Member 2664
108 entries
457892 views

 RSS
(M)
US
Immortal since Jun 17, 2010
Uplinks: 0, Generation 4
mad-scientist and computer programmer looking for something more interesting than most people accept as their future
  • Affiliated
  •  /  
  • Invited
  •  /  
  • Descended
  • BenRayfield’s favorites
    From AsylumSeaker
    Christopher Langan
    From Yissar
    Technology Progress vs....
    From XiXiDu
    The Nature of Self
    From QESelf
    View Point Room Argument...
    From Jorgen
    My Paper on Computer...
    Recently commented on
    From gamma
    Is brain a computer?
    From BenRayfield
    Elections should be done...
    From BenRayfield
    The most dangerous thing...
    From BenRayfield
    Why is there no Content...
    From BenRayfield
    How can a set of computers...
    BenRayfield’s projects
    Polytopia
    The human species is rapidly and indisputably moving towards the technological singularity. The cadence of the flow of information and innovation in...

    The Total Library
    Text that redefines...

    Start your own revolution
    Catching up with the future. All major institutions in the world today are grappling to come to terms with the internet. The entertainment...

    Proposal for a multimedia...
    A musical mindstorm on the nature of sound, light, space and subjective experience powered by locally produced energy, heralding the ending of the...
    Now playing SpaceCollective
    Where forward thinking terrestrials share ideas and information about the state of the species, their planet and the universe, living the lives of science fiction. Introduction
    Featuring Powers of Ten by Charles and Ray Eames, based on an idea by Kees Boeke.
    http://en.wikipedia.org/wiki/Newcomb%27s_paradox
    Basically, its a question about how to decide what to do of 2 choices, the one that looks more valuable here and now (take whats in both boxes), or the one that looks more valuable because others chose it and it worked out for them even though it usually looked less valuable at the time (take only whats in the first box and dont look in the second).

    The paradox is those whose strategy tells them to not take whats offered also in the second box, in the past for most of them, have found more total value (however they value possible futures) in the first box than those whose strategy is to take both boxes, but how could taking whats in an extra box reduce the total of the 2 boxes if they're already sitting there in front of each player at the time?

    Whatever the cause, if time is not what we think it is, or if however the amounts of value in the boxes is filled is caused by really advanced prediction, or whatever reason it was or is that way, the gametheory of it is the same. The world is what it is, and we must choose at each moment of here and now what to do in a variety of contexts. Once you arrive at a here and now, its too early to figure out what caused it. You can do that in later moments. In a moment itself, you can think about earlier moments, or however time is shaped which is how you arrived at that moment.

    This is about the most basic few things about intelligence, how all possible minds could work.

    If someone is playing basketball in an experiment where a tone is played depending on the height of the basketball in each moment, gradually different tone very accurately as the basketball moves, then all the players who pay attention to the world around them including the tone and basketball (both objects of value, basketball as goal of the game and tone as a good predictor of the basketball) will sync their thoughts with both. If the tone acts differently when a certain player is shooting the ball, as it rises still touching their hand, then their surprise, as in Jeff Hawkins saying "surprise goes up the hierarchy" of patterns about patterns in neocortex, will cause a different movement of the hand either toward or away from what that different tone would be associated with, a different height of the ball. The perceived speeds of time differ between handling the ball and hearing the tone which has been learned to be a good predictor of the basketball. But is it a good predictor if it changes when a certain player shoots but not for anyone else? That may depend on if the changes tend to help that player shoot better or worse or what effect it has.

    Long ago there were experiments to understand the intelligence of mice which were put in mazes to find cheese, and they reached the cheese faster in a majority of the repeats. That kind of thing needs to be continued with experiments based on getting the mouse to think a certain tone which varies by how the mouse moves, is valuable in helping it find the cheese. It could be a sum of tones, each centered on some unique frequency and varying based on how each body part of the mouse moves at that moment, what direction its turned, where it is in the maze, or many possible "feature vectors". Whatever kinds help the mouse find the cheese more often should be seen as valuable to the mouse, depending on if its smart enough to see the connection to the change in tones and its variety of body parts. The more complex the strategy of tones, the less likely the mouse is to see it as more than a random pattern.

