[agi] Emergent languages Org
Jeez there's always something new. Anyone know about this (which seems at a glance loosely relevant to Ben's approach) ? http://www.emergent-languages.org/ Overview This site provides an introduction to the research on emergent and evolutionary languages as conducted at the Sony Computer Science Laboratory in Paris and the AI-Lab at the VUB in Brussels. One of the principle objectives of this research is to identify the cognitive capabilities that artificial agents must posses to enable, in a population of such agents, the emergence and evolution of a language that exhibits characteristic features identified in natural languages. Looks like Sony- Aibo- financed. Luc Steels seems to be a principal figure. This is quite fun: http://www.csl.sony.fr/~py/clickerTraining.htm Here he explains/justifies his approach: http://www.csl.sony.fr/downloads/papers/2006/steels-06a.pdf And how did I get to all this? From, tangentially, Construction Grammar, which is yet another interesting aspect of cognitive linguistics: http://en.wikipedia.org/wiki/Construction_grammar - 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=93139505-4aa549
Re: [agi] Emergent languages Org
I havn't read any of Luc Steels stuff for a long time, but he has been researching the evolution of language using robots or software agents since the early 1990s. This is really a symbol grounding problem where the communication in some way needs to represent things or situations which the agent can perceive with its sensors. Some years ago I tried to do something similar to Pierre Oudeyers video using a humanoid robot - presenting objects and saying this is a... or what is this? or Is this a...?. I didn't go very far down this route because I found that visual recognition of objects constitutes the major part of the problem. It is possible to use SIFT features and geometric hashes (which I think is what the AIBO robot is doing in this demo) but these 2D methods just aren't very good on objects with complicated 3D shapes. Since I'm interested in making machines which are genuinely intelligent, as opposed to appearing to be intelligent in a five minute demo, I've spent most of my efforts on the 3D object recognition problem. It turns out that other things are fundamentally related to this problem, such as mapping, navigation and SLAM. On 03/02/2008, Mike Tintner [EMAIL PROTECTED] wrote: Jeez there's always something new. Anyone know about this (which seems at a glance loosely relevant to Ben's approach) ? http://www.emergent-languages.org/ Overview This site provides an introduction to the research on emergent and evolutionary languages as conducted at the Sony Computer Science Laboratory in Paris and the AI-Lab at the VUB in Brussels. One of the principle objectives of this research is to identify the cognitive capabilities that artificial agents must posses to enable, in a population of such agents, the emergence and evolution of a language that exhibits characteristic features identified in natural languages. Looks like Sony- Aibo- financed. Luc Steels seems to be a principal figure. This is quite fun: http://www.csl.sony.fr/~py/clickerTraining.htm Here he explains/justifies his approach: http://www.csl.sony.fr/downloads/papers/2006/steels-06a.pdf And how did I get to all this? From, tangentially, Construction Grammar, which is yet another interesting aspect of cognitive linguistics: http://en.wikipedia.org/wiki/Construction_grammar - 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=93142547-2c3dab
Re: [agi] Emergent languages Org
I have been collaborating with this lab on their Fluid Construction Grammar system, as described briefly in this blog post: http://texai.org/blog/2007/10/24/fluid-construction-grammar I downloaded their Common Lisp implementation and rewrote it in Java and demonstrated that I could achieve the same results as their Lisp implementation. Then I extended it to parse incrementally, e.g. word-by-word, strictly left-to-right, creating semantics at each step. I have not studied the theory of emerging languages as I am focused on what I think is their excellent production rule engine for bi-directional grammars. I would be glad to provide an introduction on the linkedin network to Pieter Wellens, who is a PhD student at the associated VUB AI-Lab. -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: Ben Goertzel [EMAIL PROTECTED] To: