible to wire a calculator into the brain, this could dramatically increase the effectiveness of certain kinds of work, right? So, if a certain nation were to make laws allowing this, and to encourage research into this, then potentially they could gain a dramatic advantage over other nations... There does therefore seem a possibility for a brain enhancement race if a case is made to some national government that within say 10-20 years effort a massively productivity-increasing brain-enhancement could be made. This is not really an AGI topic, though, so I'm cross-posting to the Singularity list and I think discussion should continue there if anyone feels like it. (Though the topic may be sufficiently obvious not to need follow-up discussion...) -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76046209-29ce80
Re: [agi] BMI/BCI Growing Fast
Bear in mind that science has used very little imagination here to date. Science only started studying consciousness ten years ago. It still hasn't started studying Thought - the actual contents of consciousness: the streams of thought inside people's heads. In both cases, the reason has been sheer prejudice and nothing to do with true science. That is absurd -- the reason is that brain-scanning doesn't work very well; and designing informative rigorous psych lab experiments to measure aspects of cognition is really hard I'm confident that well within the next 10 years, science will a) recognize Thought as a vital area of study (with the same mushrooming of study that took place with Consciousness, if not larger) and will b) understand why Thought is so important - above all, to improve human thinking. What do you think the mushrooming of Cognitive Science during the last decade has been? Science does recognize thought as a critical area of study. It's just a difficult thing to study. All right, I'm going to stop taking the bait of your absurd claims and statements, Mike (for a while at least) ... it's tempting to fall into the trap of correcting silly statements that pour into my Inbox, but it's burning too much of my time, even though I read and type quite fast... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76106993-9e1bad
Re: [agi] BMI/BCI Growing Fast
Mike: Making the general public smarter is not in the best interest of government, who wants to keep us fat dumb and (relatively) happy (read: distracted). If we're not making people smarter with currently available resources, why would we invest in research to discover expensive new technologies to make people smarter? We need that money to invest in research for expensive new technologies to allow people to be lazier. You are thinking mostly about the USA, it seems. I was thinking mostly about the People's Republic of China. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76069170-430555
Re: [agi] BMI/BCI Growing Fast
Is China pushing its people into being smarter? Are they giving incentives beyond the US-style capitalist reasons for being smart? The incentive is that if you get smart enough, you may figure out a way to get out of China ;-) Thus, they let the top .01% out, so as to keep the rest of the top 1% motivated by the hope of getting out. Clever, huh? ben g - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76321943-06efd5
Re: [agi] BMI/BCI Growing Fast
That is sarcasm ... however, it's also my serious hypothesis as to why the Chinese gov't doesn't mind losing their best brightest... -- Ben On Dec 14, 2007 7:35 PM, Robin Gane-McCalla [EMAIL PROTECTED] wrote: Is that sarcasm or an official Communist Party platform? On Dec 14, 2007 3:48 PM, Benjamin Goertzel [EMAIL PROTECTED] wrote: Is China pushing its people into being smarter? Are they giving incentives beyond the US-style capitalist reasons for being smart? The incentive is that if you get smart enough, you may figure out a way to get out of China ;-) Thus, they let the top .01% out, so as to keep the rest of the top 1% motivated by the hope of getting out. Clever, huh? ben g - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; -- Robin Gane-McCalla YIM: Robin_Ganemccalla AIM: Robinganemccalla - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=76335876-20d29a
Re: [agi] The Function of Emotions is Torture
Mike In case you're curious I wrote down my theory of emotions here http://www.goertzel.org/dynapsyc/2004/Emotions.htm (an early version of text that later became a chapter in The Hidden Pattern) Among the conclusions my theory of emotions leads to are, as stated there: * * AI systems clearly will have emotions * Their emotions will include, at least, happiness and sadness and spiritual joy * Generally AI systems will probably experience less intense emotions than humans, because they can have more robust virtual multiverse modeling components, which are not so easily bollixed up – so they'll less often have the experience of major non-free-will-related mental-state shifts * Experiencing less intense emotions does not imply experiencing less intense states of consciousness. Emotion is only one particular species of state-of-consciousness. * The specific emotions AI systems will experience will probably be quite different from those of humans, and will quite possibly vary widely among different AI systems * If you put an AI in a human-like body with the same sorts of needs as primordial humans, it would probably develop every similar emotions to the human ones * -- Ben On Dec 12, 2007 9:27 PM, Mike Tintner [EMAIL PROTECTED] wrote: I don't think you've answered my point - which perhaps wasn't put well enough. All you propose, as far as I can see, is to apply *values* to behaviour - to apply positive and negative figures to behaviours considered beneficial or detrimental, and thus affect the system's further behaviour - reinforcing it, for example. It is more or less like a value approach to investing on stocks on the stockmarket - when their value goes up or down, a formula determines whether the system buys more or less shares. But this is a purely numerical approach to altering behaviour. There is nothing a priori wrong with it - although, in fact, (although this is a more complex argument which I won't really go into), it would never actually work for AGI, which has to deal with problems where it is impossible to apply precise or reliable values. But the important point here is that these *values* are not *emotions* at all. They're fundamentally different entities and affect behaviour in fundamentally different ways - your values, for example, will not cause any pleasure or pain to a self, or have a corporeal, hormonal nature, or conflict. You, like others, are trying to invest your value system with a complexity and dignity that it simply hasn't got and has no right to. It's absurd - you might just as well talk of every plus or minus sign in a mathematical calculation as conferring pleasure or pain. It also shows a very limited understanding of emotions. Matt: Mike Tintner [EMAIL PROTECTED] wrote: Matt: I don't believe that the ability to feel pleasure and pain depends on consciousness. That is just a circular definition. http://en.wikipedia.org/wiki/Philosophical_zombie Richard:It is not circular. Consciousness and pleasure/pain are both subjective issues. They can resolved together. Both of you, in fairly standard fashion, are approaching humans and animals as if they were dissected on a table with consciousness/ emotions/ pleasure pain lying around. The reality is that we are integrated systems in which - a self is continually subjected to and feels (or to some extent may choose not to feel) emotions (involving pleasure/pain) via a (two-way) nervous system. The questions Matt has to answer is: 1) are the systems you envisage going to have a self (to feel emotions) - and if so, why? No, I am proposing a measure of reinforcement for intelligence in general, whether human, animal, or machine, all of which fall under Legg and Hutter's universal intelligence ( http://www.vetta.org/documents/ui_benelearn.pdf ), which is based on Hutter's AIXI model ( http://www.hutter1.net/ai/aixigentle.htm ). In this model, an agent and an environment are modeled by a pair of interactive Turing machines exchanging symbols. In addition, the environment sends a utility or reinforcement signal to the agent at each step. The goal of the agent is to maximize the accumulated utility. The paper on universal intelligence (UI) proposes defining intelligence as the expected accumulated utility for a randomly chosen environment (from a Solomonoff distribution of environments, i.e. self delimiting Turing machines chosen by coin flips). Hutter's AIXI model shows that the most intelligent strategy is to guess at each step that the environment is simulated by the shortest program consistent with the observed interaction so far. However, AIXI is not computable. In humans, it is natural to think of positive utility or reinforcement as a reward signal or pleasure, and negative utility as a penalty, such as pain. In this respect, humans seek to maximize expected
Re: [agi] The same old explosions?
Self-organizing complexity and computational complexity are quite separate technical uses of the word complexity, though I do think there are subtle relationships. As an example of a relationship btw the two kinds of complexity, look at Crutchfield's work on using formal languages to model the symbolic dynamics generated by dynamical systems as they approach chaos. He shows that as the parameter values of a dynamical system approach those that induce a chaotic regime in the system, the formal languages implicit in the symbolic-dynamics representation of the system's dynamics pass through more and more complex language classes. And of course, recognizing a grammar in a more complex language class has a higher computational complexity. So, Crutchfield's work shows a connection btw self-organizing complexity and computational complexity, via the medium of formal languages and symbolic dynamics. As another, more pertinent example, the Novamente design seeks to avoid the combinatorial explosions implicit in each of its individual AI learning/reasoning components, via integrating these components together in an appropriate way. This integration, via its impact on the overall system dynamics, leads to a certain degree of complexity in the self-organizing-systems sense -- Ben G On Dec 11, 2007 10:09 AM, Richard Loosemore [EMAIL PROTECTED] wrote: Mike Tintner wrote: Essentially, Richard others are replaying the same old problems of computational explosions - see computational complexity in this history of cog. sci. review - no? No: this is a misunderstanding of complexity unfortunately (cf the footnote on p1 of my AGIRI paper): computational complexity refers to how computations scale up, which is not at all the same as the complexity issue, which is about whether or not a particular system can be explained. To see the difference, imagine an algorithm that was good enough to be intelligent, but scaling it up to the size necessary for human-level intelligence would require a computer the size of a galaxy. Nothing wrong with the algorithm, and maybe with a quantum computer it would actually work. This algorithm would be suffering from a computational complexity problem. By contrast, there might be proposed algorithms for iimplementing a human-level intelligence which will never work, no matter how much they are scaled up (indeed, they may actually deteriorate as they are scaled up). If this was happening because the designers were not appreciating that they needed to make subtle and completely non-obvious changes in the algorithm, to get its high-level behavior to be what they wanted it to be, and if this were because intelligence requires complexity-generating processes inside the system, then this would be a complex systems problem. Two completely different issues. Richard Loosemore Mechanical Mind Gilbert Harman Mind as Machine: A History of Cognitive Science. Margaret A. Boden. Two volumes, xlviii + 1631 pp. Oxford University Press, 2006. $225. The term cognitive science, which gained currency in the last half of the 20th century, is used to refer to the study of cognition-cognitive structures and processes in the mind or brain, mostly in people rather than, say, rats or insects. Cognitive science in this sense has reflected a growing rejection of behaviorism in favor of the study of mind and human information processing. The field includes the study of thinking, perception, emotion, creativity, language, consciousness and learning. Sometimes it has involved writing (or at least thinking about) computer programs that attempt to model mental processes or that provide tools such as spreadsheets, theorem provers, mathematical-equation solvers and engines for searching the Web. The programs might involve rules of inference or productions, mental models, connectionist neural networks or other sorts of parallel constraint satisfaction approaches. Cognitive science so understood includes cognitive neuroscience, artificial intelligence (AI), robotics and artificial life; conceptual, linguistic and moral development; and learning in humans, other animals and machines. click for full image and caption Among those sometimes identifying themselves as cognitive scientists are philosophers, computer scientists, psychologists, linguists, engineers, biologists, medical researchers and mathematicians. Some individual contributors to the field have had expertise in several of these more traditional disciplines. An excellent example is the philosopher, psychologist and computer scientist Margaret Boden, who founded the School of Cognitive and Computing Sciences at the University of Sussex and is the author of a number of books, including Artificial Intelligence and Natural Man (1977) and The Creative Mind (1990). Boden has been active in cognitive science pretty much from the start and has known many of
Re: [agi] AGI communities and support
Thanks Bob. But I meant, it looks more likely that robots will achieve - and have already taken the first concrete steps to achieve - the goals of AGI - the capacity to learn a range of abilities and activities. Can you point to any single robot that has demonstrated the capability to learn a range of abilities and activities? I don't think you can. It seems to me that all you're saying is, basically, that robots have been used to do a lot of different things. So have software programs. But, a collection of highly specialized intelligences does not comprise a general intelligence. It may be (one way to make) a significant part of a general intelligence architecture, but if so, I would argue the hard part is being left out... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73930833-e36fbf
Re: [agi] AGI communities and support
Yes I expect to see more narrow AI robotics in future, but as time goes on there will be pressures to consolidate multiple abilities into a single machine. Ergonomics dictates that people will only accept a limited number of mobile robots in their homes or work spaces. Physical space is at a premium, and you don't want multiple devices getting in your way all the time. In a similar manner it's inconvenient to have to carry around multiple electronic gadgets - really you just want a single gadget which does lots of things. My prediction is that, when robotics firms finally get around to working on multi-purpose robots that need to carry out multiple intersecting, interacting activities in messy real-world environments -- THEN they will all of a sudden become intensively interested in work on cognitive architecture and AGI. Work that is of very limited interest to them now since they are working on robots with very specialized functionalities. I don't think that a robot assembling parts in a factory has any more to do with AGI than Google's statistical NLP engine does -- probably less. Embodiment may be a very useful ingredient for AGI systems, but, it's quite possible to do narrow robotics and that is what nearly all roboticists are doing, for the same practical reasons that nearly all AI software researchers are doing narrow AI. I agree that the DARPA challenge robots are interesting and are pushing more in the AGI direction than nearly any commercial robotics. It's worthy of note that the most interesting robotics work, from an AGI view, seems to be either -- gov't funded research stuff like the DARPA challenge (most entrants are universities whose research projects live off gov't funding) -- blue-sky RD projects by big companies with money to burn, such as Honda's Asimo project I.e. the AGI meets robotics meme is VERY far from affecting the commercial robotics industry, it would seem. Which is a shame. I would like to break past this barrier by taking Webkinz as an inspiration, and making intelligent robotic toys that do most of their thinking on a remote server farm (and interoperate with intelligent agents in virtual worlds). But this is going to be a number of years in coming, I'm sure The cost of really good robotics equipment remains high (from the perspective of a commercially viable robot toy), though decreasing rapidly. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73963833-0c76f0
Re: [agi] AGI communities and support
So I reckon roboticists ARE actually focussed on an AGI challenge - whereas, as I've pointed out before, there is nothing comparable in pure AGI. To my knowledge, none of the work on the ICRA Robotic Challenge is at this point taking a strong AGI approach And with all those millions of investment bucks - I expect to see some results/ genuine progress in the not too distant future. Millions of investment bucks doesn't go that far in robotics, unfortunately. I hope to see progress too, but I believe you're way optimistic about the current state of robotics research. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73975970-c08d44
Re: [agi] Do we need massive computational capabilities?
