Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Thursday October 19 2017, @02:39PM   Printer-friendly
from the Zeroing-in-on-AI dept.

Google DeepMind researchers have made their old AlphaGo program obsolete:

The old AlphaGo relied on a computationally intensive Monte Carlo tree search to play through Go scenarios. The nodes and branches created a much larger tree than AlphaGo practically needed to play. A combination of reinforcement learning and human-supervised learning was used to build "value" and "policy" neural networks that used the search tree to execute gameplay strategies. The software learned from 30 million moves played in human-on-human games, and benefited from various bodges and tricks to learn to win. For instance, it was trained from master-level human players, rather than picking it up from scratch.

AlphaGo Zero did start from scratch with no experts guiding it. And it is much more efficient: it only uses a single computer and four of Google's custom TPU1 chips to play matches, compared to AlphaGo's several machines and 48 TPUs. Since Zero didn't rely on human gameplay, and a smaller number of matches, its Monte Carlo tree search is smaller. The self-play algorithm also combined both the value and policy neural networks into one, and was trained on 64 GPUs and 19 CPUs over a few days by playing nearly five million games against itself. In comparison, AlphaGo needed months of training and used 1,920 CPUs and 280 GPUs to beat Lee Sedol.

Though self-play AlphaGo Zero even discovered for itself, without human intervention, classic moves in the theory of Go, such as fuseki opening tactics, and what's called life and death. More details can be found in Nature, or from the paper directly here. Stanford computer science academic Bharath Ramsundar has a summary of the more technical points, here.

Go is an abstract strategy board game for two players, in which the aim is to surround more territory than the opponent.

Previously: Google's New TPUs are Now Much Faster -- will be Made Available to Researchers
Google's AlphaGo Wins Again and Retires From Competition


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 4, Insightful) by vux984 on Thursday October 19 2017, @06:31PM (1 child)

    by vux984 (5045) on Thursday October 19 2017, @06:31PM (#584712)

    Novel as far as I know doesn't mean useful or great

    The fact that the strategies were discovered and reinforced as valid by repeated play does tell us the AI found them useful.

    Might just be so utterly useless it has been written off by humans or it's something that requires a form of thinking humans can't or won't do

    Now we're just moving the goal posts. "So, it found a new strategy that had never been formally recognized that it is actively using to help it win, well... I'll only be impressed if humans can use it!"
    Frankly, take it as a small compliment to the human race that that it didn't find 'one weird trick that always wins' that we'd somehow missed for a few thousand years. Seriously what did you EXPECT?

    I'm sure it was great for their software but for the game? And for humans? Probably not so much.

    Are the goalposts even on the field anymore? They set out to beat humans at go with a machine, something that was only recently projected to be something still a long ways away. And they succeeded, decisively. That's impressive. Now the new generation requires only a fraction of the hardware and resources the previous generation needed, and not only still cleans up humans, but also cleans up the previous system. That's impressive.

    This is in general the issue or problem, I have, with AlphaGo. It hasn't really done anything, except to play games.

    That's precisely the task for which it was made.

    But the current AlphaGo ego stroke is really not all that awesomesauce they claim it to be.

    What claim did they make that you are so offended by?
    Go was considered something that couldn't be defeated by machine until very recently. It was considered that the immense number of possible moves, and the difficulty that even humans had at quantifying the strength of a move made it a difficult problem. I'm very impressed they solved it. I don't think for a second that this means we're on the brink of a sentient AI, but I'm still impressed.

    Starting Score:    1  point
    Moderation   +2  
       Insightful=2, Total=2
    Extra 'Insightful' Modifier   0  
    Karma-Bonus Modifier   +1  

    Total Score:   4  
  • (Score: 2) by rylyeh on Thursday October 19 2017, @10:34PM

    by rylyeh (6726) <kadathNO@SPAMgmail.com> on Thursday October 19 2017, @10:34PM (#584953)

    People who play Go know that there are few, if any, truly novel openings that have not been analyzed long ago.
    For AG to discover Any new openings that work is, itself, amazing. Remember the Bobby Fischer opening for Chess? [URL:https://en.wikipedia.org/wiki/Bobby_Fischer/]

    --
    O friend and companion of night, thou who rejoicest in the baying of dogs {here a hideous howl burst forth}...