Why are super top professional Go players so stupid?

[Laskos]

I learned in one-two hours of analysis more than they have learnt in decades of study and practice. I guess they are natively not that gifted. Lack of vitamins.


How to play or analyze Go games:

In Windows:

1/ Download Lizzie --- Leela Zero Interface
https://github.com/featurecat/lizzie/releases
Unpack the CPU or GPU (better) versions somewhere.

2/ Download Leela Go Zero --- a Go program with no human provided knowledge, using MCTS and a deep residual convolutional neural network
https://github.com/leela-zero/leela-zero/releases
Unpack it and place the resulting folder inside the Lizzie folder.

3/ Download the latest weights
http://zero.sjeng.org
Don't unpack them, just place the "gz" file in the Lizzie folder.

4/ Open the config.txt file in the Lizzie folder, and modify the "network-file" and "engine-command" entries according to your imported weights and engine path and command.

5/ Double-click on the lizzie.jar executable. The GUI will open, and if everything in config.txt is fine, it will initialize the engine and the weights file, and will start analyze the blank opening position of Go. Long-press and hold "x" to see the options.



All is set to play and analyze Go games. On my RTX 2070 GPU, I estimate that my (Kai's) playing strength is above AlphaGo which beat Lee Sedol 4-1, and somewhat below AlphaGo Zero, which crushed Ke Jie, both these human players being living legends of Go (but as I said, pretty stupid).

I took a recent very top game of Ke Jie versus Lee Sedol, where Lee Sedol (White) won. Here is the SGF file which you can import in Lizzie:

Code: Select all

(
;PB[Ke Jie]BR[9p]PW[Lee Sedol]WR[9p]RE[W+R]DT[2018-05-21]HA[0]KM[7.5]SZ[19]AP[WGo.js:2]FF[4]GM[1]CA[UTF-8]
;B[pd]
;W[dp]
;B[qp]
;W[cc]
;B[np]
;W[fq]
;B[qf]
;W[fd]
;B[bo]
;W[cp]
;B[ck]
;W[ci]
;B[cn]
;W[ob]
;B[pb]
;W[ej]
;B[oc]
;W[jd]
;B[he]
;W[hd]
;B[cd]
;W[dc]
;B[fe]
;W[ge]
;B[gf]
;W[gd]
;B[di]
;W[dj]
;B[cj]
;W[ch]
;B[ei]
;W[dl]
;B[cl]
;W[dg]
;B[fi]
;W[fg]
;B[ff]
;W[gg]
;B[hg]
;W[hh]
;B[ed]
;W[de]
;B[ee]
;W[dd]
;B[ec]
;W[ih]
;B[ie]
;W[id]
;B[ig]
;W[ef]
;B[fb]
;W[nc]
;B[kg]
;W[nb]
;B[le]
;W[lc]
;B[hb]
;W[ib]
;B[nd]
;W[qk]
;B[qm]
;W[qh]
;B[gp]
;W[hr]
;B[iq]
;W[hq]
;B[hp]
;W[ip]
;B[fp]
;W[ep]
;B[gq]
;W[gr]
;B[fr]
;W[eq]
;B[ir]
;W[er]
;B[kq]
;W[io]
;B[ko]
;W[ok]
;B[pj]
;W[pk]
;B[ri]
;W[ki]
;B[qi]
;W[ni]
;B[mk]
;W[nm]
;B[mi]
;W[mh]
;B[ld]
;W[mc]
;B[li]
;W[lh]
;B[kh]
;W[oh]
;B[ng]
;W[nh]
;B[rh]
;W[lk]
;B[jb]
;W[ia]
;B[kc]
;W[kb]
;B[kj]
;W[ji]
;B[ml]
;W[md]
;B[ne]
;W[pm]
;B[jj]
;W[ij]
;B[oj]
;W[nj]
;B[kl]
;W[ik]
;B[ll]
;W[me]
;B[mf]
;W[lg]
;B[lf]
;W[kf]
;B[je]
;W[jg]
;B[jf]
;W[jh]
;B[kd]
;W[jc]
;B[qn]
;W[nk]
;B[gn]
;W[em]
;B[dk]
;W[ek]
;B[dm]
;W[en]
;B[el]
;W[fl]
;B[hm]
;W[hn]
;B[fk]
;W[dl]
;B[lb]
;W[ja]
;B[el]
;W[qc]
;B[pc]
;W[dl]
;B[oa]
;W[la]
;B[el]
;W[hf]
;B[if]
;W[dl]
;B[na]
;W[mb]
;B[el]
;W[kg]
;B[ke]
;W[dl]
;B[hs]
;W[el]
;B[bi]
;W[fs]
;B[bh]
;W[bg]
;B[im]
;W[go]
;B[on]
;W[qg]
;B[rg]
;W[pf]
;B[qe]
;W[bp]
;B[ap]
;W[gm]
;B[om]
;W[nl]
;B[db]
;W[cb]
;B[ca]
;W[ba]
;B[da]
;W[ah]
;B[bj]
;W[jn]
;B[jm]
;W[kn]
;B[in]
;W[lo]
;B[ho]
;W[kp]
;B[lq]
;W[mn]
;B[fo]
;W[mp]
;B[mq]
;W[co]
;B[bn]
;W[jl]
;B[gl]
;W[fm]
;B[hk]
;W[fj]
;B[il]
;W[jk]
;B[hj]
;W[km]
;B[eo]
;W[dn]
;B[do]
;W[gk]
;B[gi]
;W[hl]
;B[cf]
;W[df]
;B[gl]
;W[cm]
;B[cg]
;W[dh]
;B[bm]
;W[hi]
;B[kk]
;W[ii])