On Dec 7, 2007 7:09 AM, Mike Tintner [EMAIL PROTECTED] wrote: Matt,:AGI research needs special hardware with massive computational capabilities. Could you give an example or two of the kind of problems that your AGI system(s) will need such massive capabilities to solve? It's so good - in fact, I would argue, essential - to ground these discussions. Problems that would likely go beyond the capability of a current PC to solve in a realistic amount of time, in the current NM architecture, would include for instance: -- Learning a new type of linguistic relationship (in the context of link grammar, this would mean e.g. learning a new grammatical link type) -- Learning a new truth value formula for a probabilistic inference rule -- Recognizing objects in a complex, rapidly-changing visual scene (Not that we have written the code to let the system solve these particular problems yet ... but the architecture should allow it...) I don't think we need more than hundreds of PCs to deal with these things, but we need more than a current PC, according to the behavior of our current algorithms. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73544012-c56a06
Re: [agi] Do we need massive computational capabilities?
On Dec 7, 2007 10:21 AM, Bob Mottram [EMAIL PROTECTED] wrote: If I had 100 of the highest specification PCs on my desktop today (and it would be a big desk!) linked via a high speed network this wouldn't help me all that much. Provided that I had the right knowledge I think I could produce a proof of concept type AGI on a single PC today, even if it ran like a tortoise. It's the knowledge which is mainly lacking I think. I agree that at the moment hardware is NOT the bottleneck. This is why, while we've instrumented the Novamente system to be straightforwardly extensible to a distributed implementation, we haven't done much actual distributed processing implementation yet. We have build commercial systems incorporating the NCE in simple distributed architectures, but haven't gone the distributed-AGI direction yet in practice -- because, as you say, it seems likely that the key AGI problems can be worked out on a single machine, and you can then scale up afterwards. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73609156-15fdf3
Re: [agi] None of you seem to be able ...
On Dec 6, 2007 8:06 PM, Ed Porter [EMAIL PROTECTED] wrote: Ben, To the extent it is not proprietary, could you please list some of the types of parameters that have to be tuned, and the types, if any, of Loosemore-type complexity problems you envision in Novamente or have experienced with WebMind, in such tuning and elsewhere? Ed Porter A specific list of parameters would have no meaning without a huge explanation which I don't have time to give... Instead I'll list a few random areas where choices need to be made, that appear localized at first but wind up affecting the whole -- attention allocation is handled by an artificial economy mechanism, which has the same sorts of parameters as any economic system (analogues of interest rates, rent rates, etc.) -- program trees representing internal procedures are normalized via a set of normalization rules, which collectively cast procedures into a certain normal form. There are many ways to do this. -- the pruning of (backward and forward chaining) inference trees uses a statistical bandit problem methodology, which requires a priori probabilities to be ascribed to various inference steps Fortunately though in each of the above three examples there is theory that can guide parameter tning (different theories in the three cases -- dynamic systems theory for the artificial economy; formal computer science and language theory for program tree reduction; and Bayesian stats for the pruning issue) Webmind AI Engine had too many parameters and too much coupling between subsystems. We cast parameter optimization as an AI learning problem but it was a hard one, though we did make headway on it. Novamente Engine has much coupling btw subsystems, but no unnecessary coupling; and many fewer parameters on which system behavior can sensitively depend. Definitely, minimization of the number of needful-of-adjustment parameters is a very key aspect of AGI system design. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73598324-4bf78b
Re: [agi] Do we need massive computational capabilities?
Clearly the brain works VASTLY differently and more efficiently than current computers - are you seriously disputing that? It is very clear that in many respects the brain is much less efficient than current digital computers and software. It is more energy-efficient by and large, as Read Montague has argued ... but OTOH sometimes it is wy less algorithmically efficient For instance, in spite of its generally high energy efficiency, my brain wastes a lot more energy calculating 969695775755/ 8884 than my computer does. And e.g. visual cortex, while energy-efficient, is horribly algorithmically inefficient, involving e.g. masses of highly erroneous motion-sensing neurons whose results are averaged together to give reasonably accurate values.. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73893310-401039
Re: [agi] None of you seem to be able ...
On Dec 5, 2007 6:23 PM, Mike Tintner [EMAIL PROTECTED] wrote: Ben: To publish your ideas in academic journals, you need to ground them in the existing research literature, not in your own personal introspective observations. Big mistake. Think what would have happened if Freud had omitted the 40-odd examples of slips in The Psychopathology of Everyday Life (if I've got the right book!) Obviously, Freud's reliance on introspection and qualitative experience had plusses and minuses. He generated a lot of nonsense as well as some brilliant ideas. But anyway, I was talking about style of exposition, not methodology of doing work. If Freud were a professor today, he would write in a different style in order to get journal publications; though he might still write some books in a more expository style as well. I was pointing out that, due to the style of exposition required in contemporary academic culture, one can easily get a false impression that no one in academia is doing original thinking -- but the truth is that, even if you DO original thinking, you are required in writing your ideas up for publication to give them the appearance of minimal originality via grounding them exorbitantly in the prior literature (even if in fact their conception had nothing, or very little, to do with the prior literature). I'm not saying I like this -- I'm just describing the reality. Also, in the psych literature, grounding an idea in your own personal observations is not acceptable and is not going to get you published -- unless of course you're a clinical psychologist, which I am not. The scientific heavyweights are the people who are heavily grounded. The big difference between Darwin and Wallace is all those examples/research, and not the creative idea. That is an unwarranted overgeneralization. Anyway YOU were the one who was harping on the lack of creativity in AGI. Now you've changed your tune and are harping on the lack of {creativity coupled with a lot of empirical research} Ever consider that this research is going on RIGHT NOW? I don't know why you think it should be instantaneous. A number of us are doing concrete research work aimed at investigating our creative ideas about AGI. Research is hard. It takes time. Darwin's research took time. The Manhattan Project took time. etc. And what I didn't explain in my simple, but I believe important, two-stage theory of creative development is that there's an immense psychological resistance to moving onto the second stage. You have enough psychoanalytical understanding, I think, to realise that the unusual length of your reply to me may possibly be a reflection of that resistance and an inner conflict. What is bizarre to me, in this psychoanalysis of Ben Goertzel that you present, is that you overlook the fact that I am spending most of my time on concrete software projects, not on abstract psychological/philosophical theory Including the Novamente Cognition Engine project which is aimed precisely at taking some of my creative ideas about AGI and realizing them in useful software As it happens, my own taste IS more for theory, math and creative arts than software development -- but, I decided some time ago that the most IMPORTANT thing I could do would be to focus a lot of attention on implementation and detailed design rather than just generating more and more funky ideas. It is always tempting to me to consider my role as being purely that of a thinker, and leave all practical issues to others who like that sort of thing better -- but I consider the creation of AGI *so* important that I've been willing to devote the bulk of my time to activities that run against my personal taste and inclination, for some years now And fortunately I have found some great software engineers as collaborators. P.S. Just recalling a further difference between the original and the creative thinker - the creative one has greater *complexes* of ideas - it usually doesn't take just one idea to produce major creative work, as people often think, but a whole interdependent network of them. That, too, is v. hard. Mike, you can make a lot of valid criticisms against me, but I don't think you can claim I have not originated an interdependent network of creative ideas. I certainly have done so. You may not like or believe my various ideas, but for sure they form an interdependent network. Read The Hidden Pattern for evidence. -- Ben Goertzel - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73146505-9fe3b7
Re: [agi] None of you seem to be able ...
There is no doubt that complexity, in the sense typically used in dynamical-systems-theory, presents a major issue for AGI systems. Any AGI system with real potential is bound to have a lot of parameters with complex interdependencies between them, and tuning these parameters is going to be a major problem. The question is whether one has an adequate theory of one's system to allow one to do this without an intractable amount of trial and error. Loosemore -- if I interpret him correctly -- seems to be suggesting that for powerful AGI systems no such theory can exist, on principle. I doubt very much this is correct. -- Ben G On Dec 6, 2007 9:40 AM, Ed Porter [EMAIL PROTECTED] wrote: Jean-Paul, Although complexity is one of the areas associated with AI where I have less knowledge than many on the list, I was aware of the general distinction you are making. What I was pointing out in my email to Richard Loosemore what that the definitions in his paper Complex Systems, Artificial Intelligence and Theoretical Psychology, for irreducible computability and global-local interconnect themselves are not totally clear about this distinction, and as a result, when Richard says that those two issues are an unavoidable part of AGI design that must be much more deeply understood before AGI can advance, by the more loose definitions which would cover the types of complexity involved in large matrix calculations and the design of a massive supercomputer, of course those issues would arise in AGI design, but its no big deal because we have a long history of dealing with them. But in my email to Richard I said I was assuming he was not using this more loose definitions of these words, because if he were, they would not present the unexpected difficulties of the type he has been predicting. I said I though he was dealing with more the potentially unruly type of complexity, I assume you were talking about. I am aware of that type of complexity being a potential problem, but I have designed my system to hopefully control it. A modern-day well functioning economy is complex (people at the Santa Fe Institute often cite economies as examples of complex systems), but it is often amazingly unchaotic considering how loosely it is organized and how many individual entities it has in it, and how many transitions it is constantly undergoing. Unsually, unless something bangs on it hard (such as having the price of a major commodity all of a sudden triple), it has a fair amount of stability, while constantly creating new winners and losers (which is a productive form of mini-chaos). Of course in the absence of regulation it is naturally prone to boom and bust cycles. So the system would need regulation. Most of my system operates on a message passing system with little concern for synchronization, it does not require low latencies, most of its units, operate under fairly similar code. But hopefully when you get it all working together it will be fairly dynamic, but that dynamism with be under multiple controls. I think we are going to have to get such systems up and running to find you just how hard or easy they will be to control, which I acknowledged in my email to Richard. I think that once we do we will be in a much better position to think about what is needed to control them. I believe such control will be one of the major intellectual challenges to getting AGI to function at a human-level. This issue is not only preventing runaway conditions, it is optimizing the intelligence of the inferencing, which I think will be even more import and diffiducle. (There are all sorts of damping mechanisms and selective biasing mechanism that should be able to prevent many types of chaotic behaviors.) But I am quite confident with multiple teams working on it, these control problems could be largely overcome in several years, with the systems themselves doing most of the learning. Even a little OpenCog AGI on a PC, could be interesting first indication of the extent to which complexity will present control problems. As I said if you had 3G of ram for representation, that should allow about 50 million atoms. Over time you would probably end up with at least hundreds of thousand of complex patterns, and it would be interesting to see how easy it would be to properly control them, and get them to work together as a properly functioning thought economy in what ever small interactive world they developed their self-organizing pattern base. Of course on such a PC based system you would only, on average, be able to do about 10million pattern to pattern activations a second, so you would be talking about a fairly trivial system, but with say 100K patterns, it would be a good first indication of how easy or hard agi systems will be to control. Ed Porter -Original Message- From: Jean-Paul Van Belle [mailto:[EMAIL PROTECTED] Sent: Thursday, December 06,
Re: [agi] None of you seem to be able ...
