Agreed, internal simulation is necessary, and then real world testing (& correction/adjustment) must follow. But at what level of granularity must the simulation proceed? Why must it be a physical/visual simulation, rather than (or in conjunction with) a symbolic one?
I don't care for strict Boolean logic anymore than you. I think it's an oversimplification. Things aren't just true or false. They can be senseless/meaningless or true half the time or true unless you get surprised by something, etc. etc. etc. But there are ways to represent that in semantic networks. Given a non-Boolean representation of truth, it becomes possible to weigh conflicting information and alternative explanations or expectations. And putting these into a semantic network that is capable of representing a story or chain of causally related events (or even a branching tree if you're thinking about alternative possibilities) allows an event-based symbolic simulation. Furthermore, in cases where physical movement in a 3D space is required, a physical model can be consulted for finer-grained modeling when an action's consequences aren't clearly discernable from the semantic information. People often visualize things unrealistically. I'm reading Steven Pinker's How the Mind Works, and he mentions an experiment where people predicted that a ball shot out of a corkscrew tube would continue on a corkscrew path. Of course, when shown a video of it actually happening that way, the same people laughed at the absurdity of the image. If you've already thought about the scenario, however, and you're asked the same question again, you don't need to visualize it all over again to know it's BS. You just pop out with, "Of course not!" and laugh, because you've stored the information away semantically for rapid contextually-based retrieval. This semantic information is accumulated about all kinds of things, all the time. I think the key difference between people and animals is that we've evolved means for manipulating semantic information directly, taking a shortcut past all the physical modeling, and, more importantly, *communicating* that semantic information directly via speech, taking a shortcut past all the physical experience. How much computation does it take to run a 3D simulation, vs. bumping a few fuzzy truth values up against each other to get a rough estimate? Why wouldn't evolution engineer that computational efficiency into us? -- Sent from my Palm Pre On Oct 20, 2012 2:16 PM, Mike Tintner <[email protected]> wrote: Aaron: The human brain is defined by this "box", the drive to identify and collect useful new concepts, ...These are hardcoded into the human condition by our genes. They are algorithms. Why can't we copy them, or improve on them, in software?. Why can’t science work without experiments or investigation? Why can’t we go back to pre-science, ivory tower, in-a-box/book, natural philosophy? I have outlined at length before why concepts are fluid schemas/outlines - and not just outlines on a mental canvas but “outlines of [embodied] action”. DRAW ME A CHAIR HAND ME THAT BOX are not and never could be hardcoded or algorithmic or disembodied – they are there like all concepts to direct/embrace an infinite diversity of possible actions in an infinite diversity of possible real world situations, incl. an infinite diversity of *new* actions and situations. And a real world AGI – you – must not just check like logic whether those ideas are logically *true* but, like science, first simulate them in an embodied way internally in order to start to *realise* them and see whether they are *realistic* when compared with the relevant *evidence* you have about such actions, and then like science must put them into real world action and test whether they *really* work., But no time now... I mainly wanted to say thanks for your personal response – sure I bash too much. From: [email protected] Sent: Saturday, October 20, 2012 7:33 PM To: AGI Subject: Re: [agi] ONE EXAMPLE This is by far the best post I've seen come out of you, Mike. You started talking about what you think instead of just bashing what other people think. If you continue this way, it might be possible to come to the point of mutual understanding, if not agreement, instead of just frustrated gnashing of teeth on both sides. Having said that, I have a couple of points I disagree with you on. 1) Being on a desk, in a pocket, or roving around in a robot body is not what makes the difference between the chef/program that remixes the same tired old ingredients/concepts to synthesize new recipes/behaviors/thoughts and the chef/program that goes looking for new ingredients/concepts to make something fundamentally new. What makes the difference is *interaction* with the real world. A robot that's blind and deaf isn't going to be able to incorporate new concepts any better than a PC on a desk that's equally blind and deaf. It's the ability to sense the world that's key to developing new concepts. Where you'll probably disagree with me, however, is that I think natural language in the form of text might be sufficient sensory input to develop a limited range of new concepts. This