Could have just posted this under the 'The End is Near' thread. . . .
Quote:
Algorithmic optimization is better done by computers than by humans in complex closed systems.
Rocag said:
How good is it at Global Thermonuclear War?
dargscisyhp said:Quote:
Algorithmic optimization is better done by computers than by humans in complex closed systems.
One of the interesting things about this experiment to me was that the machine considered several openings, but ultimately landed on human ones.
dargscisyhp said:Quote:
Algorithmic optimization is better done by computers than by humans in complex closed systems.
I'm not sure it actually shows this. Could be that human understanding of Chess is lacking.
One of the interesting things about this experiment to me was that the machine considered several openings, but ultimately landed on human ones.
I would like to see the results of this when it is limited to maybe looking at 5-10 positions a second. This is probably more in the human range. I wonder what it's performance would be like at that point.
amercer said:dargscisyhp said:Quote:
Algorithmic optimization is better done by computers than by humans in complex closed systems.
I'm not sure it actually shows this. Could be that human understanding of Chess is lacking.
One of the interesting things about this experiment to me was that the machine considered several openings, but ultimately landed on human ones.
I would like to see the results of this when it is limited to maybe looking at 5-10 positions a second. This is probably more in the human range. I wonder what it's performance would be like at that point.
I wonder if you could do the reverse experiment and crowdsource moves from the 1000 best human players for a match, thereby upping the human computation power.
To your other point about us maybe not understanding chess, all the top players practice against computers they can't beat. So maybe we are learning something.
kurt vonnegut said:
Could have just posted this under the 'The End is Near' thread. . . .
AstroAg17 said:
It's the first scenario. Alpha Zero during its learning phase experimented with pretty much all of the most popular openings. Here are graphs of which openings fell into and out of its favor over time.
Ulrich said:
I like to think that this validates my carelessness with pawns and frenetic style.
AgLiving06 said:AstroAg17 said:
It's the first scenario. Alpha Zero during its learning phase experimented with pretty much all of the most popular openings. Here are graphs of which openings fell into and out of its favor over time.
Gotcha. So it was aware of the popular openings, but didn't have to use them. During it's learning it figured out that they were in fact the "best" openings.
That's interesting on a couple fronts because it seems to validate that humans are on the right track with our openings, but that we make sufficient mistakes on a completely transparent game to lose anyways.
Aggrad08 said:
Did it find any openings that we underutilize or didn't know about as effective openings?
dargscisyhp said:AgLiving06 said:AstroAg17 said:
It's the first scenario. Alpha Zero during its learning phase experimented with pretty much all of the most popular openings. Here are graphs of which openings fell into and out of its favor over time.
Gotcha. So it was aware of the popular openings, but didn't have to use them. During it's learning it figured out that they were in fact the "best" openings.
That's interesting on a couple fronts because it seems to validate that humans are on the right track with our openings, but that we make sufficient mistakes on a completely transparent game to lose anyways.
As Astro has said, it was not aware of any openings beforehand. However, in the course of its learning it did stumble onto our openings, and analyzed them to quite a degree it seems. Also, the board is anything but transparent. It's hidden behind the wall of combinatorics. There are more possible board arrangements than atoms in the universe.
Wede01 said:
This may be the end of Nick Saban and Bill Billicek.
Couldn't you enter the rules of football and generate never before seen plays, formations, and strategies? Crazy thought.
dargscisyhp said:Wede01 said:
This may be the end of Nick Saban and Bill Billicek.
Couldn't you enter the rules of football and generate never before seen plays, formations, and strategies? Crazy thought.
Football would be a lot more complex. It's not a perfect information game, there are a lot more factors to consider (players' particular skillsets as opposed to set in stone rules for piece moves, weather where no analogue in Chess exists, perhaps physics itself to be able to simulate a game etc.) and chance almost certainly plays a part. For these reasons, I imagine we are a long way away from that.