Conclusion: there is a danger that the complexity that even Ben agrees must be present in AGI systems will have a significant impact on our efforts to build them. But the only response to this danger at the moment is the bare statement made by people like Ben that I do not think that the danger is significant. No reason given, no explicit attack on any component of the argument I have given, only a statement of intuition, even though I have argued that intuition cannot in principle be a trustworthy guide here. But Richard, your argument ALSO depends on intuitions ... I'll try, though, to more concisely frame the reason I think your argument is wrong. I agree that AGI systems contain a lot of complexity in the dynamical- systems-theory sense. And I agree that tuning all the parameters of an AGI system externally is likely to be intractable, due to this complexity. However, part of the key to intelligence is **self-tuning**. I believe that if an AGI system is built the right way, it can effectively tune its own parameters, hence adaptively managing its own complexity. Now you may say there's a problem here: If AGI component A2 is to tune the parameters of AGI component A1, and A1 is complex, then A2 has got to also be complex ... and who's gonna tune its parameters? So the answer has got to be that: To effectively tune the parameters of an AGI component of complexity X, requires an AGI component of complexity a bit less than X. Then one can build a self-tuning AGI system, if one does the job right. Now, I'm not saying that Novamente (for instance) is explicitly built according to this architecture: it doesn't have N components wherein component A_N tunes the parameters of component A_(N+1). But in many ways, throughout the architecture, it relies on this sort of fundamental logic. Obviously it is not the case that every system of complexity X can be parameter-tuned by a system of complexity less than X. The question however is whether an AGI system can be built of such components. I suggest the answer is yes -- and furthermore suggest that this is pretty much the ONLY way to do it... Your intuition is that this is not possible, but you don't have a proof of this... And yes, I realize the above argument of mine is conceptual only -- I haven't given a formal definition of complexity. There are many, but that would lead into a mess of math that I don't have time to deal with right now, in the context of answering an email... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73243865-194e0e
Re: Last word for the time being [WAS Re: [agi] None of you seem to be able ...]
Show me ONE other example of the reverse engineering of a system in which the low level mechanisms show as many complexity-generating characteristics as are found in the case of intelligent systems, and I will gladly learn from the experience of the team that did the job. I do not believe you can name a single one. Well, I am not trying to reverse engineer the brain. Any more than the Wright Brothers were trying to reverse engineer a bird -- though I do imagine the latter will eventually be possible. You know, I sympathize with you in a way. You are trying to build an AGI system using a methodology that you are completely committed to. And here am I coming along like Bertrand Russell writing his letter to Frege, just as poor Frege was about to publish his Grundgesetze der Arithmetik, pointing out that everything in the new book was undermined by a paradox. How else can you respond except by denying the idea as vigorously as possible? It's a deeply flawed analogy. Russell's paradox is a piece of math and once Frege was confronted with it he got it. The discussion between the two of them did not devolve into long, rambling dialogues about the meanings of terms and the uncertainties of various intuitions. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=73249230-63bddf
Re: [agi] None of you seem to be able ...
Tintner wrote: Your paper represents almost a literal application of the idea that creativity is ingenious/lateral. Hey it's no trick to be just ingenious/lateral or fantastic. Ah ... before creativity was what was lacking. But now you're shifting arguments and it's something else that is lacking ;-) You clearly like producing new psychological ideas - from a skimming of your work, you've produced several. However, I didn't come across a single one that was grounded or where any attempt was made to ground them in direct, fresh observation (as opposed to occasionally referring to an existing scientific paper). That is a very strange statement. In fact nearly all my psychological ideas are grounded in direct, fresh **introspective** observation --- but they're not written up that way because that's not the convention in modern academia. To publish your ideas in academic journals, you need to ground them in the existing research literature, not in your own personal introspective observations. It is true that few of my psychological hypotheses are grounded in my own novel lab experiments, though. I did a little psych lab work in the late 90's, in the domain of perceptual illusions -- but the truth is that psych and neuroscience are not currently sophisticated enough to allow empirical investigation of really interesting questions about the nature of cognition, self, etc. Wait a couple decades, I guess. In terms of creative psychology, that is consistent with your resistance to producing prototypes - and grounding your invention/innovation. Well, I don't have any psychological resistance to producing working software, obviously. Most of my practical software work has been proprietary for customers; but, check out MOSES and OpenBiomind on Google Code -- two open-source projects that have emerged from my Novamente LLC and Biomind LLC work ... It just happens that AGI does not lend itself to prototyping, for reasons I've already tried and failed to explain to you We're gonna launch trainable, adaptive virtual animals in Second Life sometime in 2008 But I won't consider them real prototypes of Novamente AGI, even though in fact they will use several aspects of the Novamente Cognition Engine software. They won't embody the key emergent structures/dynamics that I believe need to be there to have human-level cognition -- and there is no simple prototype system that will do so. You celebrate Jeff Hawkins' prototype systems, but have you tried them? He's built (or, rather Dileep George has built) an image classification engine, not much different in performance from many others out there. It's nice work but it's not really an AGI prototype, it's an image classifiers. He may be sort-of labeling it a prototype of his AGI approach -- but really, it doesn't prove anything dramatic about his AGI approach. No one who inspected his code and ran it would think that it did provide such proof. There are at least two stages of creative psychological development - which you won't find in any literature. The first I'd call simply original thinking, the second is truly creative thinking. The first stage is when people realise they too can have new ideas and get hooked on the excitement of producing them. Only much later comes the second stage, when thinkers realise that truly creative ideas have to be grounded. Arguably, the great majority of people who may officially be labelled as creatives, never get beyond the first stage - you can make a living doing just that. But the most beautiful and valuable ideas come from being repeatedly refined against the evidence. People resist this stage because it does indeed mean a lot of extra work , but it's worth it. (And it also means developing that inner faculty which calls for actual evidence). OK, now you're making a very different critique than what you started with though. Before you were claiming there are no creative ideas in AGI. Now, when confronted with creative ideas, you're complaining that they're not grounded via experimental validation. Well, yeah... And the problem is that if one's creative ideas pertain to the dynamics of large-scale, complex software systems, then it takes either a lot of time or a lot of money to achieve this validation that you mention. It is not the case that I (and other AGI researchers) are somehow psychologically undesirous of seeing our creative ideas explored via experiment. It is, rather, the case that doing the relevant experiments requires a LOT OF WORK, and we are few in number with relatively scant resources. What I am working toward, with Novamente and soon with OpenCog as well, is precisely the empirical exploration of the various creative ideas of myself, others whose work has been built on in the Novamente design, and my colleagues... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to:
Re: [agi] None of you seem to be able ...
More generally, I don't perceive any readiness to recognize that the brain has the answers to all the many unsolved problems of AGI - Obviously the brain contains answers to many of the unsolved problems of AGI (not all -- e.g. not the problem of how to create a stable goal system under recursive self-improvement). However, current neuroscience does NOT contain these answers. And neither you nor anyone else has ever made a cogent argument that emulating the brain is the ONLY route to creating powerful AGI. The closest thing to such an argument that I've seen was given by Eric Baum in his book What Is Thought?, and I note that Eric has backed away somewhat from that position lately. I think it should be obvious that AGI isn't going to happen - and none of the unsolved problems are going to be solved - without major creative leaps. Just look even at the ipod iphone - major new technology never happens without such leaps. The above sentence is rather hilarious to me. If the Ipod and Iphone are your measure for creative leaps then there have been loads and loads of major creative leaps in AGI and narrow-AI research. Anyway it seems to me that you're not just looking for creative leaps, you're looking for creative leaps that match your personal intuition. Perhaps the real problem is that your personal intuition about intelligence is largely off-base ;-) As an example of a creative leap (that is speculative and may be wrong, but is certainly creative), check out my hypothesis of emergent social-psychological intelligence as related to mirror neurons and octonion algebras: http://www.goertzel.org/dynapsyc/2007/mirrorself.pdf I happen to think the real subtlety of intelligence happens on the emergent level, and not on the level of the particulars of the system that gives rise to the emergent phenomena. That paper conjectures some example phenomena that I believe occur on the emergent level of intelligent systems. Loosemore agrees with me on the importance of emergence, but he feels there is a fundamental irreducibility that makes it pragmatically impossible to figure out via science, math and intuition which concrete structures/dynamics will give rise to the right emergent structures, without doing a massive body of simulation experiments. I think he overstates the degree of irreducibility. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=72114408-ae9503
Re: [agi] None of you seem to be able ...
On Dec 4, 2007 8:38 PM, Richard Loosemore [EMAIL PROTECTED] wrote: Benjamin Goertzel wrote: [snip] And neither you nor anyone else has ever made a cogent argument that emulating the brain is the ONLY route to creating powerful AGI. The closest thing to such an argument that I've seen was given by Eric Baum in his book What Is Thought?, and I note that Eric has backed away somewhat from that position lately. This is a pretty outrageous statement to make, given that you know full well that I have done exactly that. You may not agree with the argument, but that is not the same as asserting that the argument does not exist. Unless you were meaning emulating the brain in the sense of emulating it ONLY at the low level of neural wiring, which I do not advocate. I don't find your nor Eric's nor anyone else's argument that brain-emulation is the golden path very strongly convincing... However, I found Eric's argument by reference to the compressed nature of the genome, more convincing than your argument via the hypothesis of irreducible emergent complexity... Sorry if my choice of words was not adequately politic. I find your argument interesting, but it's certainly just as speculative as the various AGI theories you dismiss It basically rests on a big assumption, which is that the complexity of human intelligence is analytically irreducible within pragmatic computational constraints. In this sense it's less an argument than a conjectural assertion, albeit an admirably bold one. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=72126612-7f96e4
Re: Hacker intelligence level [WAS Re: [agi] Funding AGI research]
Thus: building a NL parser, no matter how good it is, is of no use whatsoever unless it can be shown to emerge from (or at least fit with) a learning mechanism that allows the system itself to generate its own understanding (or, at least, acquisition) of grammar IN THE CONTEXT OF A MECHANISM THAT ALSO ACCOMPLISHES REAL UNDERSTANDING. When that larger issue is dealt with, a NL parser will arise naturally, and any previous work on non-developmental, hand-built parsers will be completely discarded. You were trumpeting the importance of work that I know will be thrown away later, and in the mean time will be of no help in resolving the important issues. Richard, you discount the possibility that said NL parser will play a key role in the adaptive emergence of a system that can generate its own linguistic understanding. I.e., you discount the possibility that, with the right learning mechanism and instructional environment, hand-coded rules may serve as part of the initial seed for a learning process that will eventually generate knowledge obsoleting these initial hand-coded rules. It's fine that you discount this possibility -- I just want to point out that in doing so, you are making a bold and unsupported theoretical hypothesis, rather than stating an obvious or demonstrated fact. Vaguely similarly, the grammar of child language is largely thrown away in adulthood, yet it was useful as scaffolding in leading to the emergence of adult language. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=72129171-2bf67a
Re: [agi] None of you seem to be able ...
