Hacker News new | past | comments | ask | show | jobs | submit login
ACM Prize in Computing Awarded to AlphaGo Developer (acm.org)
166 points by rbanffy on April 1, 2020 | hide | past | favorite | 22 comments



As a long-time Go player and ML guy who even built his own (albeit shitty) Go AI in college, I'm a bit biased, but watching the AlphaGo-Lee Sedol match really felt like watching our generation's moon landing.

If y'all haven't already, there's a new AlphaGo documentary made by DeepMind on YT: https://www.youtube.com/watch?v=WXuK6gekU1Y. Brought tears to my eyes. Both a triumph for humanity in building an unbelievable machine like this, and a loss for humanity in that the infinite mystery of Go will be diminished, and again a triumph for humanity in Lee Sedol's brilliant win in Game 4...


FYI Michael Redmond has been doing free commentaries on Alpha Go’s games against itself which are really fun to watch: https://www.youtube.com/playlist?list=PLqpN3-2FP-kIxXhhdDds9...


Really great documentary, can't recommend it enough.


The documentary is excellent. I just wanted to mention that it's from a couple of years ago; it's just only now been put on YouTube.

https://www.imdb.com/title/tt6700846/


Q&A after winning game 4:

> What were you thinking when you made that play [move 78]?

> Lee Sedol: Move 78 was the only move I could see. There was no other placement. It was the only option for me, so I put it there.


This is one of the things that bugs me a little in the AI vs Human games. Reading the board is such a difficult skill in GO and humans get fatigued. We have to visually track all the places we could move. It only takes missing one of the top 10 moves on any given turn to lose against Alpha. It would be interesting to have AG or AZ play Lee when he has a day for each move. I'm not saying Alpha wouldn't still win but the computer is so fast and can consider so many things that a human can't hold only in their thoughts.

I do think it is a great achievement how far AI or ML has come. Not to take anything away from the team's accomplishments.


My understanding of high level go is that eventually how good you are comes down to something along the lines of your midichlorian count.


Thank you for this! It was a great piece.


Surely a great achievement. Although I feel a little bad for the team. I assume Silver didn't write all the code or design the entire algorithm single-handedly. But the recognition and the reward accrues to him alone. Ah, unjust hierarchical society.

I guess the other members of the team are being compensated well enough at DeepMind that $250k would be more icing than cake, but it still feels weird to see that Silver is the only person named in the article when a number of other world class researchers worked with him on this problem.


He's been working on Go for a very long time. Since he was a PhD student. Though I am sure the team helped, don't think he wasn't the main driving influence behind the algorithms and featurization of the problem.


Also, he's written an incredible series of groundbreaking papers throughout his career, going back to 2005. His papers tend to hold up very well. At this point, I carefully read any paper with his name on it, and I believe he very much deserves the honor.


Isn't the point of AlphaGo that Go is a minor detail?


Doesn't sound right as the point of AlphaGo itself. Generalization to other games came later, under other names (AlphaZero and MuZero).


Of course. I could imagine that he contributed 50%, 75%. I still feel bad for the team though!


I almost knew exactly who it was going to be when they mentioned AlphaGo Developer.

For those who aren't that well versed in RL, I recommend watching his lectures at UCL (https://www.youtube.com/watch?v=2pWv7GOvuf0). Really clear explanations that went hand in hand when I was reading Sutton and Barto's introductory book.


Another link with pdf with lectures, exam and assigment and a link to video of lectures in (1)

(1) https://www.davidsilver.uk/teaching/


I struggled for years to develop an AI that played the computer strategy for Empire.

http://www.classicempire.com/

I'd just run games, look at results, and endlessly tweak the strategy. Recently I learned how neural networks worked, and realize I could finally make a computer strategy that was competent. It could be trained by playing zillions of games against itself.

My only defense is that training a neural network was impractical on the machines Empire was developed on.

It's hard to resist going back to Empire and doing this.


How you ended up implement it in the code? Is it just a heuristic implemented as a series of if/else/case statements? Minimax? Thank you

EDIT: sorry, I found out that source code is available, I will try to find it

EDIT2: Looks like it's up for sale, not open


https://github.com/DigitalMars/Empire-for-PDP-10

It's the same algorithm. Mainly a bunch of ad-hoc heuristics.


Congratulations David Silver! I am currently learning Reinforcement Learning, as a coincidence I am watching David Silver's Introduction to Reinforcement Learning[0]. He explains it very clearly. I would recommend it for those who want to start to learn RL. Thank you David! Excellent!

[0] - https://www.youtube.com/playlist?list=PLqYmG7hTraZDM-OYHWgPe...


Well deserved! AlphaGo was simply a revolution.


Not to AlphaGo, whew, we're not there yet.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: