JIM> I am not able to understand why this is truly relevant to solving the contemporary problem of AGI. In my opinion, massive interrelations seem like a necessity and combinatorial complexity the problem.
SERGIO> In the post that follows, I try to explain where that complexity comes from, how it relates to AGI, and how it can be dealt with. I also explain how, the more efforts are directed towards "explaining" that complexity or those interrelations, the more the complexity grows and the more uncertain the interrelations become. The process of dealing with complexity by programming computers, does not converge. It is a long post, but it still falls short of answering your question in full. That, I will do in another post, as I anticipate that the present one will give rise to some discussion. Here is the new post. PRESENTATION OF 08-18-2012. In a political campaign, one candidate accuses the other of something terrible. Depending on which side you are, you may react like, "At last! He smashed him!" or "He has no shame! Such dirt!" The entire country is now living in two very different universes, one where the accused is finally destroyed, the other where he is not. But within hours, the offended candidate answers the accusation. He survives. He has found an invariant behavior, one that does not change under a transformation from one universe to another. Don't we always carry a cat in a box? The last thing I said in this sequence of presentations, is: "Since you are using your brain to do all that reasoning, I must conclude that there must be something in your brain that allows you to do that. My attention, therefore, immediately switches away from the reasoning itself, and towards 'that' which is in your brain and allows you to reason that way." This statement puts me in a position where I want to find 'that' what exists in brains and does the reasoning, but I have promised not to reason myself. For, if I were to reason about 'that', then the same argument would apply to my own reasoning. So I can't reason. But I can search, or observe. Of course reasoning can help to search better, but the essence of searching as opposed to reasoning is that you find something you can't explain at the time. You walk in the woods and, unexpectedly, run into a treasure. That's a discovery. Later, you combine the discovery with additional knowledge, such as "the pirates had been in the area" and write a book explaining how the pirates visited the area and hid a treasure and how you found it. Next, I will explain my search, and findings, in the context of entropy. At the time of the search, and the findings, I had no idea that entropy had anything to do with it. For the presentation, it is better to start with entropy. I want to bring to issue a few well-known concepts and their not-so-well-known relationships, assuming little background from the reader. Entropy has vast connotations, but my presentation is restricted to the context of AGI only. ENTROPY AND INFORMATION The first lesson to be learned about information is that information is a property of a physical system. Information does not exist by itself. There is always a physical system or media that carries it. It can be an optical disk, a computer's memory, a brain's memory. Information can travel. A file being copied to a computer, a fiber-optic cable that carries television signals, a beam of light coming from a star and carrying with it the history of that star, which astronomers can decode. But even when it travels, there is always something physical that carries it, a neuron, a bit of memory, an electron travelling in a cable, a beam of light. I like to think of information as a *modulation* of a physical system. Information has energy of its own. The energy in information was directly measured in an experiment conducted only 5 months ago, in March 2012. Information is measured in bits, energy is measured in Joules, and the energy of one bit of information is 3 x 10^-21 Joules. If information travels, the energy goes with it. If it is stored on a media, the energy is in the media. If you learn something or a computer learns something, the energy is in your brain or in the computer. There is a law of nature that energy can not change, it can only flow from one place to another, and information is a flow of energy. Entropy is frequently introduced as a measure of uncertainty in information. But entropy is also a physical quantity, a property of the state of a physical system. Any physical system has entropy, and given the state of that system, the entropy of the system can be calculated as a unique function of that state. Entropy too can travel, meaning it can flow from one place to another. There is also a law of nature for entropy, but unlike the law for energy which bans change, the one for entropy bans only decrease. The entropy of a system can not decrease. Entropy can flow from one system to another, and it can not disappear, but it is perfectly possible for the entropy of a system to increase on its own, even if there is no flow of entropy entering that system. The entropy of a system can increase as the result of a process of acquiring information, for example, learning. The information carries energy, the influx of energy causes an increase in the entropy of the system, and the system's uncertainty increases accordingly. Consider now a description of the system in terms of variables and states (see my earlier post). When energy flows in, the result is an increase in the number of states and in the number of possible transitions from one state to another. The increase in the number of transitions carries with it an increase in the uncertainty about which transitions will actually take place, (there are more possibilities), and hence the increase in entropy. This is a thermodynamic process, and can not be avoided. Consider Douglas Hofstadter looking at his mother. His brain has just received signals coming from 100,000,000 dots of light on his retinae. With the information, also came additional energy, additional entropy, and additional uncertainty. Hofstadter's brain now has more information, more energy, more entropy, and more uncertainty. Hofstadter is now more ignorant and more confused than he was just before he saw his mother. Think of a developer programming a computer. Perhaps a carefully designed AGI machine. Of course, the computer with the program is a physical system. As the developer writes more and more program into it, and makes the computer "better informed", she is also making it more confused (so much for the nieceties of our language), and it becomes for her more and more difficult to "understand" the program. This process is very well known in software development, I am not making it up. It constitutes a hard limit that can't be overcome. There is an easy solution known as refactoring. It is a process where a human developer extracts entropy from the computer, making it less uncertain, less confused, and more understandable. The problem: a human is required to do that. A machine can't do it. If an AGI machine is supposed to have, one day, a human level of intelligence, then it should be able to do by itself the same that the human developer does, namely, to extract entropy from its own program. Currently, I don't know of any efforts, but mine, directed to systematically extract entropy from computer programs by machine. Leaving the entropy in the program results in a confused, uncertain machine, incapable of finding an invariant behavior as the owner of the cat in the box did. Certainly not an AGI machine. Sergio From: Jim Bromer [mailto:[email protected]] Sent: Friday, August 17, 2012 12:27 PM To: AGI Subject: Re: [agi] Uncertainty, causality, entropy, self-organization, and Schroedinger's cat. Sergio: And no, there is no limit on time of execution either. This is still unpublished, so I can only give you a hint. Assuming a neural-network computer simulation where each element of the causal set is represented by exactly one individual neuron, and assuming near-neighbor coupling, the time of execution is constant and independent of size. This is massive parallelism. I am just curious, are you still with me? May I ask a quiz to verify? You don't have to answer, the answer is below. Aside from the obvious fact that this is going to be fast, what is the real, profound significance of this result? Jim: I am not able to understand why this is truly relevant to solving the contemporary problem of AGI. In my opinion, massive interrelations seem like a necessity and combinatorial complexity the problem. Jim Bromer AGI | <https://www.listbox.com/member/archive/303/=now> Archives <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> | Modify Your Subscription <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> AGI | Archives | Modify Your Subscription <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> ------------------------------------------- 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