Richard, Well, I'm really sorry to have offended you so much, but you seem to be a mighty easy guy to offend! I know I can be pretty offensive at times; but this time, I wasn't even trying ;-) The argument I presented was not a conjectural assertion, it made the following coherent case: 1) There is a high prima facie *risk* that intelligence involves a significant amount of irreducibility (some of the most crucial characteristics of a complete intelligence would, in any other system, cause the behavior to show a global-local disconnect), and The above statement contains two fuzzy terms -- high and significant ... You have provided no evidence for any particular quantification of these terms... your evidence is qualitative/intuitive, so far as I can tell... Your quantification of these terms seems to me a conjectural assertion unsupported by evidence. 2) Because of the unique and unusual nature of complexity there is only a vanishingly small chance that we will be able to find a way to assess the exact degree of risk involved, and 3) (A corollary of (2)) If the problem were real, but we were to ignore this risk and simply continue with an engineering approach (pretending that complexity is insignificant), The engineering approach does not pretend that complexity is insignificant. It just denies that the complexity of intelligent systems leads to the sort of irreducibility you suggest it does. Some complex systems can be reverse-engineered in their general principles even if not in detail. And that is all one would need to do in order to create a brain emulation (not that this is what I'm trying to do) --- assuming one's goal was not to exactly emulate some specific human brain based on observing the behaviors it generates, but merely to emulate the brainlike character of the system... then the *only* evidence we would ever get that irreducibility was preventing us from building a complete intelligence would be the fact that we would simply run around in circles all the time, wondering why, when we put large systems together, they didn't quite make it, and No. Experimenting with AI systems could lead to evidence that would support the irreducibility hypothesis more directly than that. I doubt they will but it's possible. For instance, we might discover that creating more and more intelligent systems inevitably presents more and more complex parameter-tuning problems, so that parameter-tuning appears to be the bottleneck. This would suggest that some kind of highly expensive evolutionary or ensemble approach as you're suggesting might be necessary. 4) Therefore we need to adopt a Precautionary Principle and treat the problem as if irreducibility really is significant. Whether you like it or not - whether you've got too much invested in the contrary point of view to admit it, or not - this is a perfectly valid and coherent argument, and your attempt to try to push it into some lesser realm of a conjectural assertion is profoundly insulting. The form of the argument is coherent and valid; but the premises involve fuzzy quantifiers whose values you are apparently setting by intuition, and whose specific values sensitively impact the truth value of the conclusion. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=72135696-ff196d
[agi] Re: A global approach to AI in virtual, artificial and real worlds
What makes anyone think OpenCog will be different? Is it more understandable? Will there be long-term aficionados who write books on how to build systems in OpenCog? Will the developers have experience, or just adolescent enthusiasm? I'm watching the experiment to find out. Well, OpenCog has more than one possible development avenue associated with it. On the one hand, I have some quite specific AGI design ideas which I intend to publish next year (major aspects of the Novamente AGI design), which are suited to be implemented within OpenCog. As I believe these ideas are capable to lead to the development of AGI at the human-level and beyond (though there are many moderate-sized research problems that must be solved along the way, and yes I realize the possibility that one of these blows up and becomes a show-stopper, but I'm betting that won't happen and I've certainly thought about it a lot...) ... thus I believe OpenCog has big potential in this regard, if folks choose to develop it in that way. On the other hand OpenCog may also be quite valuable as a platform for the development of other folks' AGI ideas, potentially ones quite different from my own. I don't know what will develop and neither do any of us, I would suppose... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=72154694-0749bf
Re: Hacker intelligence level [WAS Re: [agi] Funding AGI research]
OK, understood... On Dec 4, 2007 9:32 PM, Richard Loosemore [EMAIL PROTECTED] wrote: Benjamin Goertzel wrote: Thus: building a NL parser, no matter how good it is, is of no use whatsoever unless it can be shown to emerge from (or at least fit with) a learning mechanism that allows the system itself to generate its own understanding (or, at least, acquisition) of grammar IN THE CONTEXT OF A MECHANISM THAT ALSO ACCOMPLISHES REAL UNDERSTANDING. When that larger issue is dealt with, a NL parser will arise naturally, and any previous work on non-developmental, hand-built parsers will be completely discarded. You were trumpeting the importance of work that I know will be thrown away later, and in the mean time will be of no help in resolving the important issues. Richard, you discount the possibility that said NL parser will play a key role in the adaptive emergence of a system that can generate its own linguistic understanding. I.e., you discount the possibility that, with the right learning mechanism and instructional environment, hand-coded rules may serve as part of the initial seed for a learning process that will eventually generate knowledge obsoleting these initial hand-coded rules. It's fine that you discount this possibility -- I just want to point out that in doing so, you are making a bold and unsupported theoretical hypothesis, rather than stating an obvious or demonstrated fact. Vaguely similarly, the grammar of child language is largely thrown away in adulthood, yet it was useful as scaffolding in leading to the emergence of adult language. The problem is that this discussion has drifted away from the original context in which I made the remarks. I do *not* discount the possibility that an ordinary NL parser may play a role in the future. What I was attacking was the idea that a NL parser that does a wonderful job today (but which is built on a formalism that ignores all the issues involved in getting an adaptive language-understanding system working) is IPSO FACTO going to be a valuable step in the direction of a full adaptive system. It was the linkage that I dismissed. It was the idea that BECAUSE the NL parser did such a great job, therefore it has a very high probability of being a great step on the road to a full adaptive (etc) language understanding system. If the NL parser completely ignores those larger issues I am justified in saying that it is a complete crap shoot whether or not this particular parser is going to be of use in future, more complete theories of language. But that is not the same thing as making a blanket dismissal of all parsers, saying they cannot be of any use as (as you point out) seed material in the design of a complete system. I was objecting to Ed's pushing this particular NL parser in my face and insisting that I should respect it as a substantial step towards full AGI . and my objection was that I find models like that all show and no deep substance precisely because they ignore the larger issues and go for the short-term gratification of a parser that works really well. So I was not taking the position you thought I was. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=72155184-923590
Re: [agi] Funding AGI research
On Nov 30, 2007 7:57 AM, Mike Tintner [EMAIL PROTECTED] wrote: Ben: It seems to take tots a damn lot of trials to learn basic skills Sure. My point is partly that human learning must be pretty quantifiable in terms of number of times a given action is practised, Definitely NOT ... it's very hard to quantify when a child is practicing crawling versus just rehearsing the component arm/leg movements, wiggling around, etc. I can imagine that quantifying this sort of thing in a really meaningful way must be fairly difficult... I wonder whether anyone's counting. I agree it's a worthwhile effort, though. I don't think anyone has counted this sort of thing because it would require constant surveillance of the child. The data being gathered in Deb Roy's Human Speechome project should actually be useful for this -- he's got a video camera on his young child nearly 24 hours a day... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=70743699-cc240c
Re: FW: [agi] AGI DARPA-style
Yeah, I've been following that for a while. There are some very smart people involved, and it's quite possible they'll make a useful software tool, but I don't feel they have a really viable unified cognitive architecture. It's the sort of architecture where different components are written in different programming languages based on unrelated ideas and are hooked together in an overall architecture, interacting with each other as black boxes. No emergent intelligence via inter-component interactions is likely to arise in this approach And of course the lack of embodiment makes any solution to the symbol-grounding problem unlikely to emerge... -- Ben On Nov 30, 2007 12:59 PM, Ed Porter [EMAIL PROTECTED] wrote: Also checkout http://caloproject.sri.com/publications/ for a list of CALO related publications Ed Porter -Original Message- From: Ed Porter [mailto:[EMAIL PROTECTED] Sent: Friday, November 30, 2007 12:58 PM To: 'agi@v2.listbox.com' Subject: RE: [agi] AGI DARPA-style Checkout AGI DARPA-style: Software That Learns from Users-- A massive AI project called CALO could revolutionize machine learning at http://www.technologyreview.com/Infotech/19782/?a=f Ed Porter - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=70909538-e0531f
Re: [agi] Funding AGI research
[What related principles govern the Novamente's figure's trial and error learning of how to pick up a ball?] Pure trial and error learning is really slow though... we are now relying on a combination of -- reinforcement from a teacher -- imitation of others' behavior -- trial and error -- active correction of wrong behavior by a teacher ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=70641251-aaef7a
Re: [agi] Funding AGI research
On Nov 29, 2007 11:35 PM, Mike Tintner [EMAIL PROTECTED] wrote: Presumably, human learning isn't that slow though - if you simply count the number of attempts made before any given movement is mastered at a basic level (.e.g crawling/ walking/ grasping/ tennis forehand etc)? My guess would be that, for all the frustrations involved, we need relatively few attempts. Maybe in the hundreds or thousands at most? It seems to take tots a damn lot of trials to learn basic skills, and we have plenty of inductive bias in our evolutionary wiring... But then it seems increasingly clear that we use maps/ graphics/ schemas to guide our movements - have you read the latest Blakeslee book on body maps? So does Novamente, it uses an internal simulation-world (among other mechanism)... but that doesn't magically make learning rapid, though it makes it more tractable... ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=70644788-023e28
Re: Re[12]: [agi] Funding AGI research
On Nov 30, 2007 12:03 AM, Dennis Gorelik [EMAIL PROTECTED] wrote: Benjamin, That proves my point [that AGI project can be successfully split into smaller narrow AI subprojects], right? Yes, but it's a largely irrelevant point. Because building a narrow-AI system in an AGI-compatible way is HARDER than building that same narrow-AI component in a non-AGI-compatible way. Even if this is the case (which is not) that would simply mean several development steps: 1) Develop narrow AI with non-reusable AI component and get rewarded for that (because it would be useful system by itself). Obviously, most researchers who have developed useful narrow-AI components have not gotten rich from it. The nature of our economy and society is such that most scientific and technical innovators are not dramatically financially rewarded. 2) Refactor non-reusable AI component into reusable AI component and get rewarded for that (because it would reusable component for sale). 3) Apply reusable AI component in AGI and get rewarded for that. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=70648456-e5f42e
Re: Re[12]: [agi] Funding AGI research
So far only researchers/developers who picked narrow-AI approach accomplished something useful for AGI. E.g.: Google, computer languages, network protocols, databases. These are tools that are useful for AGI RD but so are computer monitors, silicon chips, and desk chairs. Being a useful tool for AGI RD does not make something constitute AGI RD. I do note that I myself have done (and am doing) plenty of narrow AI work in parallel with AGI work. So I'm not arguing against narrow AI nor stating that narrow AI is irrelevant to AGI. But your view of the interrelationship seems extremely oversimplified to me. If it were as simple as you're saying, I imagine we'd have human-level AGI already, as we have loads of decent narrow-AI's for various tasks. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=70647705-610230
Re: Re[10]: [agi] Funding AGI research
EDI must admit, I have never heard cortical column described as containing 10^5 neurons. The figure I have commonly seen is 10^2 neurons for a cortical column, although my understanding is that the actual number could be either less or more. I guess the 10^5 figure would relate to so-called hypercolumns. The term cortical column is vague http://en.wikipedia.org/wiki/Cortical_column There are minicolumns (around 100 neurons each) and hypercolumns (around 100 minicolumns each). Both are called columns.. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=69432075-16cf74
Re: Re[4]: [agi] Funding AGI research
Still, this is the most resource-intensive part of the Novamente system (the part that's most likely to require supercomputers to achieve human-level AI). Why is it the most resource intensive, is it the evolutionary computational cost? Is this where MOSES is used? Correct, this is one place that MOSES is used... - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=69229200-f1b2ef
Re: Re[10]: [agi] Funding AGI research
