Eureka Machine Discovers Newton's Laws in 2 Hours

Discussion in 'Intelligence & Machines' started by madanthonywayne, Jun 15, 2010.

  1. madanthonywayne Morning in America Registered Senior Member

    Messages:
    12,461
    Scientists have invented a device they refer to as the "Eureka Machine". It is a computer that sifts thru tons of data looking for underlying patterns. It was assigned to study the motion of a dual pendulum and, within a few hours, came up with Newton's laws of motion.

    It was later given data on yeast cells and came up with a bunch of equations that the Scientists don't yet understand. The equations work, they predict things like what happens to protein X when enzyme A increases. But the Scientists don't understand why they work or what they really mean.

    This marks the first time a machine has independently made scientific discoveries. It's pretty cool, but also a little scary. It reminds me of something Agent Smith said:
    Which is why the Matrix was redesigned to this, the peak of your civilization. I say your civilization because as soon as we started thinking for you it really became our civilization which is of course what this is all about. Evolution, Morpheus, evolution. Like the dinosaur. Look out that window. You had your time. The future is our world, Morpheus. The future is our time.

    Is this the begining of the singularity?

    http://www.guardian.co.uk/science/2009/apr/02/eureka-laws-nature-artificial-intelligence-ai
     
  2. Google AdSense Guest Advertisement



    to hide all adverts.
  3. soullust Registered Senior Member

    Messages:
    1,380
    Hey we may have given birth to the next dominant life form on earth.
     
  4. Google AdSense Guest Advertisement



    to hide all adverts.
  5. Michael 歌舞伎 Valued Senior Member

    Messages:
    20,285
    *beep beep*

    I watched the video and it does look pretty impressive at first glance. I'm not sure what to make of it really.
     
  6. Google AdSense Guest Advertisement



    to hide all adverts.
  7. Pandaemoni Valued Senior Member

    Messages:
    3,634
    It doesn't make me apprehensive. If there is one thing computers are better at than people, it's sifting data and finding correlations. The thing is, scientists don't know why the computer's yeast equations work, but neither does the computer, it just saw a correlation in the data and worked out a precise formula that matched it.

    If the computer had worked out the relationships *and* the mechanisms undelying it, that would be different (and I tend to assume that day is coming). Perhaps then I will be apprehensive. As a fictional computer once said:

     
  8. DNA100 Registered Senior Member

    Messages:
    259
    Impressive Pattern Recognition!

    But like Pandaemoni pointed out,it doesn't quite find out the underlying mechanisms.
     
  9. Stryder Keeper of "good" ideas. Valued Senior Member

    Messages:
    13,105
    As Lorenz found with Weather prediction system, sometimes human error can account for the most dramatic of chaotic results. If the Pattern Recognition systems isn't fed the right data to begin with, then it will plot abstract erroneous output.

    Sometimes this might be that the wrong data is entered, other times it will be down to data not existing to begin with. (When that data doesn't exist, you're left with a fuzzy logic created variable that you have to then try and make sense of.)

    What I would find of personal interest however is if the system Cornell has been working on is applied Transhumanly (Namely apply it as Cybernetic Mnemonics) You could have apply the knowledge of a human brain to the workings of an AI, especially important for structuring various scientists that can occasionally have issues explaining themselves or expressing themselves in a format that the consensus can interpret.
     
  10. Search & Destroy Take one bite at a time Moderator

    Messages:
    1,467
    Awesome thread.

    It makes me wonder if I can hook up 20 PS3s on a super computing parallel network, install a neural network, and generate a profitable profit-seeking model for the financial markets...
     
  11. Stryder Keeper of "good" ideas. Valued Senior Member

    Messages:
    13,105
    Thats kind of the reason for the current economic downturn, as if you end up with a programmatic formula, you end up with something that can be exploited.
     
  12. Search & Destroy Take one bite at a time Moderator

    Messages:
    1,467
    What % of the troubled economy to do you pin on speculators?
     
  13. Stryder Keeper of "good" ideas. Valued Senior Member

    Messages:
    13,105
    Initially it was a rather large percentage, this was proven in regards to banks borrowing and lending. With the financial turmoil caused by Iceland in the 3rd or 4th quarter of last year a number of banks collapsed causing a number of countries to look very closely at how banks are moderated in regards to borrowing and lending.

