Comment Re:I say it all the time, Vision to 3d world = AI (Score 1) 78
Interesting. I never heard of SHRDLU before, but I have heard of Cyc wikipedia said came after. From what I've heard of CYC, CYC only takes words in relation to each other, but does not have a physics space in which to have "imagination space". Because of this, CYC doesn't sound like it should ever achieve AI.
However SHRDLU does sound promising. I would think if they enhanced the physics and vocabulary on SHRDLU, it could become "imagination space" for future AI. I knew this part could be created independently of "vision recognition" part, but I've felt that it'd be better if the vision recognition was done first. After all, if you're databasing objects on their real world dimensions, they'd work better in the physics imagination space because you have the exact dimensions and contours of the object. I mean, if you're gonna sit there and database in many different objects based on their geometry, might as well have it done automatically. Of course I wasn't thinking to start small with basic objects in a solid physics engine.
Interestingly enough: I think this is an area which advances could be made now. I wouldn't exactly start with SHRDLU because it is early 1970s technology. I'd shop around for good physics engines first. I have no money to buy physics engines, so I'd personally have to go with the best free one... So maybe I'm incapable of doing this research due to lack of funding. Anyway, someone else can do it: Find a good physics engine, then start databasing "Nouns" "Adjectives" "Verbs" and "Adverbs" maybe looking at initial SHRDLU code just for advice. The thing that makes me interested now is that I realize that many objects are not exactly the same objects: Trees are trees, but even the same variety of tree has different growth stages and limb orientations. So for generalization terms, it could be good to look at things like "Spheres" "Cubes" "Pyramids" and so on.
Thank you for putting me onto this. I think it is something that could be solved now with enough time and energy. The reason I put it off was because I made the assumption that the 3d digitizer could double as an "Object databaser" for inputting objects into the database. I originally didn't want to teach the AI about a world of objects that bare no relation to exact real world objects, but it is fine to have approximate objects and approximate physics to start with.
However SHRDLU does sound promising. I would think if they enhanced the physics and vocabulary on SHRDLU, it could become "imagination space" for future AI. I knew this part could be created independently of "vision recognition" part, but I've felt that it'd be better if the vision recognition was done first. After all, if you're databasing objects on their real world dimensions, they'd work better in the physics imagination space because you have the exact dimensions and contours of the object. I mean, if you're gonna sit there and database in many different objects based on their geometry, might as well have it done automatically. Of course I wasn't thinking to start small with basic objects in a solid physics engine.
Interestingly enough: I think this is an area which advances could be made now. I wouldn't exactly start with SHRDLU because it is early 1970s technology. I'd shop around for good physics engines first. I have no money to buy physics engines, so I'd personally have to go with the best free one... So maybe I'm incapable of doing this research due to lack of funding. Anyway, someone else can do it: Find a good physics engine, then start databasing "Nouns" "Adjectives" "Verbs" and "Adverbs" maybe looking at initial SHRDLU code just for advice. The thing that makes me interested now is that I realize that many objects are not exactly the same objects: Trees are trees, but even the same variety of tree has different growth stages and limb orientations. So for generalization terms, it could be good to look at things like "Spheres" "Cubes" "Pyramids" and so on.
Thank you for putting me onto this. I think it is something that could be solved now with enough time and energy. The reason I put it off was because I made the assumption that the 3d digitizer could double as an "Object databaser" for inputting objects into the database. I originally didn't want to teach the AI about a world of objects that bare no relation to exact real world objects, but it is fine to have approximate objects and approximate physics to start with.