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

Reimagining Chess with AlphaZero

cacm.acm.org

Post reply on HN
Print view View original

81–90 of 116 posts

Re: Reimagining Chess with AlphaZero

#81

Post by kzrdude »

Andrew_nenakhov wrote: post #76
kzrdude wrote: post #62

Earlier quoted context omitted.

You can't reach top 1% without being very good at taking advantage of time pressure, though, or am I wrong? The term rapid chess includes many time controls - both with and without increment, so I'm not sure how to understand you.

It doesn't matter, if there is a time increment, there are still rarely any draws. Maybe 1 in 15 or 20 games. Of course, I'm at the very bottom of that 1%, maybe things are somewhat different higher up. And seriously speaking, the more time given, the higher the probability of a draw: games are usually won or lost because of blunders. If you have time to evaluate all moves and play very patiently, and your opponent d…

I see.

I found the rapid world championship, and at the top level the number of draws is certainly higher. (And it's played with an increment).

See this table for example: https://chess-results.com/tnr600852.aspx?lan=6&art=5&flag=30 highlighting ½ with the search gives an idea of the number of draws.

Re: Reimagining Chess with AlphaZero

#83

Post by __s »

Tenoke wrote: post #75

If you want say less draws or no advantage for White or less memorization, the easiest solution to me seems to be to just have engines evaluate a few trillion random positions and then filter out those with 0 advantage and low chance of draws in explored lines and then use those as the starting positions.

As done in Checkers: http://www.quadibloc.com/other/bo010201.htm

& how TCEC (Top Chess Engine Championship) runs

Re: Reimagining Chess with AlphaZero

#84

Post by __s »

beardyw wrote: post #35

It would be interesting to see if there is anything which would let black start on equal terms.

Allow Black to choose White's first move. They aren't allowed to pick f3 (huh, looks like Stockfish greatly prefers f3 to g4)

Or more, based on https://lichess.org/analysis you'd have to ban a couple more moves like a4/h4, but you could opt to restrict picking first moves within Stockfish range at some depth of -0.1 to 0.1 (looks like c4 is enough to be 0.1 at depth 47)

This may increase the draw rate

If you want to keep engine analysis out of the rules, you can go the "evenly split candy between siblings" route: after White's first move & before Black's response, Black may elect to trade positions

Re: Reimagining Chess with AlphaZero

#85

Post by tasuki »

__s wrote: post #84
beardyw wrote: post #35

It would be interesting to see if there is anything which would let black start on equal terms.

Allow Black to choose White's first move. They aren't allowed to pick f3 (huh, looks like Stockfish greatly prefers f3 to g4) Or more, based on https://lichess.org/analysis you'd have to ban a couple more moves like a4/h4, but you could opt to restrict picking first moves within Stockfish range at some depth of -0.1 to 0.1 (looks like c4 is enough to be 0.1 at depth 47) This may increase the draw rate If you want to…

That seems needlessly unfair towards white. Isn't the pie rule (one player plays a move, the other chooses which colour they play) strictly better?

Re: Reimagining Chess with AlphaZero

#86

Post by __s »

tasuki wrote: post #85
__s wrote: post #84

Earlier quoted context omitted.

Allow Black to choose White's first move. They aren't allowed to pick f3 (huh, looks like Stockfish greatly prefers f3 to g4) Or more, based on https://lichess.org/analysis you'd have to ban a couple more moves like a4/h4, but you could opt to restrict picking first moves within Stockfish range at some depth of -0.1 to 0.1 (looks like c4 is enough to be 0.1 at depth 47) This may increase the draw rate If you want to…

That seems needlessly unfair towards white. Isn't the pie rule (one player plays a move, the other chooses which colour they play) strictly better?

Yes, but it'll get _much_ closer to 0 than e4 or d4, which are already close enough to 0 to produce good games won by black. I wouldn't call it needlessly unfair

It also means White still gets the edge of being able to prepare for the first move vs their opponent, ie they may've studied 1.Nc3 lines much deeper in preparation, whereas currently White gets the edge of choosing the first move & it's advantage

ie White gets to decide whether they prefer their e4 prep or d4 prep against their opponent's prep as Black, so if White doesn't want to play against their opponent's Sicilian they play d4. This is a theoretically moot but real advantage. See Fischer's performance with c4: https://www.chessgames.com/perl/explorer?pid=19233&side=whit...

See also how well 1.g3 or 1.b3 have performed while Stockfish gives both a solid 0 eval

In short, while I started with Yes, I mean No. Because b3/g3 are equal enough that White gets an edge by deciding whether the game will be a g3 or a b3 game

Re: Reimagining Chess with AlphaZero

#87

Post by tasuki »

mwattsun wrote: post #25
monsieurbanana wrote: post #15

Earlier quoted context omitted.

There is only one way to fix chess and it's called Go. Of course, this isn't exactly chess anymore but like any design decision there are no perf- Wait a minute, Go is perfect. If we forget about ko rules.

I was really impressed by "AlphaGo - The Movie", despite not knowing much about Go, because the way they told the story, people from the Asian countries were so sure their champion would win a Go tournament against a computer, that were stunned into a new consciousness about artificial intelligence when the computer won. This started an arms race in AI is what I gather. Edit: Full movie for free on YouTube: AlphaGo -…

I'm from Europe and was also sure the human would win.

The only thing that prevented me from losing a bunch of money on prediction markets was logistics.

Re: Reimagining Chess with AlphaZero

#88

Post by j4yav »

matharmin wrote: post #58
qsort wrote: post #56

Earlier quoted context omitted.

I'm particularly surprised that the stalemate = win rule doesn't result in more decisive games. Intuitively, it would seem that an entire class of endgames that are classically draws would become winning for the stronger side. Perhaps computers are so strong they can "see through" the rule change?

From the posted stats, the Stalemate=win rule does result in more decisive games (fewer draws). But it has an even bigger effect in giving white a further advantage over black.

What about white draws with stalemate and black wins?

Re: Reimagining Chess with AlphaZero

#89

Post by alasdair_ »

RcouF1uZ4gsC wrote: post #80

I think the best thing that can be done for human vs human chess is to reduce the number of draws. I think this can be done with a very simple rule change: In a draw, instead of a both players receiving half a point, one player will randomly receive a full point. The expected value of a draw is still the same (0.5 points per draw) but I think it does change the psychology of it.

That would make most major tournaments almost completely luck based.

It would be interesting to see what happens if draws were only worth a third of a point. Presumably that would encourage one of the players to risk more trying to win.

Re: Reimagining Chess with AlphaZero

#90

Post by bgroat »

thom wrote: post #45

I feel compelled to say that, Coronavirus shenanigans aside, Wijk an Zee has just finished and was a fantastic event with many interesting and decisive games. Classical chess does not need fixing, even at the highest level. It’s fun that we can run these experiments though. I’d love to test out some of the variants a friend and I came up with! One of my favourites was starting with all the pieces off the board and a…

When I was in high school I attempted to code a "Fog-of-War" chess where different pieces had different sight ranges.

The King had no sight individually, but their range was the sight-range of all pieces within one piece from them (because they could relay information)

Post reply on HN