Nearly any AGI component can be used within a narrow AI, That proves my point [that AGI project can be successfully split into smaller narrow AI subprojects], right? Yes, but it's a largely irrelevant point. Because building a narrow-AI system in an AGI-compatible way is HARDER than building that same narrow-AI component in a non-AGI-compatible way. So, given the pressures of commerce and academia, people who are motivated to make narrow-AI for its own sake, will almost never create narrow-AI components that are useful for AGI. And, anyone who creates narrow-AI components with an AGI outlook, will have a large disadvantage in the competition to create optimal narrow-AI systems given limited time and financial resources. Still, AGI-oriented researcher can pick appropriate narrow AI projects in a such way that: 1) Narrow AI project will be considerably less complex than full AGI project. 2) Narrow AI project will be useful by itself. 3) Narrow AI project will be an important building block for the full AGI project. Would you agree that splitting very complex and big project into meaningful parts considerably improves chances of success? Yes, sure ... but demanding that these meaningful parts -- be economically viable and/or -- beat competing, somewhat-similar components in competitions dramatically DECREASES chances of success ... That is the problem. An AGI may be built out of narrow-AI components, but these narrow-AI components must be architected for AGI-integration, which is a lot of extra work; and considered as standalone narrow-AI components, they may not outperform other similar narrow-Ai components NOT intended for AGI-integration... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=69277648-780726
Re: Re[8]: [agi] Funding AGI research
My claim is that it's possible [and necessary] to split massive amount of work that has to be done for AGI into smaller narrow AI chunks in such a way that every narrow AI chunk has it's own business meaning and can pay for itself. You have not addressed my claim, which has massive evidence in the history of AI research to date, that narrow AI chunks with AGI compatibility are generally much harder to build than narrow AI chunks intended purely for standalone performance, and hence will very rarely be the best economic choice if one's goal is to make a narrow-AI chunk serving some practical application within (the usual) tight time and cost constraints. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=69291210-f650cd
Re: Re[4]: [agi] Funding AGI research
Well, there is a discipline of computer science devoted to automatic programming, i.e. synthesizing software based on specifications of desired functionality. State of the art is: -- Just barely, researchers have recently gotten automated program learning to synthesize an nlogn sorting algorithm based on the goal of sorting a large set of lists as rapidly as possible... -- OTOH, automatic synthesis of logic circuits automatically carrying out various tasks is now a fairly refined science, see e.g. Koza's GP III book All in all we are nowhere near having AI software that can automatically synthesize large, complex software programs. Automated program learning is part of the Novamente system but the architecture is designed so that only small programs need to be learned, carrying out particular internal or external tasks/functions. Still, this is the most resource-intensive part of the Novamente system (the part that's most likely to require supercomputers to achieve human-level AI). -- Ben On Nov 26, 2007 7:14 AM, Mike Tintner [EMAIL PROTECTED] wrote: John: I kind of like the idea of building software that then builds AGI. What are the best current examples of (to any extent) self-building software ? - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=68550543-6ccd8f
Re: Re[6]: [agi] Funding AGI research
Linas: I find it telling that no one is saying I've got the code, I just need to scale it up 1000-fold to make it impressive ... Yes, that's an accurate comment. Novamente will hopefully reach that point in a few years. For now, we will need (and use) a lotta machines for commercial product deployment purposes. But for RD purposes, it's all about solving a large number of moderate-sized computer science and AI research problems, that are connected together via the overall NM AGI design. Once these problems are all worked through and we have a completed Novamente codebase then we will be far better able to evaluate what our hardware requirements actually are. I am pretty sure they will be large. But right now, having masses of hardware wouldn't accelerate our progress all that much. What is useful to us is money to pay the right brains to solve the long list of apparently-not-that-huge technical problems between here and a completed Novamente system. And of course there is always a nonzero risk that one of these apparently-not-that-huge technical problems will turn out to be huge; but, a lot of thinking has gone on over a number of years in a serious attempt to avoid this... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=68394009-e1d34e
Re: Re[6]: [agi] Funding AGI research
Cassimatis's system is an interesting research system ... it doesn't yet have lotsa demonstrated practical functionality, if that's what you mean by work... He wants to take a bunch of disparately-functioning agents, and hook them together into a common framework using a common logical interlingua I think this approach is unlikely to lead to the various agents involved quelling, rather than exacerbating, each others' intrinsic combinatorial explosions... I think it is unlikely to lead to sufficiently coherent system-wide emergent dynamics to give rise to an effective phenomenal self ... But given the primitive state of AGI theory at the moment, I can't *prove* that these complaints are correct, of course... -- Ben G On Nov 25, 2007 7:22 PM, Edward Porter [EMAIL PROTECTED] wrote: A few days ago there was some discussion on this list about the potential usefulness of narrow AI to AGI. Nick Cassimatis, who is speaking at AGI 2008, has something he calls Polyscheme which is described partially at the following AGIRI link: http://www.agiri.org/workshop/Cassimatis.ppt It appears to use what are arguably narrow AI modules in a coordinated manner to achieve AGI. Is this a correct interpretation? Does it work? And, if so, how? I can imagine how multiple narrow AI's could be used to create a more general AGI if there were some AGI glue to represent and learn the relationships between the different AGI modalities. Cassimatis mentions tying these different modalities together using relations involving times, space, events, identity, causality and belief. (But I don't remember much description of how it does it.) Arguably these are enough dimensions to create generalized representations, provided there is some generalized means for representing all the important states and representations in each of the Narrow AI modalities and the relationships between them in each of these dimensions and compositions and generalizations formed from such relationships. Is that what Cassimatis is talking about? Ed Porter This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=68488723-9c2917
Re: Re[4]: [agi] Funding AGI research
Are you asking for success stories regarding research funding in any domain, or regarding research funding in AGI? Any domain, please. OK, so your suggestion is that research funding, in itself, is worthless in any domain? I don't really have time to pursue this kind of silly argument, sorry... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=67380099-81533c
Re: Re[6]: [agi] Funding AGI research
No. My point is that massive funding without having a prototype prior to funding is worthless most of the times. If prototype cannot be created at reasonably low cost then fully working product most likely cannot be created even with massive funding. Well, this seems to dissolve into a set of vagaries... How much funding is massive varies from domain to domain. E.g. it's hard to do anything in nanotech without really expensive machinery. For AGI, $10M is a lot of money, because the main cost is staff salaries, plus commodity hardware. For nanotech, $10M isn't all that much, since specialized hardware is needed to do many kinds of serious work. And, what counts as a prototype often depends on one's theoretical framework. Do you consider there to have been a prototype for the first atom bomb? I don't think there was, but there were preliminary experiments that, given the context of the framework of theoretical physics, made the workability of the atom bomb seem plausible. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=67466565-8e64f2
Re: Re[6]: [agi] Funding AGI research
On Nov 20, 2007 11:22 PM, Dennis Gorelik [EMAIL PROTECTED] wrote: Jiri, AGI is IMO possible now but requires very different approach than narrow AI. AGI requires properly tune some existing narrow AI technologies, combine them together and may be add couple of more. That's massive amount of work, but most AGI research and development can be shared with narrow AI research and development. Unfortunately, I don't think this is quite true... There is plenty overlap btw AGI and narrow AI but not as much as you suggest... Also: Narrow AI technologies are not meant to be combined together, so to build AGI out of narrow-AI components, you need to create narrow-AI components that are **specifically designed** for integration into an AGI system... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=67471374-384e50
Re: Re[8]: [agi] Funding AGI research
Could you describe a piece of technology that simultaneously: - Is required for AGI. - Cannot be required part of any useful narrow AI. The key to your statement is the word required Nearly any AGI component can be used within a narrow AI, but, the problem is, it's usually a bunch easier to make narrow AI's using components that don't have any AGI value... Another way to go -- use existing narrow AIs as prototypes when building AGI. I don't really accept any narrow-AI as a prototype for an AGI. This is an example of what I meant when I said that what counts as a prototype is theory-dependent, I suppose... I think there is loads of evidence that narrow-AI prowess does not imply AGI prowess, so that a narrow-AI can't be considered a prototype for an AGI.. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=67477308-6a7310
Re: Re[2]: [agi] Funding AGI research
On Nov 18, 2007 12:50 AM, Dennis Gorelik [EMAIL PROTECTED] wrote: Benjamin, Do you have any success stories of such research funding in the last 20 years? Something that resulted in useful accomplishments. Are you asking for success stories regarding research funding in any domain, or regarding research funding in AGI? Obviously, there are not yet any real success stories regarding research funding in AGI. But this is because -- Due to advances in computing hardware and cognitive science, AGI is just now (meaning, in the last 3-8 years, say), for the first time, at a stage where serious advances can be made -- The small amount of $$ put into AGI research has almost entirely focused on an even smaller set of conceptual approaches (GOFAI), which are deeply problematic for reasons already extensively discussed on this list There were no success stories regarding manned spaceflight before Apollo ... there were no success stories for genome-sequencing before it was first done, etc. You seem to have gone from -- advocating funding only for development rather than research to -- advocating funding only for research in mature fields where there have already been dramatic successes I don't think either of these is an adequate funding strategy. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66341486-50b99a
Re: [agi] Funding AGI research
Novamente as a whole is definitely a research project, albeit one with a very well fleshed out research plan. I have a strong hypothesis about how the project will come out, and arguments in favor of this hypothesis; but I don't have the level of confidence I'd have in, say, the stability of a bridge built according to known principles of mechanical engineering; or the functionality of a word-processing program built according to good specs. On the other hand our virtual animals product development is mostly (difficult) software development, with a few bits of (strictly delimited) research contained therein The outcome there is way more determinate. ben On Nov 18, 2007 1:15 PM, Joshua Fox [EMAIL PROTECTED] wrote: What you are advocating is to fund Development but not Research. Ben, I favor funding for both R and D. Would you put the Novamente project in the R or the D phase? If a prototype is a good way to distinguish the two, is there a prototype for Novamente? And if it is still in the research phase, is it reasonable to give time estimates for a project which is still in that phase? Joshua - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66343974-5ff050
Re: [agi] Funding AGI research
Proactively minimizing risk in as many areas as possible make a venture much more salable, but most AI ventures tend to be very apparently risky at many levels that have no relation to the AI research per se and the inability of these ventures to minimize all that unnecessary risk is a giant mark against them. Cheers, J. Andrew Rogers I've often heard it said that VC's consider three kinds of risk -- people -- market -- technology and if you have risk in more than one of those areas, you won't get funded. Of course like all such saws this is a big simplification but it has some relevance to this context anyway. AGI is always going to be viewed as a major technology risk, unless one comes into the fundraising process with an extremely strong prototype (and maybe even then). Mitigaging the people-risk requires getting experienced businesspeople on board, which is generically difficult for an AGI company because of the bad reputation AI has. Mitigating the market risk means finding a market niche where incremental work toward AGI is of dramatically more economic value than narrow-AI technology. I think this is really the hard part. (Or, as an alternative, it involves finding gullible investors and making them believe that the incremental work toward AGI will be of dramatically more economic value than narrow-AI technology -- but most good AGI researchers don't have taste for this particular flavor of dishonesty...) In most practical app areas, you can make something that can be spun to customers as almost-as-good-as- a-fractional-AGI, via using clever narrow-AI techniques. The obvious example is Google which isn't nearly as good as a good NLP search engine will be, but is almost-as-useful as a partially-mature NLP search engine, and was a lot easier/cheaper to prototype and initially roll out. As I've said before, I am bullish on virtual worlds and gaming as an area where early-stage AGI tech can have dramatically more economic value than cleverly crafted narrow-AI. Humanoid robotics is clearly another such area, but a trickier area to get started in right now. But I'm not saying these are the only examples. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66346729-733843
Re: [agi] Funding AGI research
I have not heard a *creative* new idea here that directly addresses and shows the power to solve even in part the problem of creating general intelligence. To be quite frank, the most creative and original ideas inside the Novamente design are quite technical. I suspect you don't have the background to understand them. However, I can see there's something else underlying your remarks: a sort of mystification of the power of intelligence. I think you'll be shocked when the workings of the brain are finally unveiled and what is revealed is that there are a lot of cleverly evolved modules carrying out various particular functions, wired together in an architecture designed to enable their synergy ... and some of them wired to observe and improve one another in virtuous cycles, using their particular recognition/learning algorithms ... but no magic trick, no super-secret algorithm of thought ... In short, I think you'll look at the brain and say hhmmpph, that's not very creative ;-p I think you are looking for an essence of intelligence that is just not there. The essence is in the emergent phenomena that come about when a sufficiently rich set of components is wired together in a way allowing them to inter-adapt and inter-improve in the context of controlling an agent that needs to achieve complex goals in a complex environment. There is no magic trick of thought at the center of it all, that is just waiting for some Einstein of Cognition to unveil it. But you can keep waiting for this magic trick to be revealed, if you like, while some of the rest of us work on actually creating AGI on rational bases ... and more or less gracefully putting up with your continual complaints that because we're not done yet, our approaches must obviously be worthless ;-) -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66437203-2462b5
