Is it possibly to functionally transfer knowledge from a neural network to another?

Discussion in 'Intelligence & Machines' started by Buckaroo Banzai, Jan 3, 2018.

  1. someguy1 Registered Senior Member

    Messages:
    727
    Because the people who study computations don't know how to instantiate any other type of computation in the physical world.

    So that any OTHER type of computation may well exist; but it has not yet been discovered. You're just talking about a hypothetical future leap in technology; but you THINK you are saying something much more tangible than that. You're not. You're making an unjustified leap. The science isn't there. Yes you can CALL the secret sauce of consciousness a "computation." And you can say that you mean it in a sense more general than just a TM. The problem is -- you haven't got such a thing. Nobody does.

    The point isn't to deny there could be a mode of computation that's not a Turing machine. It's that nobody's ever found an example of such.

    The rest of your post was interesting. I would love to respond point by point. But having reread my own previous post, I think it came out pretty well and I'm content to let that post stand as my thesis. With more words I could only repeat what I've said.

    I'm going to take a break from this very interesting thread for a while. The conversation has helped me to clarify some of my own thoughts about these matters. Thanks to everyone who chatted. This has been a thoughtful thread.
     
    Last edited: Jan 22, 2018
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. someguy1 Registered Senior Member

    Messages:
    727
    There is no known model of computation that goes beyond what a TM can do. This is the Church-Turing thesis. The theoretical models that go beyond, such as oracle machines, involve transfinite principles that can not be instantiated in the real world using known physics.

    Neural nets do not go beyond what TMs can do. They just do the same things differently.

    You want to imagine "levels" of computation ... but we don't know any that apply to the real world. If you assume transfinite ordinals you can get an endless hierarchy of levels of computation. In fact Turing studied ordinal models of computation in his doctoral thesis, before publishing his paper on Turing machines.

    The problem is that all these models depend on transfinite ordinals. and there are no transfinite ordinals in physics. Whether that holds true in the future, no one can say.

    There ARE NO LEVELS of computation in the real world that we know of. Nobody's figured out yet what that could possibly look like.

    Not neural nets, not quantum computers. Neither of those approaches allows you to compute anything that a TM can't. This is my strong understanding and will remain so till someone shows me otherwise.
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. Write4U Valued Senior Member

    Messages:
    20,069
    OK, question:
    Can we introduce bio-chemistry into computers?
     
    Last edited: Jan 22, 2018
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. Michael 345 New year. PRESENT is 72 years oldl Valued Senior Member

    Messages:
    13,077
    I might be way way way off here but what would you think of the following?

    Our brain is a computer (biological composition) (operating system, content - completely immaterial)

    Stay with me

    We are going to use this computer to teach another computer. We know this computer has the same (biological composition) and suspect the operating system could be the same

    Get where I'm going

    Human teaching ape

    Psychologist Francine "Penny" Patterson teaching Koko ape

    Can we consider Koko brain the biological computer?

    I understand a lot of the teaching took place from what we look at as plain computers

    Human (brain) to plain computer to Koko (brain)

    Problem what stops the human brain getting all of its content into Koko brain?

    So would wiring up Koko brain to either plain computer or even Penny's brain enable a more efficient upload / download ie more information into Koko's brain

    Would the increased information be able to be used by Koko's brain?

    Not sure if any of my thought bubble helps but.......

    Please Register or Log in to view the hidden image!



    Please Register or Log in to view the hidden image!

     
  8. billvon Valued Senior Member

    Messages:
    21,635
    Agreed. Neural networks are not traditional Turing machines, but a Turing machine can simulate a neural network (albeit slowly.)
    We know that simulation can accurately simulate the real world, with sufficient fidelity to predict the outcome of experiments in that real world. As computational power has increased, the complexity of systems that can be simulated has increased as well.
     
  9. Write4U Valued Senior Member

    Messages:
    20,069
    I agree, if it takes a human brain about roughly 20 -30 years to fully "mature", then comparing a "learning' computer to a brain should involve an actual teaching period also.
    What struck my interest is that the brain is not a single lump of neurons that do stuff, but that there are distinct areas which process specific types of information, IOW, the human brain is a compound structure, with many specialized areas for processing specific types of information.

    and the interesting part is that each of these area has a connection to the frontal lobe which is where sentience (decicion making) resides.
    https://www.psychologytoday.com/blog/eyes-the-brain/201108/when-is-the-brain-fully-mature
     
  10. iceaura Valued Senior Member

    Messages:
    30,994
    I said it was describable - formulatable - in propositional logic.
    And you never did actually explain why you thought that was not the case - your discussion veered into digital approximations of unsolvable equations, and did not return.
    I'm not setting it aside, I'm pointing to some of the aspects involved that need to be considered. The brain as a neural net is self-remodeling - that is duplicable, in principle, and more or less takes care of your objection that the brain is not a box of hardware running an algorithm.
    No, I (not Write4U) asserted the opposite and more relevant claim: that a machine can be designed that flies like a bird. That the functionality of bird flight can be transferred to a machine, and imho probably one managed by a neural net.
    I'm normally arguing that anyone who thinks the functionality of a given human brain can be transferred to a computer of some kind any time soon is probably underestimating the task. AlphaGo or no AlphaGo.
     
  11. iceaura Valued Senior Member

    Messages:
    30,994
    They will be "deterministic" only in the loose sense in which predictability and repeatability are not implied (chaos, quantum uncertainty), as the brain is. They will be programming themselves, and probably remodeling their hardware connections themselves. My guess is they won't be all that easy to "understand". Right now, the programmers of AlphaGo have trouble understanding its moves - and uncertainty attends their deductions, when they have made them. And AlphaGo is quite simple compared to anything approaching a full emulation of function.
     
  12. Write4U Valued Senior Member

    Messages:
    20,069
    Just for the record, I agree with you. Moreover I gave an example of the frigate bird wich uses the principle of "lift" with exquisite conservation of energy. It is more of a glider using air currents (lift) than a power driven airplane which requires "thrust" to produce lift, because of its weight.

    But according to the aerodynamics of flight, what difference does the propulsion method make when using wings to create lift?

    When sailing my sailboat I use my sails (vertical wings) to create horizontal "lift" to tack "upwind". The airstream (wind) itself is the propulsion force. In the absence of wind I end up in the "doldrums", unless I use my outboard motor.
     
    Last edited: Jan 23, 2018

Share This Page