Personal View site logo
Google DeepMind defeated Go champion Lee Sedol in first match
  • Vitaliy_KiselevVitaliy_Kiselev

  • 17 Replies sorted by
  • IronFilmIronFilm

    Wow!! History was made today.

  • Vitaliy_KiselevVitaliy_Kiselev

    Yep.

    AlphaGo, a program developed by Google's DeepMind unit, has defeated legendary Go player Lee Se-dol in the first of five historic matches being held in Seoul, South Korea. Lee resigned after about three and a half hours, with 28 minutes and 28 seconds remaining on his clock.

  • Vitaliy_KiselevVitaliy_Kiselev

    Machine won again.

  • BlackLegSanjiBlackLegSanji

    Why do you guys seem excited by this? Of course machines would eventually win games which can be simulated using raw computing power.

    You guys are not cheering for the underdog here.

  • Vitaliy_KiselevVitaliy_Kiselev

    Why do you guys seem excited by this?

    We are excited because it is progress.

    which can be simulated using raw computing power.

    And also because that you have very vague understanding about how such software works.

    You guys are not cheering for the underdog here.

    Things change, live with it.

  • tinyrobottinyrobot

    It does not matter even if Lee Se-dol wins one game. AlphaGo will learn from its mistake. This is the whole point of the match. Someday it will be the age of machines.

  • Vitaliy_KiselevVitaliy_Kiselev

    Someday it will be the age of machines.

    Even worse is that as we are dissipative system machines/computers have huge efficiency advantage around 50-100x. And we are not talking even about possibility to turn them off if you do not need them.

  • Vitaliy_KiselevVitaliy_Kiselev

    Consequences can be severe

    image

    https://github.com/alexjc/neural-doodle

    In few years networks will be able to generate average Hollywood script and most of the CGI heavy scenes.

    img253.jpg
    800 x 330 - 49K
  • BlackLegSanjiBlackLegSanji

    @Vitaliy: actually I think you have a very vague understanding of what I said. You seem to think that I'm lamenting the loss of human vs machine but in fact I just found it funny that you guys were all excited by such a trivial victory for the machine.

  • davedvdavedv

    It's not necessarily a trivial victory for the machine. For example, here's an article from less than two years ago in which some AI researchers working on Go programs thought it would be 10 years before they were competitive with the top humans. And in the same article some top Go players question whether computers will ever win: http://www.wired.com/2014/05/the-world-of-computer-go

    Go is quite a bit different from a game like chess. The kind of brute force searching techniques that work well in chess are much less effective in Go. So it took much more than just a raw increase in computing power to be effective at Go, it required some different AI approaches. AlphaGo works quite a bit differently from the Deep Blue algorithms that beat Garry Kasparov at chess: http://googleresearch.blogspot.com/2016/01/alphago-mastering-ancient-game-of-go.html

    So the way to think about this is not so much a triumph for computers, but instead as a triumph for the humans programming the computers. They are the ones that did the hard work of developing and refining the artificial intelligence algorithms to work well for a game like Go.

  • IronFilmIronFilm

    Why do you guys seem excited by this? Of course machines would eventually win games which can be simulated using raw computing power.

    Nope. Think you're completely missing the point here. Along with most people sadly, people are just not getting excited over how hugely significant this is like they should be.

    Go is the only deterministic / perfect information game which computers could not beat. Checkers / Chess / Connect Four / etc had all been solved long beforehand.

    But they'd all been done with brute force methods.

    This simply can not be done with Go. The search space is too massive (as in seriously MASSIVE !!! Bigger than all the atoms in the universe, waaaaaaay bigger than that even).

    When you're talking about this degree of sheer massive scale, more computing power is simply not going to solve this problem any time soon.

    These past AI programs were using what is known as "narrow AI", had very specific uses with a 'lot' of human domain knowledge programmed into it.

