I don’t believe robots are going to watch vids as we do. Understanding the syntax of human level vision is every bit as complicated as - and utterly interrelated with - understanding human language. Any idea that any machine is going to do either in other than an extremely long time is fantasy through and through – and involves an extreme failure to understand the nature of both. (I’ve no idea for example where Stephen Reed has got in the last years – but I can safely predict “nowhere AGI-wise”)
The importance for AGI of recognizing that image schemas structure all action is that they have PRACTICAL VALUE – they *can* be introduced and used to control simple actions on simple objects by simple robots in simple environments – which is where practical, realistic AGI will begin – e.g. industrial and domestic robots. How can fluid outlines be introduced into current computer/robotic systems? By instructing robots to “manoeuver” – to “do something ALONG THESE LINES” – rather than to do something along a specific line – iow to explore the range of lines of movements of the robot’s hands and other limbs in a given situation – explore, for example, the range of their grips and handling, or the range of their possible treads and steps. Image schemas serve as loose GUIDELINES FOR ACTION- and from guiding simple actions they can go on to guide extremely complex courses of action. A simple arrow can serve as an instruction to reach out for an object – a very simple journey. . It can and does also serve as an instruction to go from London to New York – a representation of a very complex journey. From: Jim Bromer Sent: Monday, July 23, 2012 10:51 PM To: AGI Subject: Re: [agi] Image schemas control all forms of action [Lakoff replies] But, one of the things that you (Mike) haven't yet understood is that deriving usable information for an AGI program from video is just as difficult as deriving it from text or mathematics or a IO system that "rewards" a program, and so on. If effective intelligence was as simple as watching a movie and associating some text or a phrase with an action that was recognized in the video then AGI would already be a done deal. The problem is figuring out a way to get a computer to do complicated things like that reliably. There are some data objects in most IO media that are easy to discern. Some of them (not all of them) can sometimes (not all the time) be used to reliably indicate that a 'conceptualizable' action is taking place (or a 'conceptualizable' object is in a scene) but those moments of clarity are not dense enough to build a solid foundation for higher AGI. Jim Bromer On Mon, Jul 23, 2012 at 5:17 PM, Mike Tintner <[email protected]> wrote: From: George Lakoff Sent: Monday, July 23, 2012 10:11 PM To: Mike Tintner Subject: Re: [Cogling-L] The scope of image schemas Narayanan's X-schemas (or process schemas) characterize all events and actions and actually control physical actions. So you're right about that. We are now working on entity schemas, but we're not there yet. George On Sun, Jul 22, 2012 at 11:34 AM, Mike Tintner <[email protected]> wrote: Lakoff:The idea behind image metaphors is simple. Images are structured by image schemas. A given image has multiple image schemas linked via neural binding to form a composite image schema ? or more than one. Metaphors map one image to another by mapping the source image schemas to the identical image schemas in the target George, Your exposition was v. useful. Can you/should you not extend the scope of image schemas? They structure presumably under *Images* : both *Verbal Images* & *Graphic/Photographic/Sensory Images*. and not just word images but : *Words/Language/Concepts" - period; *all words* are structured by image schemas, no? And from that one can one go on to argue - no? - that they structure *Moves/Movement* - period - that, for example, our reaching for a cup is structured by a schema. After all, language is used principally to structure actions: "Hand me that cup" - "Go to the other room". It makes sense that image schemas should structure not just verbally-mediated action, but all action, however mediated. The same mirror neurons that respond to (image-schema-structured) verbal accounts of action, also respond when just watching direct sensory images of agents executing those actions. Concepts/schemas arguably structure all the actions of living creatures. Comments? P.S. Personally, I think it's helpful to think of image schemas as "[loose] outlines" - esp. in connection with actions. Comments? 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
