Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Go master Lee Se-dol says he quits, unable to win over AI Go players

en.yna.co.kr

Post reply on HN
Print view View original

301–310 of 554 posts

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#301

Post by jmcgough »

dragontamer wrote: post #273
btilly wrote: post #260

Earlier quoted context omitted.

You're thinking too much like a human. No, you're thinking too much like someone who understands chess but not Go. Suppose that we built an opening book with a trillion reasonable courses of action on it. Each one analyzed well. As you have discovered, you will only go a few ply into the game. And all of the positions that you will be directed towards will have only a small edge. Instead put a tiny fraction of the co…

> Instead put a tiny fraction of the computing power necessary to build this book into self-training. Do you believe that AlphaZero could continue to improve dramatically with another 6-months of training? Or if it can improve at another 6-months after that? At some point, the network will reach a local maximum, and it will be unable to improve beyond that. Characterizing AlphaZero's moves through big-data analysis i…

> Anyone with an opening book of AlphaZero's mistakes will be able to push AlphaZero into a mistaken position.

This only works if AlphaZero never retrains on previous games.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#302

Post by codebje »

dragontamer wrote: post #193
CydeWeys wrote: post #188

Earlier quoted context omitted.

AlphaZero uses Monte Carlo Tree Search. It definitely has randomness built in. I'm with the other guy; I don't see how a person could make the AI play any better.

> AlphaZero uses Monte Carlo Tree Search. It definitely has randomness built in. That's not what MCTS means. MCTS refers to the bandit problem which formulated the search parameters. MCTS always chooses the "most interesting" path to explore. (Where "most interesting" is the path that balances explore-and-exploit hyper-parameters). AlphaZero improved upon MCTS by deferring to the neural net as the hyper-parameter. Bu…

MCTS is a random algorithm, and AlphaGo is no exception.

The AI selects a move. What state is the board in now? It doesn't know, because the opponent also selected a move.

MCTS models this with a probability distribution of the states, and samples from this distribution repeatedly to build an estimate of the effectiveness of each move it could make.

But what's the probability of each move made by the opponent? And after the simulation has looked as many moves ahead as it can in the time constraints, how good a position is it in?

These are the same question, really - what's the chance of winning from this board state. In Chess you can use a heuristic algorithm to figure it out. In Go, you can't. But you can use a neural network to learn an approximation that improves as it sees more games complete.

AlphaGo does this. MCTS is a random sampling technique, and the neural net informs its probability distributions, but doesn't make it deterministic.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#303

Post by Ma8ee »

neohaven wrote: post #93
SchoolboyBlue wrote: post #21

Earlier quoted context omitted.

I would counter with the fact that in physical endeavors it's apparent to us that we are not #1 - our household cats are more agile than us. It's in matters of intellect that humans still believe they are #1. AlphaGO's achievement in another field would have similar effects, e.g.: - An AI that diagnoses sickness better than any doctor - An AI that generates text which humans believe more beautiful than any other poet…

As a person who likes music, making it, listening to it, breaking it down and hacking it... Making a classical arrangement that evokes a particular expression in the listener is the job of the musician. If an AI system helps you explore the possibilities there, it's more like a studio musician that's able to improvise. You're still the person, the human, the emotional filter, that picks "This sounds right" or "This d…

Imagine the day when you can’t find a more satisfying note than the one the computer already have chosen.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#304

Post by sixstringtheory »

mastax wrote: post #271
jimbokun wrote: post #175

Earlier quoted context omitted.

> Come on, none of us are really "human" anymore since the advent of cell phones. We all use our cyborg-capabilities to search the internet and fact-check ourselves every day. Programmers use stack-overflow to teach themselves programming and remember obscure details (using our cyborg capabilities to tag, search, and sift through information ever faster and faster). Sure and any of us could jump on a motorcycle and f…

> Sure and any of us could jump on a motorcycle and fly past Usain Bolt in the 100m, but that kind of misses the whole point of the competition. An accurate metaphor would be a person on a motorcycle competing against another person on a motorcycle.

Or rather, two people showing up to a footrace with motorcycles. It sounds a bit silly to call it a footrace at that point. Nothing wrong with motorcycle racing, but gotta be honest about it.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#305

Post by ammon »

btilly wrote: post #260
dragontamer wrote: post #233

Earlier quoted context omitted.

> Go does not have openings. There are 381 opening moves in Go, but really only 96 because of symmetry. 96 (opening moves) x 380 responses x 379 x 378 x 377 == ~2 Trillion positions after 5 ply. These 2-trillion positions will easily fit in a 16TB hard drive for $400. That's 8-bytes per position, so you probably can get there with more symmetries and some compression applied. ---------- You're thinking too much like…

You're thinking too much like a human. No, you're thinking too much like someone who understands chess but not Go. Suppose that we built an opening book with a trillion reasonable courses of action on it. Each one analyzed well. As you have discovered, you will only go a few ply into the game. And all of the positions that you will be directed towards will have only a small edge. Instead put a tiny fraction of the co…

I want to add that opening books are not THAT effective in computer chess either. Yes, they are significantly more effective than in computer go (because of the smaller branching factor and greater role of tactics). However, exponential increase in game tree size is crazy, even in chess. Thus, opening books really can't take you THAT far into a chess game (10 moves). An engine with a worse book but better search/eval will usually win.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#306

Post by v64 »

arcticfox wrote: post #186
scarmig wrote: post #154

Earlier quoted context omitted.