Re: [agi] Multi-agent learning
On Nov 18, 2007 6:45 PM, Lukasz Stafiniak [EMAIL PROTECTED] wrote: Ben, Have you already considered what form of multi-agent epistemic logic (or whatever extension to PLN) Novamente will use to merge knowledge from different avatars? Well, standard PLN handles this in principle via ContextLinks which allow contextual (e.g. perspectival) specialization of knowledge. However, **control** of this kind of process is the tricky thing, really.. melding knowledge from different minds is computationally costly and one has to decide when one wants to do it... Related, do you consider some form of privacy policy, or do you put the responsibility for not leaking secrets on avatars' owners? For our initial virtual animals: There is a collective memory among all the AI animals, and also an individual memory for each animal, and there's a policy for deciding what goes in which... In later products we will likely allow more flexibility, and let users control how much they want their agents to share and/or take from the collective agent unconscious. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66449659-1f8a96
Re: [agi] Funding AGI research
Hi, The majority of VC's do, as you say, want a technology that is sewn up, from the point of view of technical feasibility. But this is not always true. There is always a gray area at the fringe of feasibility where the last set of questions has not been *fully* answered before money is thrown at it. I believe this happened in a number of projects during the dot-com insanity. A lot of things were possible in that period that aren't possible during normal business conditions... If I am right in this last idea, VCs have a stark choice: if they want AGI, they have to relax their insistence on a project that does not have that last research step. If they insist on something stronger, they can kiss goodbye to ever getting an AGI. Well, VC's don't give a crap about AGI, at least not in their capacity as VC's. They just want to make $$ in a certain way, according to a certain risk profile... So it is only via an unusual combination of factors that a VC is going to invest in an AGI project. And of course, if this unusual combination occurs ONCE, and yields successful results ... then every VC will want to jump on board and invest in AGI as quickly as possible and at a generous valuation ;-) -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66454614-75c7de
Re: [agi] Funding AGI research
There are a lot of worthwhile points in your post, and a number of things I don't fully agree with, but I don't have time to argue them all right now... Instead I'll just pick two points: 1) The Babbages and Leibnizes of a given historical period are often visible only in HINDSIGHT. You can't say that there are no Babbages or Leibnizes of AGI around right now ... there could be some on this very list, unrecognized by you, but who will be recognized by all a few decades from now... 2) I don't think it's true that Babbage's or Leibniz's machines were specced out so much better than, say, Novamente. Relative to the technology of their time, plenty of details were left unspecified -- it just seems obvious to us now, in hindsight, how to fill in those details. It wasn't obvious to all their contemporaries. And while, in hindsight, the workability of their machines seems obvious to us, to their contemporaries it must have seemed like the workability of their machines required a huge leap of intuition. They had no rigorous mathematical proof of the workability of their machines, nor did they have working prototypes. They had conceptual arguments that pushed the boundaries of the science of their times, and seemed like nonsense to many of their contemporaries. 3) I don't agree that AGI is primarily a computer science problem, any more than, say, building a car is primarily a metalworking problem. AGI requires computer science problems to be solved as part of its solution; but IMO the essence of AGI-creation is not computer science. This seems to be a genuine difference of scientific intuition btw the two of us. Plenty of others whom I respect appear to share the same opinion as you. -- Ben G On Nov 18, 2007 11:04 PM, J. Andrew Rogers [EMAIL PROTECTED] wrote: On Nov 18, 2007, at 7:06 PM, Benjamin Goertzel wrote: Navigating complex social and business situations requires a quite different set of capabilities than creating AGI. Potentially they could be combined in the same person, but one certainly can't assume that would be the case. I completely agree. But if we are to assume that AGI requires some respectable amount of funding, as seems to be posited by many people, then it seems that it will require a person with broader skills than the stereotypical computer science nerd. In that case, maybe AGI is not accessible to someone who is unwilling or unable to be anything but a computer science nerd. As if the pool of viable AGI researchers was not small enough already. And, I don't think it's fair to say that if you're smart enough to solve AGI, you should be able to quickly make a pile of money doing some kind of more marketable technical-computer-science, and fund the AGI yourself. This assumes a lot of things, for instance that AGI is the same sort of problem as technical-computer-science problems, so that if someone can do AGI better than others, they must be able to do technical- computer-science better than others too. But I actually don't think this is true; I think that AGI demands a different sort of thinking. I'm not so sure about this. All hard problems seem to receive similar sentiments until they are actually solved. I do think that AGI is a relatively hard problem even among the hard problems, but there are other computer science problems that had thousands of pages of literature devoted to them without much progress that when they were solved by someone turned out to be relatively simple. That 20/20 hindsight thing. To the extent that there is any special sauce in AGI, I expect it will look like one of these cases. Solving computer science problems is a pretty general skill, in part because it is a pretty shallow field in most important respects. To use AI research as an example, it is composed of only a handful of fundamental ideas from which a myriad of derivatives and mashups have been created. Most other problems in computer science have the same feature, and when problems get solved it is because someone looked at the handful of fundamentals and ignored the vast bodies of derivative products which add nothing new. Vast quantities of research does not equate to a significant quantity of ideas. AI is a little more complex than some other topics, but is still far simpler at the level of fundamentals than some people make it out to be. People are incapable of solving AGI for the same reason they are incapable of solving any of the other interesting computer science problems, which was the point I was making obliquely. It is not a different skill, it is the same skill that the vast majority of all computer science people are incompetent at. And AGI is particularly hard problem, even for that tiny minority of people capable of solving real problems in computer science. If you cannot solve interesting computer science problems that are likely to be simpler, then it is improbable
Re: [agi] Funding AGI research
On Nov 18, 2007 11:24 PM, Benjamin Goertzel [EMAIL PROTECTED] wrote: There are a lot of worthwhile points in your post, and a number of things I don't fully agree with, but I don't have time to argue them all right now... Instead I'll just pick two points: er, looks like that was three ;-) 1) The Babbages and Leibnizes of a given historical period are often visible only in HINDSIGHT. You can't say that there are no Babbages or Leibnizes of AGI around right now ... there could be some on this very list, unrecognized by you, but who will be recognized by all a few decades from now... 2) I don't think it's true that Babbage's or Leibniz's machines were specced out so much better than, say, Novamente. Relative to the technology of their time, plenty of details were left unspecified -- it just seems obvious to us now, in hindsight, how to fill in those details. It wasn't obvious to all their contemporaries. And while, in hindsight, the workability of their machines seems obvious to us, to their contemporaries it must have seemed like the workability of their machines required a huge leap of intuition. They had no rigorous mathematical proof of the workability of their machines, nor did they have working prototypes. They had conceptual arguments that pushed the boundaries of the science of their times, and seemed like nonsense to many of their contemporaries. 3) I don't agree that AGI is primarily a computer science problem, any more than, say, building a car is primarily a metalworking problem. AGI requires computer science problems to be solved as part of its solution; but IMO the essence of AGI-creation is not computer science. This seems to be a genuine difference of scientific intuition btw the two of us. Plenty of others whom I respect appear to share the same opinion as you. -- Ben G On Nov 18, 2007 11:04 PM, J. Andrew Rogers [EMAIL PROTECTED] wrote: On Nov 18, 2007, at 7:06 PM, Benjamin Goertzel wrote: Navigating complex social and business situations requires a quite different set of capabilities than creating AGI. Potentially they could be combined in the same person, but one certainly can't assume that would be the case. I completely agree. But if we are to assume that AGI requires some respectable amount of funding, as seems to be posited by many people, then it seems that it will require a person with broader skills than the stereotypical computer science nerd. In that case, maybe AGI is not accessible to someone who is unwilling or unable to be anything but a computer science nerd. As if the pool of viable AGI researchers was not small enough already. And, I don't think it's fair to say that if you're smart enough to solve AGI, you should be able to quickly make a pile of money doing some kind of more marketable technical-computer-science, and fund the AGI yourself. This assumes a lot of things, for instance that AGI is the same sort of problem as technical-computer-science problems, so that if someone can do AGI better than others, they must be able to do technical- computer-science better than others too. But I actually don't think this is true; I think that AGI demands a different sort of thinking. I'm not so sure about this. All hard problems seem to receive similar sentiments until they are actually solved. I do think that AGI is a relatively hard problem even among the hard problems, but there are other computer science problems that had thousands of pages of literature devoted to them without much progress that when they were solved by someone turned out to be relatively simple. That 20/20 hindsight thing. To the extent that there is any special sauce in AGI, I expect it will look like one of these cases. Solving computer science problems is a pretty general skill, in part because it is a pretty shallow field in most important respects. To use AI research as an example, it is composed of only a handful of fundamental ideas from which a myriad of derivatives and mashups have been created. Most other problems in computer science have the same feature, and when problems get solved it is because someone looked at the handful of fundamentals and ignored the vast bodies of derivative products which add nothing new. Vast quantities of research does not equate to a significant quantity of ideas. AI is a little more complex than some other topics, but is still far simpler at the level of fundamentals than some people make it out to be. People are incapable of solving AGI for the same reason they are incapable of solving any of the other interesting computer science problems, which was the point I was making obliquely. It is not a different skill, it is the same skill that the vast majority of all computer science people are incompetent at. And AGI is particularly hard problem, even
Re: [agi] Funding AGI research
On Nov 17, 2007 1:08 PM, Dennis Gorelik [EMAIL PROTECTED] wrote: Jiri, Give $1 for the research to who? Research team can easily eat millions $$$ without producing any useful results. If you just randomly pick researchers for investment, your chances to get any useful outcome from the project is close to zero. The best investing practise is to invest only into such teams that produced working prototype already. Dennis, What you are advocating is to fund Development but not Research. I think the history of science in the last century shows that funding Research as well as Development can be extremely valuable. Of course, each individual Research team has only a small chance of success if viewed from the big-picture perspective; but the $300M/year annual funding that was posited could fund a lot of AGI Research teams. Let's say it was used to fund 300 research teams at a rate of $1M/year per team. The odds are high that more than one of these teams would produce breakthrough discoveries within a 5-10 year period, even though betting on any individual team would be a long-shot. The US Research funding establishment is doing a good job of funding many sorts of research, but not AGI or life-extension. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=66242905-752b8c
Re: [agi] Funding AGI research