is not to say that true AGI could come out of a text-based interface, but that we could potentially get something of the same flavor but watered down. 2) You consider algorithms to be pre-planned courses of action, as opposed to improvised, ad hoc courses of action. This is true for your typical stereotype of an algorithm, e.g. using an algorithm specifically designed so the robot can make coffee or something. However, AGI is defined by the attempt to build something that comes up with these things on it's own, not one that just follows a pre-planned routine or picks from among a set of pre-planned routines. The human brain's learning mechanisms implement algorithms for recognizing new concepts and incorporating them, and the human brain's planning mechanisms implement algorithms that improvise new courses of action on the fly, often preempting those already being acted out. To change these mechanisms, you would have to change the person's DNA or rewire their brain. This ensures that we don't go haywire, which would be contrary to evolutionary fitness. These make the "box" that defines what we're capable of thinking of, including what kinds of new concepts or courses of action we're capable of coming up with. We are all limited in our level of insight. No one can think of everything, but for any one thing taken by itself, there's a good chance a person can think of it given the time. What AGIers want to build is this "box", not the contents. We're dissatisfied with standard AI because it focuses on the contents. This seems to be your complaint about AGI, as well. But I think really this is a failure of recognition, not of existence. We want to build programs that automatically recognize and incorporate new concepts, much as your chef with his new ingredients. But the chef himself is following a standard algorithm: collect and incorporate new ingredients. This is the "box" that grabs new contents instead of fuddling though the same ones over and over. We want to build programs that automatically generate new courses of action on the fly using these concepts they have accumulated through experience, much as your chef creates new recipes by recombining old and new ingredients. The chef follows a standard algorithm here, too: pick some ingredients, and pick some steps by which to combine and prepare them. This is the "box" that shakes up the contents it has already accumulated to invent a new combination thereof, which fits the needs of the moment. The human brain is defined by this "box", the drive to identify and collect useful new concepts, and the drive to utilize the concepts already collected towards meeting the person's needs. These are hardcoded into the human condition by our genes. They are algorithms. Why can't we copy them, or improve on them, in software? Yes, the algorithms are static, unchanging, but what they *do* -- identifying & collecting new concepts and using them in improvised, appropriate behavior -- is dynamic and fluid. -- Sent from my Palm Pre On Oct 20, 2012 7:24 AM, Mike Tintner <[email protected]> wrote: Jeez Ben, that is the longest most convoluted post .... You really are not registering (forget about agreeing with) the other side at all – and thinking in strictly retro, out-of-date terms. 1) Vis-a-vis your ideas – the accusation is simple: you do not have an idea for take-off – an idea that will explain how a machine will go on from one diverse task to another. Not even an attempt at an idea. Your magic sauce is eternally in the post. Such an idea would explain, say, how your robot would pick up not just one form of object, but another and another – and ultimately any object within the capacity of its effector – **all without being reprogrammed.** Or it would explain how a robot would go from one terrain to another to another – from rocky to sandy to beach – **all without being reprogrammed.** Or how an AGI would go from understanding a story about dogs, to a story about lions, giraffes, snakes etc – all without being reprogrammed. Nor BTW does the entire field have even an *attempt* at a take-off idea. 2) “AGI –gadfly”. I am not. I am an opponent of what you and most here represent which is **standalone computational AGI** (yes, I know you are making gestures towards robots, but they are not fundamental/ integral.) I OTOH am a proponent of what will be the next and first generation of **real AGI** – ROBOTICS AGI. 3) Creativity is what AGI is about – I agree with Deutsch/ he agrees with me. It is what will distinguish real AGI from narrow AI. Creativity does indeed involve a) the incorporation of, and b) adaptation, to **new elements**. And it is in no way a mechanical problem for a robot – it is merely impossible for a standalone computer. [The identification of *new elements* is key to defining creativity – precisely because any computer program can be called trivially creative. As I think you said, a chess program is trivially creative because it plays new games it has never played before. But does it introduce *new elements*? Does it introduce *new moves* or *new pieces*? No it doesn’t and can’t. So now we can define why it isn’t truly creative] By extension, most AGI-ers incl. you AFAICT, seem deeply confused about whether GA’s are creative or not. They are not – precisely because all they do is re-mix an existing set of old elements. Put this simply – let’s say you are a creative chef who wants to create a new dish. If you follow the GA approach, you will take an existing set of dishes with a limited set of ingredients, and you will endlessly remix them. You remix the same old elements. So if you want a new ice cream, you will take existing flavours and endlessly remix those. Lemon and vanilla and caramel etc. And what you will get will be new, but definitely still vanilla-y/caramel-y/lemon-y etc – still v. much of the same family. That is not true creativity or the way real world creativity, or real AGI, does or can proceed. The creative chef looks for NEW INGREDIENTS – new elements. The creative chef will be able to come up with **snail** ice cream or **eggs and bacon** ice cream or – what the heck – leather ice cream. How will he do this? No mechanical mystery – to put it v. simply : he will look around the world for new ingredients, reach out and pick them up and add them to his ice-cream mix/stew and mix them in – and see if they work. (This by the way is what improvisation is - nothing to do with those musical progs. which do not improvise at all – merely remix existing musical elements). The chef uses his robotic/embodied capacity to find new elements in the world – “objets trouves”. - both physical and mental elements. Standalone computer programs cannot do this. 4) The REAL BATTLE – so what is going on here is not a battle between you – guardian of the sacred flame – and some irritant gadfly – but a clash between two fundamentally opposed visions of how AGI should proceed a) STANDALONE, VIRTUAL WORLD AGI vs b) ROBOTIC, REAL WORLD AGI (and this can be framed also as a clash between standalone desktop computers OTOH and real-world-embodied-and-embedded tablets and mobile phones OTOH). And secondly, it is a clash between a) PRE-PLANNED COURSES OF ACTION (of wh. algorithms are one form) and b) IMPROVISED, AD HOC COURSES OF ACTION. Once you move out of your computational virtual world, . (as present robotics AGI-ers are doing), you will realise the central challenge of navigating the real, unstructured world - how on earth can you plan for a real world that is continually throwing new things at you – (new elements) – where you can never know what is around the next corner, or foresee every pitfall and stumbling block? And secondly, – though this is something you have not even begun to think about – why on earth would you *want* to plan for the real world – when there are so many new and better things to do – so many new and better ways to navigate? 5) CONCLUSION – So what is going on here is not a clash between heroic AGI-ers and some mad troll whose only apparent purpose in life is to cause them trouble , but between two grand overarching visions of what AGI is.- between a VIRTUAL, PREPLANNED, “DETACHED”, AGI and a ROBOTIC, IMPROVISED, “EMBODIED-AND-EMBEDDED” AGI. And just as Deutsch has just more or less echoed a whole set of points that I have been making for years, and squishy robots came along to support my fluid schemas, so others will come to support the still-new points that I am making. Although my vision has not been set out systematically here, it is indeed very extensive and very coherent – and has very practical consequences. A great deal of thought has gone into it. From: Ben Goertzel Sent: Saturday, October 20, 2012 4:45 AM To: AGI Subject: Re: [agi] ONE EXAMPLE *** As for Ben, he has never faced the problem of AGI in his life. Ask him what ideas he has for AGI take-off/creativity *** Mike, as you know I wrote a book on the nature of creativity in 1997 (From Complexity to Creativity) ... You unfortunately lack the scientific background to read it carefully.... And my views on AGI have been written down extensively, albeit not in sufficiently simplified language for you to understand.... ( I'm working to remedy that, with a popular-audience book on AGI in the works...) The kind of argument you are trying to make was made far more ably by George kampis in his mid-1990s book "self-modifying systems ..." , and I counter-argued his points extensively in my own book Far from new and radical, the view you're presenting is very familiar to me (and everyone else in the AGI field), we just don't agree with it Still, though you are (among other things) an under-educated, obnoxious mailing-list troll, the points you raise have been raised often enough by others with more incisive minds and better educations than you, that I feel moved to write a somewhat thorough response... The question of creativity and "radical novelty" is an interesting one that I've often discussed with others F2F... I just wrote a blog post http://multiverseaccordingtoben.blogspot.com/2012/10/can-computers-be-creative.html that addresses these issues. I preferred to write a blog post than a long email, as emails have more of a feeling of vanishing into the ether, whereas blog posts feel more persistent.. ben AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
