Aaron, Thanks for reply.
I’ll be brief, because I intend to come back to this later at length – it’s *hugely*, hugely important. Your approach to language is, broadly, standard AGI, standard philosophy, standard linguistics, standard last 2500 years of rational culture. Essentially this says that :language is primarily pictures of the world, whose meaning and truth we, as observers, must ascertain. Any application of language to action is secondary, to be tackled secondarily. This is totally upside down. Actually language is primarily enactive – consisting of guides to action rather than pictures – and there to direct action first, and fill books only secondarily. The latter is totally dependent on the second. Language (a conceptual system) begins and ends in action. Understanding this requires a revolution in our approach to language – and is an absolute sine qua non (indeed the v. foundation) of real AGI. And that’s why there are such huge problems of communication between me and y’all in this area. You can’t understand concepts until you look at their primary function, which is enactive. I always think about concepts in terms of their enaction, you all are thinking about them almost exclusively in terms of their meaning/truth decipherment – and so never really understand my points – or the central nature of an AGI. If you look at concepts in terms of meaning/truth, then it’s not wildly unreasonable to think that semantic nets might help. If you look at them in terms of action, they become absurd. All this is by way of a statement of intent, rather than a detailed argument. Re Woz, you’re new, I assume, and don’t know as most here do by now, that it refers to the Wozniak test, which does indeed require a robot to be able to respond to the command: GO TO THE KITCHEN (AND MAKE COFFEE). That is a true test of AGI (if a little complicated) and also, more specifically, of an AGI that can truly use concepts.. Really think about it - about all the kitchens, all the coffeepots, and all the intermediate *chairs* and furniture, such concepts embrace – and that a true Woz AGI must therefore be able to enactively navigate – and you’ll realise why semantic nets haven’t got a hope of working. From: [email protected] Sent: Wednesday, October 31, 2012 9:20 PM To: AGI Subject: Re: [agi] The Fundamental Misunderstanding in AGI [was Superficiality] The short term goal is to be able to represent & understand the meaning of natural language in both textual and conversational settings. This does entail being able to represent relationships between concepts across multiple contexts. The long term goal is to implement analytical thought, reasoning, and decision making on top of this representational scheme. Ultimately, this will require embodiment. I don't think the plasticity of thought which you emphasize is possible without first having an effective representation of meaning. This is why I'm starting by designing a robust representation, and only then moving on to reasoning. Of course I recognize that the system won't have real-world experience and therefore won't have informed common sense as we do. This means that initially it will make ridiculous mistakes, the sort where "everybody knows" the right answer, and it will have to be instructed on these mistakes via natural language. Later, when the system is embodied, experience will serve as the source of common sense information, and the level of natural language instruction will taper off until the system is capable of understanding without assistance. Children do not learn from words alone, and neither do I expect my system to, but it will have to make do in the interim. Only once reasoning and embodiment-derived common sense are in place will the system will be able to pass tests like you describe below, as opposed to merely representing their meaning. By the way, I don't think woz test means what you think it means. http://en.m.wikipedia.org/wiki/Wizard_of_Oz_experiment It seems the opposite of what you intend. I assume what you really mean by woz test is a real-world, task-oriented version of the Turing test. -- Sent from my Palm Pre -------------------------------------------------------------------------------- On Oct 31, 2012 5:49 AM, Mike Tintner <[email protected]> wrote: Quick question, Aaron. When you talk about “expressing”/representing these statements, is the goal of your net to *process text* - other instances of these concepts used in other texts?