Richard, Though we have theoretical disagreements, I largely agree with your analysis of the value of prototypes for AGI. Experience has shown repeatedly that prototypes displaying apparently intelligent behavior in various domains are very frequently dead-ends, because they embody various sorts of cheating. And, if AGI really is a complex emergent phenomenon that requires a certain sort of large complex system in order to come about, then one would not expect that any kind of cheap, small-scale prototype would be able to demonstrate it. As large-scale funding requires impressive prototypes, one is then faced with an irritating task of creating prototypes that fulfill the largely unrelated goals of -- looking impressive to investors who want to see prototypes -- actually being meaningful steps on the path to AGI It is actually surprisingly difficult to find ways to fulfill these two goals at the same time. I'm hoping we'll be there, with Novamente, sometime in late 2008 or early 2009. I don't think our initial virtual animals will be impressive enough as AGI to qualify -- though they'll be really cool teachable animals!! -- but, I think virtual agents with language learning facility will pass the threshold... But even once we get there (assuming we do), my own faith in the system shown-off as a path to AGI will be largely uncorrelated with its impressiveness as a prototype... ben On Nov 17, 2007 1:41 PM, Richard Loosemore [EMAIL PROTECTED] wrote: Dennis Gorelik wrote: Jiri, Give $1 for the research to who? Research team can easily eat millions $$$ without producing any useful results. If you just randomly pick researchers for investment, your chances to get any useful outcome from the project is close to zero. The best investing practise is to invest only into such teams that produced working prototype already. Serious funding is usually helpful only to scale prototype up. (See how it worked out for Google, for example). So far there is no working prototype of AGI yet, therefore there is no point to invest. On the other hand some narrow AI teams already produced some useful results. Such teams deserve investments. When narrow AI field is mature enough -- making next step to AGI would be possible for self-funding AGI research team. Although this seems like a reasonable stance, I don't think it is a strategy that will lead the world to the fast development (or perhaps any development) of a real AGI. Allow me to explain why I think this. I agree you would not just pick researchers at random, but on the other hand if you insist on a team with a working prototype this might well be a disaster. I am in a position to use massive investment straight away (and I have a project plan that says how), but the specific technical analysis of the AGI problem that I have made indicates that nothing like a 'prototype' is even possible until after a massive amount of up-front effort. There are things we can do ahead of time (and some of those are underway), but if anyone asks for a prototype that does some fraction of the task, I can only point to the technical analysis and ask the investor to understand why this is not possible. Catch 22. No prototype, no investment; no investment, no prototype. Investors are leery of sorry, no prototype! claims (with good reason, generally) but they are also not tech-savvy enough to comprehend the technical analysis that tells them that they should make an exception in this case. And even worse, the technical community (for reasons I have explained, to general annoyance ;-) ) has reasons for disliking the particular technical analysis I have offered. If I turn out to be right in my analysis, none of the people who have what they claim to be prototypes will actually reach the goal of a viable AGI. (They disagree, of course!). Richard Loosemore Wednesday, October 31, 2007, 11:50:12 PM, you wrote: I believe AGI does need promoting. And it's IMO similar with the immortality research some of the Novamente folks are involved in. It's just unbelievable how much money (and other resources) are being used for all kinds of nonsense/insignificant projects worldwide. I wish every American gave just $1 for AGI and $1 for immortality research. Imagine what this money could for all of us (if used wisely). Unfortunately, people will rather spend the money for their popcorn in a cinema. Godlike intelligence? :) Ok, here is how I see it: If we survive, I believe we will eventually get plugged into some sort of pleasure machine and we will not care about intelligence at all. Intelligence is a useless tool when there are no problems and no goals to think about. We don't really want any goals/problems in our minds. Basically, the goal is to not have goal(s) and safely experience as intense pleasure as the available design allows for as long as possible. AGI could be
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
About PolyWorld and Alife in general... I remember playing with PolyWorld 10 years ago or so And, I had a grad student at Uni. of Western Australia build a similar system, back in my Perth days... (it was called SEE, for Simple Evolving Ecology. We never published anything on it, as I left Australia in the middle of the research...) But after fiddling with stuff like this a while, it becomes clear that, just as each GOFAI or machine learning program can be pushed so far and no further; similarly each Alife program can be pushed so far and no further... One of the most fascinatng busts in that area was Tom Ray's attempt to induce robust virtual evolution of multicellular life. I forget the name of his project but he was doing it at ATR in Japan. It was a follow-up to his excellently successful Tierra program, which was the first to demonstrate biology-like reproduction in artificial organisms Anyway Tom's attempt and many others to to beyond the complexity threshold observed in Alife programs did not pan out... Overall, I came away from my flirtation with Alife with the impression that it was doomed due to the lack of a viable artificial chemistry (chemistry arguably being the source of the richness of real biology). So, there was some cool work on artificial chemistry of a sort, done by Walter Fontana and many others, which I don't remember very well... The deep question I came away with was: What exactly are the **abstract properties** of the periodic table of elements that allows it to give rise to chemical compounds and ensuing biological structures with so much complexity? And then I decided Alife was not gonna be a shortcut and turned wholly to AI insetad ;-) Thing is, I'm sure Alife can work, but the computational requirements have gotta be way way bigger than for AI. And conceptually, it doesn't seem like Alife is really a shortcut -- because puzzling out the requirements that artificial chemistry needs to have, in order to support robust artificial biology, seems just as hard or harder than building a simulated brain or a non-brain-based AGI. After all it's not like we know how real chemistry gives rise to real biology yet --- the dynamics underlying protein-folding remain ill-understood, etc. etc. So I find this a deep and fascinating area of research (the borderline btw artificial chemistry and artificial biology, more so than Alife proper), but I doubt it's a shortcut to AGI ... though it would be cool to be proven wrong ;-) -- Ben G On Nov 15, 2007 3:30 AM, Bob Mottram [EMAIL PROTECTED] wrote: Although I thought this was a good talk and I liked the fellow presenting it to me it seems fairly clear that little or no progress has been made in this area over the last decade or so. In the early 1990s I wrote somewhat similar simulations where agents had their own neural networks whose architecture was specified by a genetic algorithm, but just like the speaker I came up against similar problems. As the guy says it should be in principle possible to go all the way from simple types of creatures up to more complex ones, like humans. In practice though what tends to happen is that the complexity of the neural nets reaches a plateau from which little subsequent progress occurs. Even after allowing the system to run for tens of thousands of generations not much of interest happens. I think the main problem here is the low complexity of the environment and the agents themselves. In a real biological system there are all kinds of niches which can be exploited in a variety of ways, but in polyworld (and other similar simulations) it's all very homogeneous. Real biological creatures are coalitions of millions of cells, each of which is a chemical factory containing an abundance of nano machinery, each of which is a possible site for evolutionary change. The sensory systems of real creatures are also far richer than simply being able to detect three colours (even molluscs can do better than this), and this is obviously a limiting factor upon the development of greater intelligence. On 15/11/2007, Jef Allbright [EMAIL PROTECTED] wrote: This may be of interest to the group. http://video.google.com/videoplay?docid=-112735133685472483 This presentation is about a potential shortcut to artificial intelligence by trading mind-design for world-design using artificial evolution. Evolutionary algorithms are a pump for turning CPU cycles into brain designs. With exponentially increasing CPU cycles while our understanding of intelligence is almost a flat-line, the evolutionary route to AI is a centerpiece of most Kurzweilian singularity scenarios. This talk introduces the Polyworld artificial life simulator as well as results from our ongoing attempt to evolve artificial intelligence and further the Singularity. Polyworld is the brain child of Apple Computer Distinguished Scientist Larry Yaeger, who remains the primary
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
I think that linguistic interaction with human beings is going to be what lifts Second Life proto-AGI's beyond the glass ceiling... Our first SL agents won't have language generation or language learning capability, but I think that introducing it is really essential, esp. given the limitations of SL as a purely physical environment... ben On Nov 15, 2007 1:38 PM, Bob Mottram [EMAIL PROTECTED] wrote: Which raises the question of whether the same complexity glass ceiling will be encountered when running AGI controlled agents within Second Life. SL is probably more complex than polyworld, although that could be debatable depending upon your definition of complexity. One factor which would raise the bar would be the additional baggage being introduced into the virtual world from the first life of human participants. On 15/11/2007, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 02:30, Bob Mottram wrote: I think the main problem here is the low complexity of the environment Complex programs can only be written in an environment capable of bearing that complexity: http://sl4.org/archive/0710/16880.html - Bryan - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=65511033-66e22b
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
On Nov 15, 2007 8:57 PM, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 08:16, Benjamin Goertzel wrote: non-brain-based AGI. After all it's not like we know how real chemistry gives rise to real biology yet --- the dynamics underlying protein-folding remain ill-understood, etc. etc. Can anybody elaborate on the actual problems remaining (beyond etc. etc.-- which is appropriate from Ben who is most notably not a biochemist/chemist/bioinformatician)? Hey -- That is a funny comment -- I've published a dozen bioinformatics papers in the last 5 years, and am CEO / Chief Scientist of a bioinformatics company (Biomind LLC, www.biomind.com) I am no chemist but I'm pretty much an expert on analyzing microarray and SNP data, and various other corners of bioinformatics, having introduced some funky new techniques into the field. In fact my most popular research paper is not on AGI but rather on Chronic Fatigue Syndrome -- it was the first-ever paper giving evidence for a (weak) genetic basis for CFS. -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=65715822-29017b
Re: [agi] Polyworld: Using Evolution to Design Artificial Intelligence
No worries!! just wanted to clarify... To address your question more usefully: There is soo much evidence that chemistry is subtly important for biology in ways that are poorly understood. In neuroscience for instance the chemistry of synaptic transmission btw neurons is still weakly understood, so we still don't know exactly how poor a model the formal neuron used in computer science is As a single example you have both ionotropic and metabotropic glutamate receptors along neurons ... whose synaptic transmission properties depend on ambient chemistry in the intracellular medium in ways no one understands really.. etc. etc. etc. ;-) ben On Nov 15, 2007 10:07 PM, Bryan Bishop [EMAIL PROTECTED] wrote: On Thursday 15 November 2007 20:02, Benjamin Goertzel wrote: On Nov 15, 2007 8:57 PM, Bryan Bishop [EMAIL PROTECTED] wrote: Can anybody elaborate on the actual problems remaining (beyond etc. etc.-- which is appropriate from Ben who is most notably not a biochemist/chemist/bioinformatician)? Hey -- That is a funny comment Oh my. This is a big, big mistake on my part. I am sorry. Please accept my apologies .. and the knowledge that my parenthetical comment no longer applies. - Bryan - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=65726263-86dc00
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
Hi, No: the real concept of lack of grounding is nothing so simple as the way you are using the word grounding. Lack of grounding makes an AGI fall flat on its face and not work. I can't summarize the grounding literature in one post. (Though, heck, I have actually tried to do that in the past: didn't do any good). FYI, I have read the symbol-grounding literature (or a lot of it), and generally found it disappointingly lacking in useful content... though I do agree with the basic point that non-linguistic grounding is extremely helpful for effective manipulation of linguistic entities... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64981284-09925d
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
Richard, So here I am, looking at this situation, and I see: AGI system intepretation (implicit in system use of it) Human programmer intepretation and I ask myself which one of these is the real interpretation? It matters, because they do not necessarily match up. That is true, but in some cases they may approximate each other well.. In others, not... This happens to be a pretty simple case, so the odds of a good approximation seem high. The human programmer's intepretation has a massive impact on the system because all the inference and other mechanisms are built around the assumption that the probabilities mean a certain set of things. You manipulate those p values, and your manipulations are based on assumptions about what they mean. Well, the PLN inference engine's treatment of ContextLink home InheritanceLink Bob_Yifu friend is in no way tied to whether the system's implicit interpretation of the ideas of home or friend are humanly natural, or humanly comprehensible. The same inference rules will be applied to cases like ContextLink Node_66655 InheritanceLink Bob_Yifu Node_544 where the concepts involved have no humanly-comprehensible label. It is true that the interpretation of ContextLink and InheritanceLink are fixed by the wiring of the system, in a general way (but what kinds of properties are referred to by them may vary in a way dynamically determined by the system). In order to completely ground the system, you need to let the system build its own symbols, yes, but that is only half the story: if you still have a large component of the system that follows a programmer-imposed interpretation of things like probability values attached to facts, you have TWO sets of symbol-using mechanisms going on, and the system is not properly grounded (it is using both grounded and ungrounded symbols within one mechanism). I don't think the system needs to learn its own probabilistic reasoning rules in order to be an AGI. This, to me, is too much like requiring that a brain needs to learn its own methods for modulating the conductances of the bundles of synapses linking between the neurons in cell assembly A and cell assembly B. I don't see a problem with the AGI system having hard-wired probabilistic inference rules, and hard-wired interpretations of probabilistic link types. But the interpretation of any **particular** probabilistic relationship inside the system, is relative to the concepts and the empirical and conceptual relationships that the system has learned. You may think that the brain learns its own uncertain inference rules based on a lower-level infrastructure that operates in terms entirely unconnected from ideas like uncertainty and inference. I think this is wrong. I think the brain's uncertain inference rules are the result, on the cell assembly level, of Hebbian learning and related effects on the neuron/synapse level. So I think the brain's basic uncertain inference rules are wired-in, just as Novamente's are, though of course using a radically different infrastructure. Ultimately an AGI system needs to learn its own reasoning rules and radically modify and improve itself, if it's going to become strongly superhuman! But that is not where we need to start... -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64998317-8c4281