    There were a few other events that triggered an increase in Speculators, for instance the US having a number of states banning Online Gambling. In those states the only legal online gambling left is of course the stock market, which wouldn't be much of a gamble if it was all Arbitrarily controlled by robotic trading. (Robotic Arbitration would cause steady but slow growth, as opposed to a flash in the pan event with a sudden growth)

    The problem with using a Robotic Arbitration method however was that robotraders work to dynamic programming, in various events it's possible to cause the robotrader to react worse than a panicking investor. This applied to potentially hundreds of thousands of robotraders would in turn cause an entire market panic artificially.

    This is one of the main reasons I see the entire economic situation as being fraudulently conceived and extremely misconstrued.
     
  14. kurros Registered Senior Member

    Messages:
    793
    This a pretty cool little program they've made, but hardly anything like a technological singularity. It's just a nice application of optimisation techniques (which certainly can do cool things, after all it not too different from natural selection, check out this article for other such cool applications: http://www.damninteresting.com/on-the-origin-of-circuits)
    I wonder how well it would work for discovering things like the Einstein equations or the Standard Model. I'm thinking that it can only handle relatively simple mathematics, based on their examples, so it probably has no chance with these more advanced topics. Although, I guess math does boil down to simple operations in the end, it is just that without more compact and powerful notation it just ends up looking like a huge mess; the underlying patterns will be super hard to see. The biology example is cool though, often biological systems don't need terribly advanced mathematics to model them so it's probably true that their system could discover some neat patterns there.

    Still, if they can beef it up it would be really awesome to see it extract the Standard Model from the vast amount of particle physics data we have today. Although at best I imagine it would discover it in some horrible form like this: http://nuclear.ucdavis.edu/~tgutierr/files/sml2.pdf
    And even that makes use of a lot of compact notation

    Please Register or Log in to view the hidden image!

    .

    Oh, the other thing is that it is really only going to work for systems about which we can provide it with vast amounts of experimental data, if there something for which we have only a very minimal amount of data it takes a lot more creativity to come up with plausible explanations for it (although sometimes we end up with a vast over-abundance of explanations!)
     
  15. CheskiChips Banned Banned

    Messages:
    3,538
    From what I can gather; the algorithm seems to be based on the assumption that each component of an equation (or variable/node) has finite modifiers...then it seeks to find general rules for the nodes, then it reiterates the process over the nodes. In any case... it seems like it would have issues with dealing with equations which have greatly varying significance in node inputs. In meteorology for example some variables affects equations which are are magnitudes of orders less than others..but without them models diverge greatly.

    Amazing progress out of Cornell none the less.
     
  16. Search & Destroy Take one bite at a time Moderator

    Messages:
    1,467
    Why's that?
     
  17. CheskiChips Banned Banned

    Messages:
    3,538
    It relies on isolating inputs to predict the output of the equation. It obviously has to have a tolerance for accepting a node structure. If the input of one node is within that tolerance it seems likely that it wouldn't ever isolate the node.
     
  18. Ganymede Valued Senior Member

    Messages:
    3,322
    This isn't the beginning of the Singularity, it's merely one of the many breakthroughs that will continue to accelerate the pace of change that will eventually lead us to the singularity.

    Actually, once we reach the level of true A.I most of our own intelligence will be non-biological. At-least those who can afford to "upgrade" themselves. So the transition is going to be seamless IMO.
     
    Last edited: Jun 30, 2010
  19. jhon83 Registered Member

    Messages:
    4
    The amazing thing is that scientists make this machine and now they dont understand the machine's results.
     
  20. EndLightEnd This too shall pass. Registered Senior Member

    Messages:
    1,301
    Awesome, they should make this available to public and open source.
     
  21. Captain Kremmen All aboard, me Hearties! Valued Senior Member

    Messages:
    12,738
    Has it been fed the data from sciforums "On the Fringe" section yet?
     
  22. Captain Kremmen All aboard, me Hearties! Valued Senior Member

    Messages:
    12,738
    Was the Computer an Apple?
     
  23. Unconcept Registered Senior Member

    Messages:
    38
    Wow! I didn't think we were close to something like this
     

Share This Page