( Hence, I presume, your “truth values”). Do you have any additional goal of your machine *enacting* these statements? For example, GO TO THE KITCHEN (close to MAIN STREET below) is a form of the Woz test. Is your net intended to represent that, so a robot could *enact* it and pass the Woz test? From: Aaron Hosford Sent: Wednesday, October 31, 2012 12:32 AM To: AGI Subject: Re: [agi] The Fundamental Misunderstanding in AGI [was Superficiality] I see no contrast. These vague statements/strategies can be expressed quite nicely using soft/fuzzy/uncertain truth values applied to links in a semantic net. As for implementing the behavior they describe, that will come later, but as with any goal-oriented behavior, there will be (1) a goal description -- whether vague or precise, doesn't really matter -- and (2) a (most likely heuristic, a.k.a. imprecise) goal metric. The magic you perceive here is already accounted for. Defining imprecision or uncertainty lets you represent it effectively. If you don't define it, you can't represent it. If you can't represent it, you can't compute it. To build things you have to make things out of something. You are advocating we just conjure up uncertainty out of thin air. On Tue, Oct 30, 2012 at 2:16 PM, Mike Tintner <[email protected]> wrote: I would be surprised here, Aaron, if you are not a victim of the abuse of key terms throughout AI. Of course, there is a great deal of “fuzzy” logic – and I assume your use of “soft” is not a million miles from that. When you look into these logics, you find that actually they are being used with precision and precise values Contrast them with:. 1.WE NEED TO TAKE SOME POSITIVE ACTION HERE TO DEAL WITH THEIR THREATS. WE CAN’T LET THEM THINK WE’RE GOING TO FOOL AROUND. 2. FIRST WE NEED TO DEFINE THE PROBLEM – WE CAN’T START WORKING ON SOLUTIONS LIKE AGIERS BEFORE WE’VE EVEN DEFINED THE PROBLEM. THEN WE NEED TO DIG UP WHATEVER EVIDENCE WE CAN FIND, AND GRADUALLY GENERATE SOME IDEAS. OR WE COULD START WITH IDEAS, AND THEN CHECK OUT THE EVIDENCE. 3. LET’S GO TO THE MAIN SHOPPING STREET, AND NOSE AROUND TO SEE WHAT WE CAN FIND. These are examples of the kind of truly fluid (or soft, or vague) thinking that characterises human/real AGI thinking – and that are way beyond the compass of any logic or algo. They are also interdependently, as I think we’ve discussed, truly general – levels higher than logic and maths, .with their vague generalities as distinct from the latter’s specific generalities. From: [email protected] Sent: Tuesday, October 30, 2012 6:07 PM To: AGI Subject: Re: [agi] The Fundamental Misunderstanding in AGI [was Superficiality] You apparently missed a big chunk of the recent conversation between Jim & me. I am using "soft" truth values to represent so called facts because I assumed from the start that single, unambiguous meanings are merly artifacts of our perception. My system is built from the ground up with tools necessary for dealing with imprecision, uncertainty, and ambiguity. Jim was trying to convince me of the need for additional measures, and (within the limits of our mutual understanding) we agreed on that point. This is in direct contradition to your statements below about our use of rigid logic. I'm sure there are other groups out there doing the same for shapes and images. Where did you pick up this idea that we are stuck on simple geometric shapes and logic that only permits the simple yes/no dichotomy? And why do you think algorithms are restricted to them, too? -- Sent from my Palm Pre ------------------------------------------------------------------------------ On Oct 30, 2012 12:23 PM, Mike Tintner <[email protected]> wrote: Schema is a fluid outline – as distinct from a geometrically defined outline/pattern which is rigid.(You can geometrically define a moving wavy line – but it’s “rigidly/fixedly wavy”). The outline of a real waterdrop is a fluid outline. The outline of your hand grasping or your body moving is a fluid outline. They’re actually moving/changing – so you know that any shape they may have at a given moment is fluid and about to change. Another way to think of it is to look at any cartoon: https://www.google.com/search?num=10&hl=en&safe=off&site=imghp&tbm=isch&source=hp&biw=1362&bih=692&q=obama+cartoon&oq=obama+cartoon&gs_l=img.3..0l10.1269.3112.0.3741.13.9.0.2.2.0.131.589.8j1.9.0...0.0...1ac.1.YShcABKFARI We understand when we look at a cartoon outline of say Obama that that is an outline to be interpreted "*fluidly* and not literally. We understand that that outline is to be understood as saying “the lines of the real object are “SOMETHING LIKE” these (but not exactly and not in any way that can be precisely defined). Those outlines, you could say, stand in relation to the real thing, rather like the outline of a waterdrop or hand a few seconds ago, stands in relation to their outlines now. The brain *demonstrably* works with fluid outlines. Every icon you see: http://www.clipartlab.com/clipart_preview/clipart/icons3-2.gif is evidently not a literal rendering of the outlines of the real objects, but to be interpreted fluidly. So if the conscious brain evidently works with fluid outlines, then the unconscious brain must be able