agi@v2.listbox.com Sent: Sunday, February 3, 2008 8:03:13 AM Subject: Re: [agi] Emergent languages Org Thanks for the references... I found this paper Kaplan, F., Oudeyer, P-Y., Kubinyi, E. and Miklosi, A. (2002) Robotic clicker training, Robotics and Autonomous Systems, 38(3-4), pp. 197--206. at (near the bottom) http://www.csl.sony.fr/~py/clickerTraining.htm interesting in terms of highlighting the difference btw virtual-world and physical-robotics teaching of agents, as well as the basic difference between Novamente's Virtual Animal Brain system and real dog brains... They point out that imitation learning is rarely used for teaching animals, both because animals are bad at imitation, and because of differences between human and animal anatomy. However, Novamente is good at imitation, and in a virtual-world context the differences between human and animal anatomy can be finessed pretty easily (via simply supplying the virtual animal with suggestions about how to map specific human-avatar animations into specific animal-avatar animations). What they advocate in the paper, for teaching robots, is clicker training which is basically Skinnerian reinforcement learning with a judicious, time-variant sequence of partial rewards. At first you reward the animal for doing 1/10 of the behavior right, then after it can do that, you reward it for doing 2/10 of the behavior right, etc. In their work on language learning http://www.csl.sony.fr/~py/languageAcquisition.htm I see nothing coming remotely close to a discussion of the learning of syntax or complex semantics ... what I see is some experiments in which robots learned, through spontaneous exploration and reinforcement, the simple fact that vocalizing toward other agents is a useful thing to do. Which is certainly interesting ... but it's really just a matter of learning THAT vocal communication exists, in a setting where not that many other possibilities exist... -- Ben G On Feb 3, 2008 7:08 AM, Mike Tintner [EMAIL PROTECTED] wrote: Jeez there's always something new. Anyone know about this (which seems at a glance loosely relevant to Ben's approach) ? http://www.emergent-languages.org/ Overview This site provides an introduction to the research on emergent and evolutionary languages as conducted at the Sony Computer Science Laboratory in Paris and the AI-Lab at the VUB in Brussels. One of the principle objectives of this research is to identify the cognitive capabilities that artificial agents must posses to enable, in a population of such agents, the emergence and evolution of a language that exhibits characteristic features identified in natural languages. Looks like Sony- Aibo- financed. Luc Steels seems to be a principal figure. This is quite fun: http://www.csl.sony.fr/~py/clickerTraining.htm Here he explains/justifies his approach: http://www.csl.sony.fr/downloads/papers/2006/steels-06a.pdf And how did I get to all this? From, tangentially, Construction Grammar, which is yet another interesting aspect of cognitive linguistics: http://en.wikipedia.org/wiki/Construction_grammar - 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/?; -- Ben Goertzel, PhD CEO, Novamente LLC and Biomind LLC Director of Research, SIAI [EMAIL PROTECTED] If men cease to believe that they will one day become gods then they will surely become worms. -- Henry Miller - This list is sponsored by AGIRI:
[agi] A little more technical information about OpenCog
I got a free hour this afternoon, and posted a little more technical information about our plans for OpenCog, here: http://opencog.org/wiki/OpenCog_Technical_Information Nothing surprising or dramatic, mostly just a clear online explanation of our basic plans, as have already been discussed in various emails... -- Ben G p.s. for those who don't know what opencog is, see http://opencog.org/wiki/Main_Page -- Ben Goertzel, PhD CEO, Novamente LLC and Biomind LLC Director of Research, SIAI [EMAIL PROTECTED] If men cease to believe that they will one day become gods then they will surely become worms. -- Henry Miller -- Ben Goertzel, PhD CEO, Novamente LLC and Biomind LLC Director of Research, SIAI [EMAIL PROTECTED] If men cease to believe that they will one day become gods then they will surely become worms. -- Henry Miller - 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=93259758-229b0c
Re: [agi] Goal Driven Systems and AI Dangers [WAS Re: Singularity Outcomes...]