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
On Nov 14, 2007 1:36 PM, Mike Tintner [EMAIL PROTECTED] wrote: RL:In order to completely ground the system, you need to let the system build its own symbols Correct. Novamente is designed to be able to build its own symbols. what is built-in, are mechanisms for building symbols, and for probabilistically interrelating symbols once created... ben g V. much agree with your whole argument. But - I may well have missed some vital posts - I have yet to get the slightest inkling of how you yourself propose to do this. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=65100803-21ddd3
[agi] Human uploading
Richard, I recently saw a talk by Todd Huffman at the Foresight Unconference on the topic of mind uploading technology, and he was specifically showing off techniques for imaging slices of brain, that *do* give the level of biological detail you're thinking of. Topics of discussions were, for example, inferring synaptic strength indirectly from mitochondrial activity. So, the Connectome people may not be taking a sufficiently fine-grained approach to support mind-uploading, but others are trying... Obviously, a detailed map of the brain at the level Todd is thinking of, would be of more than peripheral interest to cognitive scientists. It would not resolve cognitive questions in itself, but would be a wonderful trove of data to use to help validate or refute cognitive theories. -- Ben G On Nov 13, 2007 10:11 AM, Richard Loosemore [EMAIL PROTECTED] wrote: Bryan Bishop wrote: On Monday 12 November 2007 22:16, Richard Loosemore wrote: If anyone were to throw that quantity of resources at the AGI problem (recruiting all of the planet), heck, I could get it done in about 3 years. ;-) I have done some research on this topic in the last hour and have found that a Connectome Project is in fact in the very early stages out there on the internet: http://iic.harvard.edu/projects/connectome.html http://acenetica.blogspot.com/2005/11/human-connectome.html http://acenetica.blogspot.com/2005/10/mission-to-build-simulated-brain.html http://www.indiana.edu/~cortex/connectome_plos.pdfhttp://www.indiana.edu/%7Ecortex/connectome_plos.pdf This is the whole brain emulation approach, I guess (my previous comments were about evolution of brains rather than neural level duplication). But (switching topics to whole brain emulation) there are serious problems with this. It seems quite possible that what we need is a detailed map of every synapse, exact layout of dendritic tree structures, detailed knowledge of the dynamics of these things (they change rapidly) AND wiring between every single neuron. When I say it seems possible I mean that the chance of this information being absolutely necessary in order to understand what the neural system is doing, is so high that we would not want to gamble on them NOT being necessary. So are the researchers working at that level of detail? Egads, no! Here's a quote from the PLOS Computational Biology paper you referenced (above): Attempting to assemble the human connectome at the level of single neurons is unrealistic and will remain infeasible at least in the near future. They are not even going to do it at the resolution needed to see individual neurons?! I think that if they did the whole project at that level of detail it would amount to a possibly interesting hint at some of the wiring, of peripheral interest to people doing work at the cognitive system level. But that is all. I think it would be roughly equivalent to the following: You say to me I want to understand how computers work, in enough detail to build my own and I reply with I can get a you a photo of a motherboard and a 500 by 500 pixel image of the inside of an Intel chip... Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64558273-86797b
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
view that the current direction of Novamente is -- pick one: a) a needle in an infinite haystack or b) too fragile to succeed -- particularly since I'm pretty sure that you couldn't convince me without making some serious additions to Novamente - Original Message - *From:* Benjamin Goertzel mailto:[EMAIL PROTECTED] *To:* agi@v2.listbox.com mailto:agi@v2.listbox.com *Sent:* Monday, November 12, 2007 3:49 PM *Subject:* Re: [agi] What best evidence for fast AI? To be honest, Richard, I do wonder whether a sufficiently in-depth conversation about AGI between us would result in you changing your views about the CSP problem in a way that would accept the possibility of Novamente-type solutions. But, this conversation as I'm envisioning it would take dozens of hours, and would require you to first spend 100+ hours studying detailed NM materials, so this seems unlikely to happen in the near future. -- Ben On Nov 12, 2007 3:32 PM, Richard Loosemore [EMAIL PROTECTED] mailto:[EMAIL PROTECTED] wrote: Benjamin Goertzel wrote: Ed -- Just a quick comment: Mark actually read a bunch of the proprietary, NDA-required Novamente documents and looked at some source code (3 years ago, so a lot of progress has happened since then). Richard didn't, so he doesn't have the same basis of knowledge to form detailed comments on NM, that Mark does. This is true, but not important to my line of argument, since of course I believe that a problem exists (CSP), which we have discussed on a number of occasions, and your position is not that you have some proprietary, unknown-to-me solution to the problem, but rather that you do not really think there is a problem. Richard Loosemore - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; http://v2.listbox.com/member/?; This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; http://v2.listbox.com/member/?; This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64606349-2f1f37
Re: [agi] What best evidence for fast AI?
For example, what is the equivalent of the activation control (or search) algorithm in Google sets. They operate over huge data. I bet the algorithm for calculating their search or activation is relatively simple (much, much, much less than a PhD theses) and look what they can do. So I think one path is to come up with applications that can use and reason with large data, having roughly world knowledge-like sparseness, (such as NL data) and start with relatively simple activation algorithms and develop then from the ground up. Google, I believe, does reasoning about word and phrase co-occurrence using a combination of Bayes net learning with EM clustering (this is based on personal conversations with folks who have worked on related software there). The use of EM helps the Bayes net approach scale. Bayes nets are good for domains like word co-occurence probabilities, in which the relevant data is relatively static. They are not much good for real-time learning. Unlike Bayes nets, the approach taken in PLN and NARS allows efficient uncertain reasoning in dynamic environments based on large knowledge bases (at least in principle, based on the math, algorithms and structures; we haven't proved it yet). -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64609544-b69ea5
Re: [agi] Human uploading
Yes, I thought I had heard of people trying more ambitious techniques, but in the cases I heard of (can't remember where now) the tradeoffs always left the approach hanging on one of the issues: for example, was he talking about scanning microchondrial activity in vivo, in real time, across the whole brain?!! The mind boggles. [Uh, and it probably would, if you were the subject]. Some people think they can do very thin slices, but they are in defuncto, not in vivo. Yes, Todd believes (like most mind uploading experts) that the most practical approach to mind uploading in the near term is to slice a dead brain and scan it in. Doing uploading on live brains is bound to be far more technologically demanding, so it makes sense to focus on uploading fresh-killed brains first. Couldn't see any good references to this. It was a talk, not a publication. Not sure if it was videotaped or not. -- Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64610913-6e5f3d
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
On Nov 13, 2007 2:37 PM, Richard Loosemore [EMAIL PROTECTED] wrote: Ben, Unfortunately what you say below is tangential to my point, which is what happens when you reach the stage where you cannot allow any more vagueness or subjective interpretation of the qualifiers, because you have to force the system to do its own grounding, and hence its own interpretation. I don't see why you talk about forcing the system to do its own grounding -- the probabilities in the system are grounded in the first place, as they are calculated based on experience. The system observes, records what it sees, abstracts from it, and chooses actions that it guess will fulfill its goals. Its goals are ultimately grounded in in-built feeling-evaluation routines, measuring stuff like amount of novelty observed, amount of food in system etc. So, the system sees and then acts ... and the concepts it forms and uses are created/used based on their utility in deriving appropriate actions. There is no symbol-grounding problem except in the minds of people who are trying to interpret what the system does, and get confused. Any symbol used within the system, and any probability calculated by the system, are directly grounded in the system's experience. There is nothing vague about an observation like Bob_Yifu was observed at time-stamp 599933322, or a fact Command 'wiggle ear' was sent at time-stamp 54. These perceptions and actions are the root of the probabilities the system calculated, and need no further grounding. What you gave below was a sketch of some more elaborate 'qualifier' mechanisms. But I described the process of generating more and more elaborate qualifier mechanisms in the body of the essay, and said why this process was of no help in resolving the issue. So, if a system can achieve its goals based on choosing procedures that it thinks are likely to achieve its goals, based on the knowledge it gathered via its perceived experience -- why do you think it has a problem? I don't really understand your point, I guess. I thought I did -- I thought your point was that precisely specifying the nature of a conditional probability is a rats-nest of complexity. And my response was basically that in Novamente we don't need to do that, because we define conditional probabilities based on the system's own knowledge-base, i.e. Inheritance A B .8 means If A and B were reasoned about a lot, then A would (as measred by an weighted average) have 80% of the relationships that B does But apparently you were making some other point, which I did not grok, sorry... Anyway, though, Novamente does NOT require logical relations of escalating precision and complexity to carry out reasoning, which is one thing you seemed to be assuming in your post. Ben - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64644318-8bbdee
Re: [agi] What best evidence for fast AI?
This is the thing that I think is relevent to Robin Hanson's original question. I think we can build 1+2 is short order, and maybe 3 in a while longer. But the result of 1+2+3 will almost surely be an idiot-savant: knows everything about horses, and can talk about them at length, but, like a pedantic lecturer, the droning will put you asleep. So is there more to AGI, and exactly how do way start laying hands on that? --linas I think that evolutionary-learning-type methods play a big role in creativity. I elaborated on this quite a bit toward the end of my 1997 book From Complexity to Creativity. Put simply, inference is ultimately a local search method -- inference rules, even heuristic and speculative ones, always lead you step by step from what you know into the unknown. This makes you, as you say, like a pedantic lecturer. OTOH, evolutionary algorithms can take big creative leaps. This is one reason why the MOSES evolutionary algorithm plays a big role in the Novamente design (the other, related reason being that evolutionary learning is better than logical inference for many kinds of procedure learning). Integrating evolution with logic is key to intelligence. The brain does it, I believe, via -- implementing logic via Hebbian learning (neuron-level Hebb stuff leading to PLN-like logic stuff on the neural-assembly level) -- implementing evolution via Edelman-style Neural Darwinist neural map evolution (which ultimately bottoms out in Hebbian learning too) Novamente seeks to enable this integration via grounding both inference and evolutionary learning in probability theory. -- Ben G -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64667888-a48aa3
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
But has a human, asking Wen out on a date, I don't really know what Wen likes cats ever really meant. It neither prevents me from talking to Wen, or from telling my best buddy that ...well, I know, for instance, that she likes cats... yes, exactly... The NLP statement Wen likes cats is vague in the same way as the Novamente or NARS relationship EvaluationLink likes ListLink Wen cats is vague The vagueness passes straight from NLP into the internal KR, which is how it should be. And that same vagueness may be there if the relationship is learned via inference based on experience, rather than acquired by natural language. I.e., if the above relationship is inferred, it may just mean that {the relationship between Wen and cats} shares many relationships with other person/object relationships that have been categorized as 'liking' before In this case, the system can figure out that Wen likes cats without ever actually making explicit what this means. All it knows is that, whatever it means, it's the same thing that was meant in other circumstances where liking was used as a label. So, vagueness can not only be important into an AI system from natural language, but also propagated around the AI system via inference. This is NOT one of the trickier things about building probabilistic AGI, it's really kind of elementary... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64674694-3ada83
Re: Essay - example of how the CSP bites [WAS Re: [agi] What best evidence for fast AI?]
So, vagueness can not only be important imported, I meant into an AI system from natural language, but also propagated around the AI system via inference. This is NOT one of the trickier things about building probabilistic AGI, it's really kind of elementary... -- Ben G - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244id_secret=64674943-4b25e0