    Similar experiments could be done with dolphins, tones representing how parts of their body are curved, and the tones would be adjusted slightly sometimes to influence the dolphin to move those ways or their opposite (depending how the gametheory of it recurses). I expect many dolphins together could find some system of tones useful if they were proportional to acceleration in directions toward or away from specific other dolphins, so they could use gps without understanding the buttons or tech of it, but I'm not sure how it could practically travel with them on a large scale. An advantage of helping many dolphins communicate with eachother through vector derivatives representing any designed feature vectors, is that form of communication is used all over the world by math and prediction software, some parts of prediction markets, and is compatible with brainwaves or at least some simplified simulations of them. I think dolphins and crickets could both benefit from talking to eachother in large groups even if individuals of either are not much intelligent by themself. They are both highly wave based minds, crickets chirping at a frequency of a known equation depending on properties of the weather, but my theory is that is the main purpose but is also a carrier wave for subtle variations of other things they may communicate. Parrots could also work. Many species and software could communicate using the common language of vector derivatives of feature vectors.

    Here's the part about Newcombs Paradox in the real world, using coin flips as an example of any strategy...

    If many people spread evenly across Earth all, relative to time at Earth's center is the best I can describe it, simultaneously throw their own coin high in the air spinning fast, and while all those coins are up each person decides on a strategy for what to do in 2 cases, then the blob of reality we call Earth is entangled together by the weight of those "possible futures" which depend on all the coins which are, before the choice of strategies can reach eachother at lightspeed, physically connected to the whole Earth and therefore to eachother, by the force applied to the wavefunction at the derivative between before and after the players chose their strategies and while the coins land. Before the coins land, but after choosing strategies that depend on all the coins together, light has time to reach from each side of Earth to the others.

    If many of the strategies are what I call "econbits", which is "minority gamble wins", then those players would act based on the belief or agreement that every player's influence on the other players (however much effect they have on the world) increases or decreases by how much of that influence is risked in each moment or round of the game. Many such moments/rounds occur continuously while the first coins thrown all into the air at once are landing. Each round applies physical force, from the derivative of "possible futures" by choosing strategies that depend on combinations of the coin flips, from that physical force onto the world in total. Its a very small force, and the world is much too high torque to push with that small force, but some parts of the world are lower torque, specificly the coins in the air which have much less mass than Earth and have a very high ratio of changing the "possible futures" divided by their mass. For example, you might choose to drive a heavy car one direction or its opposite direction depending on how this "econbits" gametheory works out, and as soon as that "possible future" becomes a statistic connected to the coins in the air, the weight of those possible futures puts force on the coins. Its a very low signal to noise ratio since they are spinning, but force has to go somewhere, and the spinning is connected to the landing through the eyes of the person who flipped it and will see it land and transfer force to that "possible future" by acting to get there.