On 1/30/08, Richard Loosemore [EMAIL PROTECTED] wrote: Kaj, [This is just a preliminary answer: I am composing a full essay now, which will appear in my blog. This is such a complex debate that it needs to be unpacked in a lot more detail than is possible here. Richard]. Richard, [Where's your blog? Oh, and this is a very useful discussion, as it's given me material for a possible essay of my own as well. :-)] Thanks for the answer. Here's my commentary - I quote and respond to parts of your message somewhat out of order, since there were some issues about ethics scattered throughout your mail that I felt were best answered with a single response. The most important reason that I think this type will win out over a goal-stack system is that I really think the latter cannot be made to work in a form that allows substantial learning. A goal-stack control system relies on a two-step process: build your stack using goals that are represented in some kind of propositonal form, and then (when you are ready to pursue a goal) *interpret* the meaning of the proposition on the top of the stack so you can start breaking it up into subgoals. The problem with this two-step process is that the interpretation of each goal is only easy when you are down at the lower levels of the stack - Pick up the red block is easy to interpret, but Make humans happy is a profoundly abstract statement that has a million different interpretations. This is one reason why nobody has build an AGI. To make a completely autonomous system that can do such things as learn by engaging in exploratory behavior, you have to be able insert goals like Do some playing, and there is no clear way to break that statement down into unambiguous subgoals. The result is that if you really did try to build an AGI with a goal like that, the actual behavior of the system would be wildly unpredictable, and probably not good for the system itself. Further: if the system is to acquire its own knowledge independently from a child-like state (something that, for separate reasons, I think is going to be another prerequisite for true AGI), then the child system cannot possibly have goals built into it that contain statements like Engage in an empathic relationship with your parents because it does not have the knowledge base built up yet, and cannot understand such a propositions! I agree that it could very well be impossible to define explict goals for a child AGI, as it doesn't have enough built up knowledge to understand the propositions involved. I'm not entirely sure of how the motivation approach avoids this problem, though - you speak of setting up an AGI with motivations resembling the ones we'd call curiosity or empathy. How are these, then, defined? Wouldn't they run into the same difficulties? Humans have lots of desires - call them goals or motivations - that manifest in differing degrees in different individuals, like wanting to be respected or wanting to have offspring. Still, excluding the most basic ones, they're all ones that a newborn child won't understand or feel before (s)he gets older. You could argue that they can't be inborn goals since the newborn mind doesn't have the concepts to represent them and because they manifest variably with different people (not everyone wants to have children, and there are probably even people who don't care about the respect of others), but still, wouldn't this imply that AGIs *can* be created with in-built goals? Or if such behavior can only be implemented with a motivational-system AI, how does that avoid the problem of some of the wanted final motivations being impossible to define in the initial state? But beyond this technical reason, I also believe that when people start to make a serious efort to build AGI systems - i.e. when it is talked about in government budget speeches across the world - there will be questions about safety, and the safety features of the two types of AGI will be examined. I believe that at that point there will be enormous pressure to go with the system that is safer. This makes the assumption that the public will become aware of AGI being near well ahead of the time, and takes the possibility seriously. If that assumption holds, then I agree with you. Still, the general public seems to think that AGI will never be created, or at least not in hundreds of years - and many of them remember the overoptimistic promises of AI researchers in the past. If a sufficient amount of scientists thought that AGI was doable, the public might be convinced - but most scientists want to avoid making radical-sounding statements, so they won't appear as crackpots to the people reviewing their research grant applications. Combine this with the fact that the keys for developing AGI might be scattered across so many disciplines that very few people have studied them all, or that sudden breakthroughs may accelerate the research, I don't think it's a
Re: [agi] Emergent languages Org
I doubt that 3D object recognition is integral to 'genuine intelligence'. Theoretically, if we had an AGI we should be able to put it in a simulated 2D world and it would still act intelligently. IMO language is integral to strong AI in the same way that logic is integral to mathematics. If you think about it, human languages are basically higher order logics with fuzzy expressions, probabilities and context. That sounds like a fabulous description of the higher order logic our brains use internally to store thoughts. I haven't read any of Steels stuff lately, either. I'm not sure if any of the language he's generating is higher order, but I wouldn't be so quick to dismiss emergent language generation as a trick for just 5 minute demos. -J On Feb 4, 2008 12:34 AM, Bob Mottram [EMAIL PROTECTED] wrote: I havn't read any of Luc Steels stuff for a long time, but he has been researching the evolution of language using robots or software agents since the early 1990s. This is really a symbol grounding problem where the communication in some way needs to represent things or situations which the agent can perceive with its sensors. Some years ago I tried to do something similar to Pierre Oudeyers video using a humanoid robot - presenting objects and saying this is a... or what is this? or Is this a...?. I didn't