Hacker News new | past | comments | ask | show | jobs | submit login
AlphaGo’s ultimate challenge: a five-game match against Lee Sedol (googleasiapacific.blogspot.com)
146 points by wyclif on Mar 9, 2016 | hide | past | favorite | 33 comments



Wow, Lee Sedol just resigned. First game goes to AlphaGo. I wasn't sure who would win the 5 matches, but I never expected AlphaGo to win the first game!


Lee was rattled. This match might end up 5-0.


Too early to tell, IMO. The next game will be the bigger game since Lee Sedol has a far better idea what he's up against. If the next game goes like this one did, I would be more inclined to agree with you for the remaining 3.


He has the human advantages of adaptability and intuition, the better to try a new strategy next game.


Adaptability is not all that human (plenty of machines learn from their mistakes and adapt to new settings). Intuition is so poorly defined that depending on what you mean machines easily have it (heuristics, Bayesian inference, etc) or it is just sufficiently vague of a notion that it does not matter.


> plenty of machines learn from their mistakes and adapt to new settings

Curious what you see as examples of this.

> Intuition is so poorly defined that depending on what you mean machines easily have it (heuristics, Bayesian inference, etc)

As a working scientist and a bayesian practitioner, I'm sceptical algorithms have intuition. From my perspective, almost all models that one codifies are extremely brittle and will produce catastrophic failures (or just nonsense) unless the user possesses enough expert knowledge or intuition to a-priori know not to use the model in this regime.

However, I agree with the spirit of the text... go is a well-defined game and adaptability and intuition will be highly limited. For instance, the human can't just turn the board over, or unplug the game!


I guess I was referring to strategy specifically. The tactics are probably well in hand for both human and AI.

For the AI, the result of the first match will result in one more game entered in the its database. If it's like chess history, it's probably slanted a little towards that player's history in particular.

But the human player is well aware of the machine studying his strategic history and weighting it. If he's well studied like the chess guys are (is that how go players study?) he could employ a strategy he thinks would be surprising to the AI, or even plan to switch strategies in the middle. If one knows they are playing a pattern matcher, you can try to lead it to a local minimum and then leave it there.

Just speculating :-)


I really hope his performance isn't going to be affected by this loss, but I think that's a lot to ask of a person. 5-0 to AlphaGo is a big possibility at this point.


A bit late, but this is the AGA feed.

https://gaming.youtube.com/watch?v=YZPKR7HzM_s

No one can believe it. Myungwan Kim 9p says it's likely Lee Sedol feels like he could have won. He also says Alpha Go is likely stronger than he is.


I know it's not the focus of this game, but it would add to the presentation if a robot handled the stones for alphago.


Especially given today's post on the Google Research Blog: http://googleresearch.blogspot.ca/2016/03/deep-learning-for-...


and used computer vision algorithm to register the opponent's moves...


I saw this in an episode of the TV show, but I forgot name of the show.

PS.

"Person of interest": https://www.youtube.com/watch?v=HkvukotSSms


That would be evil genius :) But we will see if such an addition would come handy.


I wish they would highlight the most recent placements, so it's easier to watch intermittently.


We are relaying the match with discussion and analysis on OGS! The most recent move will be highlighted on the board.

https://online-go.com/demo/114161


Thanks for this! Loved the discussion.


I wish the commentators would pay more attention to the live game board, and less to their own copy of it.


Amazing.. I think AlphaGo is going to win.


Lee just resigned. O.o


Myungwan Kim 9p will be doing live commentary at the Korean Cultural Center in Los Angeles for game 3.

The second game should be a doozy since Lee Sedol will definitely know what to expect and come in full force!


If you're based in LA, here's the event link.

https://www.kccla.org/english/calendar_view.asp?cid=4020&imo...


My money is on the human. This time.


Hopefully there's still time to edit your comment before the robots notice...


https://twitter.com/mustafasuleymn/status/707469083458068480

(cough) Ok best 2 out of three before it counts. (cough)


My pal John Conner and I say they can bring it. Got EMP's waiting for them.



Halfway through the game and it's difficult to say who's winning. AlphaGo has definitely improved during the winter.


AlphaGo wins :)


AlphaGo won!


Lee seems to be ahead so far in game 1


I am so stoked for this match, Lee Sedol is a child prodigy and a legend....I literally felt as excited as I was going in McGregor vs Diaz before the fight.

I used to play Go when I was a kid on televised matches in Korea during the 90s and have woken up early on saturdays watching every game live on tv. Then I'd go to these Go school after class and there'd be like 30 students studying and fighting.

Go is a hugely appealing game to intuitive people rather than logical people who prefer Chess. Go is an infinitely more complex and at these Pro levels a demigod like Lee Sedol have the same fanatic followings.


so I watched this last night and it was an earth shattering moment...like no fucking way sedol gonna get bamboozled by a computer right?

AlphaGo winning was the cherry on top but what was really even more intense was the actual battle in itself. It was like Lee Sedol was playing himself but a version of him that would get better and better each time Sedol attacked. AlphaGo surprisingly chose the right strategy which was to be aggressive right back.

Overall, I could identify with the commentator's excitement and sort of apprehension that the first battle against the Machines have begun and lost the first round.

Lee Sedol must have been taken back at how good AlphaGo is I think he seriously underestimated it because he had a lot of hubris and over confidence going in like 'yeah imma smack the shit out of alphago' and then it after the match is like 'damn gg'.

The biggest ground breaking realization is that deep learning has become so good that it is possible to outperform a human even in previously thought impossible problems....who would've thought a bunch of logic gates fast forward 40 years we have machines that beat us in our own games? 80 years from now what will things look like?

It's a real reckoning and I really feel the drive to learn deep learning just don't know where to start




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

Search: