Alan, The (nauseating) meat is that given knowledge, structure follows. There exists a unique map from knowledge to structure and viceversa. Every time you mention knowledge, or information, I immediately switch that to structure. You don't, which is where communication stops.
It is tough, I know. If you are still interested, don't loose patiente just yet. I always randomize the input stream coming from a camera (or from anything else for that matter), in every paper I published. The results will remain the same, and my intent is to impress that fact on the reader. But here's the meat: I randomize the causal relations, NOT the pixels. The causal relations carry facts such as: - cone 0 gives rise to signal 27 - cone 0 is adjacent to its neighbor 1 - cone 0 is adjacent to its neighbor 2 - cone 0 is adjacent to its neighbor 3 Now I am going to randomize the causal relations: - cone 0 is adjacent to its neighbor 3 - cone 0 gives rise to a signal 27 - cone 0 is adjacent to its neighbor 2 - cone 0 is adjacent to its neighbor 1 Spatial information is preserved, and the structures remain the same. Hope this helps you to see more clearly that I am not such an idiot. This is not very different from what image recognition experts usually do. They start (I think, you may know better) from an image, a set of points of color with coordinates, and try to associate points themselves, and run into a geometric nightmare. I have the relationships fixed from the start. How am I doing? Sergio -----Original Message----- From: Alan Grimes [mailto:[email protected]] Sent: Friday, July 13, 2012 6:38 PM To: AGI Subject: Re: [agi] Emergent "Inference"? Sergio Pissanetzky wrote: > This does not mean that inference "knows" everything. Inference works > from knowledge, just as the child does. The child acquires that > knowledge by learning from his "sensors", mostly vision and > sensory-motor nerves in this case, and using his inference to derive meaning from the observation. > Eventually, he will "know" what a peg is, and how to recognize one, > and how to know if it is square or round and match it with a hole, and > how to control his muscles to do all that. The inference does all that > by finding associations. The child does not need a programmer to do all that. > With the inference and the computer, it is the same. The robot will > need a camera and a mechanical arm with position sensors besides the > inference. But it will NOT need a program. It will have to learn step > by step, from its sensors. Knowledge is still necessary, but it comes > as input, not as program. Naturally. But this is an excessively general answer which completely ducks the meat of the question. The problem was specified to emphasize the necessity of processing spatial relationships. Because you're a crackpot, you are unwilling/unable to see the deficiencies of your theory. Your theory does not process spatial relationships and it leaves no room for that flaw to be patched therefore I have to discard it or make radical, unauthorized modifications to it. I'm a visual thinker and I know for certain that spatial information is critically important. > In the case of the retina, the situation is a little different. You > may have seen my recent post about the blind climber who can see > enough to climb a mountain with a camera and electrodes attached to > his togue. There is no retina, no optical nerve, not even a > vision-specific area of the brain involved there. This confirms what I > already knew from my experience with causal sets. The anatomical > details about the retina or the optical nerves, or left-right and upside down, are not needed at all. Not even as input. =\ You've been repeating this ad-nausium. My counter claim is that the post-central gyrus, where most of the nerves from the tongue end up, has a number of features in common with with the occipital lobe. My claim is that while you are trumpeting the plasticity of the brain as proof that nothing matters, my counter-claim is that the same evidence is a demonstration of my point that: A. the cortex is general. B. Spatial maps are what is important, not a specific neural pathway. C. Your system does not seem to have a way to encode spatial information therefore it sucks. Get some 3D goggles, set it up with a camera and a permutator which randomizes the input stream (maps one pixel to some random pixel), then try to see with it. If you do manage to adapt to such a thing, it only means that your brain re-organized your optic nerve to restore the spatial mapping of the outside world, and that adapting back to normal will be almost as difficult. This is neural science 101. I'm sure it would be easier for the nerves to drop in wherever, but that's not what we see, we see an exquisitely ordered and highly regular pattern of neural connections at all levels, and for pretty much all senses, except the olfactory sense. > We would be living in a fantasy world if we believed that anyone can > understand or explain or prove or guarantee all that. I sure can't. > Because of that, I have proposed a practical approach. First, before > even starting anything, we need a computer with the inference > installed on it. Second, a simple model of a retina, just a camera > with a few hundred pixels, followed with the inference. Show it an > image, see what it does. Does it compress the image as the retina > does? By how much? Compare with the real retina. If a match is found, bingo! I am sure it will. The retina is extremely lossy. It is also well understood. It is not very interesting either except in showing how little input the brain actually receives. -- E T F N H E D E D Powers are not rights. ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/18883996-f0d58d57 Modify Your Subscription: https://www.listbox.com/member/?&; d2 Powered by Listbox: http://www.listbox.com ------------------------------------------- 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
