Anastasios, Yes, our answers differ. But this is not a surprise, at least not for me. One should expect different conclusions from different theories. And it is also true that a theory must be verified by comparing predictions with experimental observations. One single experiment that contradicts the theory can do away with the entire theory, if the experiment is properly verified and reproduced.
Yes, this is a revolutionary theory. But it came from a single discovery, the discovery of self-organization in causal sets (more precisely in canonical matrices, but they are equivalent). But no, it is not an algorithm, it is an observation, experience that I gained. I did follow the usual algorithm for experimental Physics, which is to pose a question and search for an answer by experiment, and be capable of recognizing the answer when you see it. These are my only two claims: I knew where to search, and I knew how to figure out what exactly I had found once I had found it. I am nothing but an average scientist. As you correctly say, discovery must be reproducible. This one is reproducible, anyone can do it. There must be an independent verification, preferably on a scale much larger than mine. It is real easy to do, as compared for example with the monumental size of image recognition work. It is not something you can do in an afternoon, of course, it requires someone who will learn enough of the theory and make it run. Besides, there is only one program, the one for I/O and the minimization of the functional. It is the same for all problems, so it has to be written only once. I am not into proto-scientists. But the thing is, the excution time is roughly constant with size IF the size does not exceed the size of the hardware. For a (very) crude example, think of a neural network with 1M neurons, one per each pixel in a 1M pixel camera, plus a PC to feed the data into the network. The neural network (not the usual type) the only think it does is to minimize the functional. An improvement on that, would be a chip with 1M microcomputers that can be programmed. I think they are getting close to that. It can be done in a year or so (don't believe me, I am terrible as a manager). And if done, will it not attract enough attention to do the next step, say 10M, much quicker? And then 100M. I am not planning on doing any of that myself. I have to continue developing the theory, at least for now. Sergio -----Original Message----- From: Anastasios Tsiolakidis [mailto:[email protected]] Sent: Sunday, July 22, 2012 12:44 PM To: AGI Subject: Re: [agi] Re: How the Brain Works -- new H+ magazine article, by me On Sun, Jul 22, 2012 at 6:12 PM, Sergio Pissanetzky <[email protected]> wrote: There is no geometry, > addition, multiplication, axes, planes, rotations, etc. All that has > to be learned just like you and I have myself written a couple of paragraphs on the many possible starting points of an intelligence architecture, with the most agnostic ones being something akin to "total synthesis" in chemistry, for example to start from 5 elements and end up with complex proteins. The lower the starting point, the more it becomes like trying to create a human being from charcoal and water. On the other hand, we do have examples of humans who managed to understand our world, develop language etc, while missing 99% of the datastream average people have (deaf and blind babies) - their effort is very much a total synthesis, so it can be done if you have a brain. Can it be done if you have a pentium? It looks like our answers differ. Of course the world does not have "geometry". Geometry is a theory that helped Euclid, Archimedes and people like him to find how much paint and wood they need to build a house or a boat. So interaction with the world was necessary but not sufficient, otherwise my grandma would be drawing isosceles triangles. But the discovery of knowledge is only verified as knowledge by repetition both personally and socially, it really depends on the majority's ability to follow an algorithm and reach the same results as you. Then every now and then appears someone who follows the algorithm and reaches new results, or who modifies the algorithm - deciding who is in error and who is a revolutionary scientist is intractable, probably also random and incomprehensible. Flat earth etc. I am assuming you are building a proto-scientist. There is a way to drop it in the deep end of scientific endeavour: you can turn it loose on a major website, ideally with a lot of non-expiring content and a lot of updates, like news.bbc.co.uk . That would be the input. Sadly for output you have to limit yourself to occasional "visitors comments" by your system, and see how long it takes for a reasonable comment. On a Pentium LOL. Or your system may end up hacking the BBC, by submitting poisoned SQL and HTTP queries and controlling the free world, you never know. But can you find the "right" invariants without interaction, only by observation? A bit like our friend Matt wants to understand the world by compressing audio, video and text? I think it can't be done, and if it can it will take an extra 2 billion years (on an Itanium!). AT ------------------------------------------- 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/?&; 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