go very far down this route because I found that visual recognition of objects constitutes the major part of the problem. It is possible to use SIFT features and geometric hashes (which I think is what the AIBO robot is doing in this demo) but these 2D methods just aren't very good on objects with complicated 3D shapes. Since I'm interested in making machines which are genuinely intelligent, as opposed to appearing to be intelligent in a five minute demo, I've spent most of my efforts on the 3D object recognition problem. It turns out that other things are fundamentally related to this problem, such as mapping, navigation and SLAM. On 03/02/2008, Mike Tintner [EMAIL PROTECTED] wrote: Jeez there's always something new. Anyone know about this (which seems at a glance loosely relevant to Ben's approach) ? http://www.emergent-languages.org/ Overview This site provides an introduction to the research on emergent and evolutionary languages as conducted at the Sony Computer Science Laboratory in Paris and the AI-Lab at the VUB in Brussels. One of the principle objectives of this research is to identify the cognitive capabilities that artificial agents must posses to enable, in a population of such agents, the emergence and evolution of a language that exhibits characteristic features identified in natural languages. Looks like Sony- Aibo- financed. Luc Steels seems to be a principal figure. This is quite fun: http://www.csl.sony.fr/~py/clickerTraining.htm Here he explains/justifies his approach: http://www.csl.sony.fr/downloads/papers/2006/steels-06a.pdf And how did I get to all this? From, tangentially, Construction Grammar, which is yet another interesting aspect of cognitive linguistics: http://en.wikipedia.org/wiki/Construction_grammar - 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=93269357-79b910
Re: [agi] Emergent languages Org
IMO language is integral to strong AI in the same way that logic is integral to mathematics. The counterargument is that no one has yet made an AI virtual chimp ... and nearly all of the human brain is the same as that of a chimp ... I think that language-centric approaches are viable, but I wouldn't dismiss sensorimotor-centric approaches to AGI either ... looking at evolutionary history, it seems that ONE way to achieve linguistic functionality is via some relatively minor tweaks on a prelinguistic mind tuned for flexible sensorimotor learning... (tho I don't believe this is the only way, unlike some) -- 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=93270421-5eade1
Re: [agi] Emergent languages Org
On Feb 4, 2008 11:27 AM, Ben Goertzel [EMAIL PROTECTED] wrote: IMO language is integral to strong AI in the same way that logic is integral to mathematics. The counterargument is that no one has yet made an AI virtual chimp ... and nearly all of the human brain is the same as that of a chimp ... I think that language-centric approaches are viable, but I wouldn't dismiss sensorimotor-centric approaches to AGI either ... looking at evolutionary history, it seems that ONE way to achieve linguistic functionality is via some relatively minor tweaks on a prelinguistic mind tuned for flexible sensorimotor learning... (tho I don't believe this is the only way, unlike some) -- Ben Interesting! You make a very good point. I'd be interested in what you see as the path from SLAM to AGI. To me, language generation seems obvious: 1. Make a language and algorithms for generating stuff in that language. 2. Implement pattern recognition and abstraction (imo not _that_ hard if you've designed your language well) 3. Ground the language through real-world sensorimotor experiences so the utterances mirror the agents' experiences. What do you see as the equivalent path from mapping, navigation and SLAM? -J - 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=93271502-5143d2
Re: [agi] Emergent languages Org
Hi, I'd be interested in what you see as the path from SLAM to AGI. To me, language generation seems obvious: 1. Make a language and algorithms for generating stuff in that language. 2. Implement pattern recognition and abstraction (imo not _that_ hard if you've designed your language well) 3. Ground the language through real-world sensorimotor experiences so the utterances mirror the agents' experiences. What do you see as the equivalent path from mapping, navigation and SLAM? Mapping, navigation and SLAM are not the key point -- embodied learning is the point ... these are just prerequisites... The robotics path to AI is a lot like the evolutionary path to natural intelligence... Create a system that learns to achieve simple sensorimotor goals in its environment... then move on to social goals... and language eventually emerges as an aspect of social interaction... Rather than language being a separate thing that is then grounded in experience, make language **emerge** from nonlinguistic interactions ... as it happened historically See Mithen's The Singing Neanderthals for ideas about how language may have emerged from prelinguistic sound-making ... and a host of researchers for ideas about how language may have emerged from gesture (I have a paper touching on the latter at novamente.net/papers ) -- 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=93273630-9e8239
Re: [agi] Emergent languages Org
On Feb 4, 2008 12:12 PM, Ben Goertzel [EMAIL PROTECTED] wrote: The robotics path to AI is a lot like the evolutionary path to natural intelligence... Create a system that learns to achieve simple sensorimotor goals in its environment... then move on to social goals... and language eventually emerges as an aspect of social interaction... You might be right, but I'm very skeptical. I don't see how any complex behaviour can simply 'emerge' from strict algorithms like SLAM. SLAM as I know it doesn't allow for emergent behavior at all (except for its explicit mapping ability). Eventually, you will have to write something which allows for emergent behaviour and complex communication. To me, that stage of your project is the interesting crux of AGI. It should have some very interesting emergant behaviour with inputs other than the information SLAM outputs. Why not just work on that difficult part now? -J - 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=93292780-7d3b9c