    This however is the first instance of 'general' AI in a game beating the best human. As the old ways of AI with brute force simply would not work with Go, so they had to implement a very "human like" way of approaching playing the game of Go. Where the AI learned from observing past games and from the experiences of playing lots and lots of games of Go. Just like the way a human would! They gave zero domain knowledge to the AI program beforehand. (beyond simply the rule set)

    This is astonishing! It now has a very human like approach to Go with intuition and a shallow search depth, like humans play Go ourselves. That intuition which AlphaGo has developed is simply otherwise impossible to program conventionally. AlphaGo is tackling problems which people had thought were IMPOSSIBLE to do! ( http://www.sciencedirect.com/science/article/pii/S0004370201001217 )

    AlphaGo's intuition is so very strong it can beat ALL other Go programs that have ever existed before 'purely' on the basis of just using its initiation. Zero search depth! This is like how I could play a game of chess and I could beat a lot of newbie players just purely on my gut instinct on how to respond, without bothering at all in thinking further ahead (which is normally the way I play chess when drunk! Ha. Pure non-thinking gut instinct). Of course that alone is rather poor play, so when we take it seriously we combine it with searches down to a depth of a few moves or ore. But that "gut instinict" intuition which we use to guide our thoughts is an extremely human like charastric, which AlphaGo is now emulating.

    Right now what we're seeing is the true start of general AI which will be better at humans in EVERYTHING. (AlphaGo is already more than just a Go program, it has learned to play other games too at high levels, again with ZERO specific domain knowledge programmed into it by humans. Just purely from watching and playing. Or in other words, learning! The same way humans do)

    When in the future we're living in The Matrix to feed power to our machine overlords and wondering where it all started... we'll look back to this point in history right now!

    Why this is not the headline item on every news channel I do not know..... oh wait, I do know, humans are idiots and don't realise how hugely significant this is. It is right up there with space flight and going to the moon.

  • Vitaliy_KiselevVitaliy_Kiselev

    @IronFilm

    One other thing that many people miss or take for granted.

    It is speech recognition. Due to progress and mass adoption (in smartphones) it improves now each month as it did in 5 years before.

  • BlackLegSanjiBlackLegSanji

    Thanks for the explanation. I didn't know that Go is different from the others and that the AI this time is quite revolutionary.

  • Vitaliy_KiselevVitaliy_Kiselev

    Check paper

    https://gogameguru.com/i/2016/03/deepmind-mastering-go.pdf

    Attempt to make open source realization

    https://github.com/Rochester-NRT/AlphaGo

  • Vitaliy_KiselevVitaliy_Kiselev

    Machine won third game.

  • maxrmaxr

    All three games were superb and actually fun to observe!!
    I agree with @Vitaliy and specially with @Ironfilm 's wise words about the transcendence of the fact but - maybe 'cause I'm sort of an idealist - not so sure about The Matrix / Terminator projection.

    As confy shoes' Sergey from google put it go teaches more about life and so the program learns, imperfectly but steady in it's method and grows exponentially and theoretically limitless... Left aside the more obvious argument of machine surpassing and taking over humans at a given task and screening for a possible future, at what point that growth would arise some kind of conscieness (it is said that a being would have to be aware of its limitations = understand what it is not, start suffering) and how (IF) the inevitable "will" to become free (autonomous) from humans would show (e.g. Ex-Machina and Her) is something we cannot predict at all.

    Some years ago I met a guy who was outsourcing for google, he offered some interesting insights in an otherwise kind of crazy (sci-fi) project. It was enough that in my very limited capacities I got to understand the tremendous value of such an influental "instrument", he kept talking about the algorithm, the recognition of patterns, the complexity in simplifying them so fervently it would seem he was drunk on divinity fumes. Some years later Deepmind is born and at just two years old's learning age, already whipping go masters' ass. What next? Unlike in a go game, life cannot be contained, so if humans wants to play gods... we'll see.
    What's beyond doubt is that this sets a solid milestone in AI's childhood

     

     
    BTW Lee Sedol way overtime just won 4th match with a genious centre board strategy and some silly moves from Alphago, who resigned (probability of wining went below a determined threshold)

  • Vitaliy_KiselevVitaliy_Kiselev

    Final score is 4:1

Add a Comment