    This is about variations in delay in prediction markets, viewing all of gametheory as a prediction market where the currency is influence on the world. In econbits theory, trying to win is how you get others to predict your coin, to learn its patterns and balance their strategies so its about half and half chance each way. Trying to lose is how you train the others to predict any data which you put into the strategies of the others, at a cost of influence to you because losing is a mistake and you're trying to lose at that moment (which some would call market manipulation, but I offer it as a simplified model of intelligence and prediction, not necessarily to be used with those highly regulated and excessively complex kind of predictions). Instead of "market manipulation", in econbits theory, trying to lose is a trade of some of your influence for others learning any data you choose (scalar or bit vectors) which, if they are trained often enough and on enough examples, they would respond about that data like an associative memory (especially boltzmann machine math, or whatever model of intelligence) because any strategy which does not learn from the other players "mistakes" where their influence is reduced (by definition, thats what a majority of players in econbits have agreed to try to converge toward) would, if the other players are effective, lose influence for not learning from those mistakes of others. Buy low, sell high, or short high, and sell low. When your coin lands a way that less than half the coins landed, thats a win for you and influence is paid from those whose coin landed all the same way (majority gamble loses). Econbits theory has "minority gamble wins" at the core because it has no exact Nash Equilibrium therefore requires prediction of prediction of prediction... to any depth, while it has Nash Equilibrium in the limit of many players and the amounts risked on gambleUp vs gambleDown converge to near equal except a tiny friction between them potentially only 1 unit of influence different between them but on average may be squareroot of the amounts which continue to be risked. The nature of prediction markets is Nash Equilibrium would prevent any future price change even if the world changed because the balance between trades would lose its tension and therefore its value as a predictor of future value, so when the future comes other things are better predictors, including potentially other games of coin flips or other models and strategies.

    Marketforce is directly interchangible with physical force without going through a power station. Marketforce is influence, and influence is entanglement.

    There are risks to accessing intelligence of many people, crickets, parrots, dolphins, and thinking software, through the common language of vector derivatives as each a varying tone or echo or delay, but if intelligence is outlawed then only outlaws will be smartest, and I say outlawed and regulated often interchangibly because of what I explained about delay applying marketforce. If marketforce is applied against intelligence in general, that would be a much bigger risk and create many problems for everyone, basically bookburning in electronic and other advanced forms, keeping the books around so it wouldnt be called that. Evolution is proceeding. The world is waking up, minds seeing eachother and asking what is a mind and how are they built and what to use them for?

    Nearly everything as its done today could be improved in some way, explore variations, see whats possible and safe and decide together what kind of futures we want to build. A basic example is, since the tones could adjust the angle a basketball is thrown by changing tone to more or less surprise (higher and lower both surprise, center is near Nash Equilibrium), then a doctor performing surgery could have practiced with such tones as predictors of his hand and tool positions so, with the help of computers adjusting the tones many times per second, he could cut into moving organs maybe as well as the machines which are designed to move up and down with the organs being cut as if they were not moving relatively. Computers and people can both use math to influence movement and thought, in a balance I think would be described by econbits theory about how the tones or echos or delays are adjusted by recursive gametheory. There are far more advanced things to consider longterm.

    A consistent gametheory understanding of Newcombs Paradox would allow the mining of "waste heat" aka entropy into usable form across the wavefunction which all our strategies are part of.

    My answer to Newcombs Paradox is:
    In econbits theory, trying to lose is paying your risked influence to train data into the other players strategies, and opposite of that, trying to win influences them to predict you because you are trying to predict them and so on to any depth. The choice at each moment to one-box or two-box is a recursive gametheory potentially without limit in a network with cycles (everyone to everyone) because there is value in equal amounts of being predicted (as the value of data pushed in by payment of the "majority gamble loses" that round) and giving others data you want them to learn and predict based on later. Its a mobius between "buy low sell high" and "short high unshort low", without direct ability to choose buy or short because they are both the same coin which you either want to receive influence by being in minority gamble (less coins landed like yours) or pay influence by being in majority gamble (slightly more coins landed like yours). Econbits theory is about a free market which exchanges prediction and influence as equal and opposite force. One-boxing and two-boxing are like read and write in an AI system, but that doesnt mean we know which is the read and which is the write. For that, we would have to predict recursively which is the source of the currency called influence which is traded opposite the flow of data of how any coins or bits land. Eliezer Yudkowsky wrote an important paper Timeless Decision Theory which would be among the next steps in understanding these things better. Also simulating the tones and getting prototypes of the game, AI, and science network up eventually.
    Tue, Feb 10, 2015  Permanent link
    Categories: gametheory, newcombsparadox, econbits, dolphin, cricket, parrot
    Sent to project: Polytopia
      RSS for this post
      Promote
      
      Add to favorites
    Create synapse
     
          Cancel