Question: if we discovered a way for white to always force a win, would humans be able to use that to beat black AIs consistently?

It feels kind of surprising that this doesn't exist (yet). There are only a handful of reasonable moves per turn, and given the unbelievable amounts of moves that offline chess engines can process given time and server farms, I'm just surprised that doesn't overwhelm the reasonable branches at some point into forcing a win. Maybe someday.

For what it's worth, endgame table databases exist that give perfect play for when there are a limited number of pieces left on the board.

As of now, we have databases for perfect play when 7 pieces remain on the board (including the two kings). The 8-piece tablebase is computationally possible, but I don't believe a comprehensive release has come out yet. Even the current 7-piece tables are incomplete because situations like lone king vs. six opposing pieces haven't been explicitly calculated due to their obviousness.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#307

Post by btilly »

patcon wrote: post #293
btilly wrote: post #260

Earlier quoted context omitted.

You're thinking too much like a human. No, you're thinking too much like someone who understands chess but not Go. Suppose that we built an opening book with a trillion reasonable courses of action on it. Each one analyzed well. As you have discovered, you will only go a few ply into the game. And all of the positions that you will be directed towards will have only a small edge. Instead put a tiny fraction of the co…

fwiw, I'm with you. Excuse the gratuitous war analogy, but I suspect the approach of OP is akin to talking about which first 8 steps to take (north-east, SW, S) when going into battle -- it's such a small scope of the whole event that it's pointless, and talking about those first steps makes one seem naive to the actual holistic task

The analogy is a very appropriate one.

If the OP spent a month learning Go, I am sure that it would make sense to him as well. Work through a series like https://senseis.xmp.net/?LearnToPlayGoSeries while playing Go regularly against a variety of opponents. Before book 3 it should be obvious.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#308

Post by bhl »

p1necone wrote: post #280
pjscott wrote: post #262

Earlier quoted context omitted.

"Solved", in this case, means "computers can play the game at levels no human can beat."

That's not the normal meaning of solved in regards to game theory.

I believe with respect to game theory, solving a game like Go would require finding a strategy that obeys the one shot deviation principle. The result would rather be boring to watch however, because the conclusion of every game played under this strategy would either be draw, or based on which player starts off first.

[1] https://en.wikipedia.org/wiki/One-shot_deviation_principle

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#309

Post by dmix »

eyegor wrote: post #51

Am I the only one reading this article with the take away that the ai is not his primary reason for retirement? I understand that the title has its own conclusion but it seems overly sensational to me. > He actually quit the KBA in May 2016 and is now suing the association for the return of his membership fee. > "... [I] have something else to do," he said, asserting his only dream for now is to rest and spend time w…

He was handed the perfect excuse.

Re: Go master Lee Se-dol says he quits, unable to win over AI Go players

#310

Post by dragontamer »

btilly wrote: post #296
dragontamer wrote: post #273

Earlier quoted context omitted.

> Instead put a tiny fraction of the computing power necessary to build this book into self-training. Do you believe that AlphaZero could continue to improve dramatically with another 6-months of training? Or if it can improve at another 6-months after that? At some point, the network will reach a local maximum, and it will be unable to improve beyond that. Characterizing AlphaZero's moves through big-data analysis i…

AlphaGo spent months training and continued to improve for the whole time. Its improvement slowed, just as it takes more work for humans to get from master to international master than it does from D-level player to C-level player. But it did not stop improving. And then AlphaZero was better than AlphaGo after around a day of self-training. Furthermore you are arguing for an opening book without considering how small…

We have LeelaZero numbers: https://ogs-forums.s3.amazonaws.com/original/2X/2/21de8caa52...

The plateau is real. I'm not sure if continuous self-play will lead to continuous progress for all of eternity. The system is clearly slowing down in self-learning.

AlphaZero's plateau is also well documented: https://i.imgur.com/NMNp6Kq.png

--------

I'm not trying to cast doubt upon reinforcement learning / MCTS / Neural Nets in the game of Go. It is clearly the best methodology we got today.

But anyone who has any experience with neural nets knows about the local-maxima problem. ALL neural nets reach a local maxima eventually. Once this point is reached, you have to rely upon other methodologies to improve playing strength.

Assuming Elo-growth for all time using a singular methodology is naive. We will go very, very far with Deep Learning, but are you satisfied with that limit? Other open questions remain: Go is very far away from being a solved game, even with a magical machine that plays 2000+ Elo stronger than humans.

Post reply on HN