to. But this requires a whole different mentality from the geometric/logical mentality – there, things have to be precise. You can’t understand a point as being loosely round about a given location. You can’t understand a given logical symbol as meaning “loosely something like this object”. If you do all your equations and deductions will be buggered. And if you just llsten to people here, they continually (naturally given their tools) crave precision, single, unambiguous meanings, correct answers. The fluid mentality is: “hang loose, dude; don’t be so uptight; go with the flow” - it’s fluid and adaptable, and continuously changing with unlimited potential to change further and produce multiple-to-infinite versions (within certain constraints).. Algorithms are utterly rigid and haven’t produced and never will a produce a single new element – or new fluid conformation. From: Mike Archbold Sent: Tuesday, October 30, 2012 4:13 PM To: AGI Subject: Re: [agi] The Fundamental Misunderstanding in AGI [was Superficiality] On Tue, Oct 30, 2012 at 6:16 AM, Mike Tintner <[email protected]> wrote: Mike A: All of Mike T's arguments seem to me to stem from a standpoint of extreme empiricism. He doesn't seem to acknowledge anything other than precisely what is under consideration. Even though a chair top can look different in all cases, in all cases there IS a constant, and that is that the essence of a chair persists. Philosophers have long fought with these issues, and as most know it was Kant who came closest (arguably) to reconciling the empiricists and the rationalizers. No I’m not a pure empiricist. (The philosophical/psychological background is loosely important – recent comments seem unaware that this is one of the most controversial areas). The difference is indeed about rationality – about what *kind* of schema/classificatory devices the mind (human or any real world mind) must impose on its images of objects. Rationality – and everyone here, except for me, is in effect a rationalist – presupposes a CONSTANT schema – just as you have said, and just as Plato implied 2,500 years ago. That’s because you are still intellectually living in the age of text, where everything you see is constant and unchanging. You wouldn't even be able to communicate at all if there were no constants. I'm not sure what you by schema in this context but I think you mean some kind of form or set-of-properties relevant to some object or thing. Nobody says you have to have 100% constants. Indeed, that is ridiculous. But, you are arguing using a false dichotomy, it seems to me: either CONSTANTS or FLUID, or roughtly rationalist vs. empiricist. The reality is however that both are needed to process reality, the constant and the changing/unique, and it doesn't matter if we are talking about language, thought, or physical objects. Move into the new millennium of movies, which are now a sine qua non, and you realise that everything is FLUID/MOVING – and different individual versions of things are different from (and in effect fluid versions of) others. There is no constant, essential waterdrop or human being, or chair or apple – especially in a world in which all things may be and usually are transformed by external means in all kinds of way – like being stepped on, smashed, burned or fragmented - if you just look, that lack of a constant is self-evident. But you don’t look – you a priori seek to impose the constant frameworks of language, maths and logic on a fluid world – determined to defend them to the death – despite the fact that they obviously are a complete, never failing to fail, bust for conceptualisation/recognition and anything AGI. For a fluid, transformational world and objects, you need fluid, transformational schemas – but there is nothing in the “languages” you know about them, and you’re not open to new ideas. I get the continuous feeling that you think that just because we express something as an algorithm or in conversational language nothing further can emerge from it.... is that right??? Fluid schemas are doubly essential because – the other thing that all here forget – an AGI of any kind must get to know and classify objects *piecemeal/gradually*, developmentally. The first chair or dog you see may not be at all a typical or common one. All the current approaches to AGI assume a *full knowledge/fully developed mind* - with well structured concept graphs and a fully developed grammar - which has in effect already learned more or less all it really needs to know - quite, quite absurd. Every approach in the field is only appropriate to a fully knowledgeable narrow AI routine/subsystem, not to a real world AGI, complete system gradually, fluidly getting to know the world. AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription 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
