On Sunday, December 5, 2021 at 4:33:23 AM UTC, Tim Chow wrote:
In the position below, XG 3-ply plays 11/5* 8/5. (For Stick's benefit,
I'll explain that when I say that XG 3-ply plays 11/5* 8/5, I mean that
XG 3-ply plays 11/5* 8/5, and *not* that if you run all the candidate
plays through an XG 3-ply analysis, then 11/5* 8/5 comes out on top.)
Seems plausible, right? Someone---I forget who---has said, "If you're deciding between making the 5pt and some other play, just make the 5pt.
It will save you a vast amount of harm over your bg career." But check
out the rollout.
XGID=-AABCaB-B--B--b--cbbBbab--:0:0:1:63:1:5:0:7:10
X:Player 2 O:Player 1
Score is X:1 O:5 7 pt.(s) match.
+13-14-15-16-17-18------19-20-21-22-23-24-+
| O O O | | O X O O O |
| O O O | | O X O O |
| O | | |
| | | |
| | | |
| |BAR| |
| | | |
| | | |
| | | X |
| X X | | X X X |
| X X | | X O X X X X |
+12-11-10--9--8--7-------6--5--4--3--2--1-+
Pip count X: 111 O: 107 X-O: 1-5/7
Cube: 1
X to play 63
1. 3-ply 11/5* 8/5 eq:+0.400
Player: 56.99% (G:20.77% B:0.30%)
Opponent: 43.01% (G:13.07% B:0.55%)
2. 3-ply 11/5* 4/1 eq:+0.236 (-0.164)
Player: 51.19% (G:18.08% B:0.20%)
Opponent: 48.81% (G:13.18% B:0.51%)
3. 1-ply 8/2 8/5* eq:+0.290 (-0.111)
Player: 51.25% (G:16.46% B:0.24%)
Opponent: 48.75% (G:10.09% B:0.40%)
4. 1-ply 11/5* 5/2 eq:+0.283 (-0.117)
Player: 52.05% (G:14.01% B:0.18%)
Opponent: 47.95% (G:7.41% B:0.27%)
eXtreme Gammon Version: 2.19.207.pre-release, MET: Kazaross XG2
-------
Rollout
-------
XGID=-AABCaB-B--B--b--cbbBbab--:0:0:1:63:1:5:0:7:10
X:Player 2 O:Player 1
Score is X:1 O:5 7 pt.(s) match.
+13-14-15-16-17-18------19-20-21-22-23-24-+
| O O O | | O X O O O |
| O O O | | O X O O |
| O | | |
| | | |
| | | |
| |BAR| |
| | | |
| | | |
| | | X |
| X X | | X X X |
| X X | | X O X X X X |
+12-11-10--9--8--7-------6--5--4--3--2--1-+
Pip count X: 111 O: 107 X-O: 1-5/7
Cube: 1
X to play 63
1. Rollout¹ 11/5* 5/2 eq:+0.486
Player: 54.57% (G:17.84% B:0.27%)
Opponent: 45.43% (G:10.07% B:0.68%)
Confidence: ±0.011 (+0.475..+0.496) - [100.0%]
2. Rollout¹ 11/5* 8/5 eq:+0.412 (-0.074)
Player: 57.91% (G:22.38% B:0.29%)
Opponent: 42.09% (G:15.22% B:1.09%)
Confidence: ±0.009 (+0.403..+0.421) - [0.0%]
3. Rollout¹ 8/2 8/5* eq:+0.363 (-0.123)
Player: 55.03% (G:21.42% B:0.25%)
Opponent: 44.97% (G:11.49% B:0.78%)
Confidence: ±0.009 (+0.354..+0.372) - [0.0%]
Although I agree with the consensus that the best proxy for optimal play is the XG rollout,
it doesn't follow that human improvement should be gauged by matching the rollout rather than
matching XG multi-ply analysis.
After all, we can't do randomized rollouts OTB but strong players can do an approximate multi-ply analysis.
If one player's reasoning leads to a match with XG analysis and an other player's reasoning leads to a different play
which matches the rollout, I would think it would tend to be that the first player's reasoning was generally better, even though
it led to the wrong answer.
Life is full of scenarios where the best type of reasoning available leads to the wrong answer.
Here's an interesting philosophy puzzle on this topic.
Jane correctly reads the fuel gauge of her car and ascertains from this reading that her tank is full of petrol.
However, the fuel gauge has a fault, which Jane doesn't know about, and that fault leads to the gauge reading "full" regardless
of how much petrol is actually in the tank.
Fortunately for Jane, the tank does happen to be full.
Does Jane know that the tank is full?
Paul
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)