Here is a pretty faithful analysis of me of this game (green line, 50,000 playouts per move) and the final position:

KeJie_LeeSedol4.jpg

Both players blundered slightly and even not very slightly pretty often, hard to believe how stupid they are, but Ke Jie blundered decisively in wrongly judged life and death competing problems.

The mandatory moves

117. H1
and
125. M18

were missed by Ke Jie (I would have never missed them), and his whole position deteriorated irreversibly after these elementary blunders.

With my little pet Lizzie, I can now confidently enter the history as a Go legend.

Maybe I will now analyze the Meijin Retirement Game and the games of Go Seigen, to see how stupid these legends were.

[Laskos]

The assessments are confirmed by the best FaceBook network ELFv2, a 20x256 very strong net, which is about 2 times faster than Leela's own 40x256 net. At 200,000 playouts per move the eval graph shows the following, similar things to the original post:

KeJie_LeeSedol8.jpg

Similarly, both Ke Jie and Lee Sedol blunder quite heavily and often, but Ke Jie decisively with moves 117 and 125.
But the first to blunder heavily, and decisively if Ke Jie knew how to play, was Lee Sedol, right in the opening, move 36:

KeJie_LeeSedol9.jpg

Blue stone at E15 is the correct move by White (Lee Sedol) which results in a balanced game. Lee Sedol moved to the hollow white stone F13, a losing move (10% winrate).

[Raphexon]

Well there are more possible moves in Go (than in Chess) so naturally it would be logical to assume that a GO player with imperfect knowledge is more likely to blunder than (for example) a Chess player with imperfect knowledge.

[Laskos]

Well there are more possible moves in Go (than in Chess) so naturally it would be logical to assume that a GO player with imperfect knowledge is more likely to blunder than (for example) a Chess player with imperfect knowledge.

Yes, also the nature of blunders often seems to me (as far as I understand) to combine some tactical and positional considerations. It's most often not like they blunder a pawn in Chess, nevertheless, the consequences can be severe, but they can be offset by a counter-blunder by the opponent. I am pretty amazed having this thing on my PC as free open source software, that's why I posted here. These human players I am writing here about are Gods, and my stupid irreverencies please me .

[Ras]

Why are super top professional Go players so stupid?

Because it's much more difficult to find the right moves over the board. In chess, it's been for a long time that every patzer can start up Stockfish and, lo and behold, find much better moves than the champions played in the actual match.

The only thing that's new here: software is so advanced now that patzers can do the same with Go.

[Laskos]

Well there are more possible moves in Go (than in Chess) so naturally it would be logical to assume that a GO player with imperfect knowledge is more likely to blunder than (for example) a Chess player with imperfect knowledge.

Yes, also the nature of blunders often seems to me (as far as I understand) to combine some tactical and positional considerations. It's most often not like they blunder a pawn in Chess, nevertheless, the consequences can be severe, but they can be offset by a counter-blunder by the opponent. I am pretty amazed having this thing on my PC as free open source software, that's why I posted here. These human players I am writing here about are Gods, and my stupid irreverencies please me .

Also, on another note, in top human Chess many of the won games are due to at most 1-2 more serious blunders. In top human Go there is (exaggerating a bit) some sort of Brownian motion of blunders, where the "central limit theorem" of blunders can be roughly applied. I guess the strength distributions for top players of these two different games might differ.

[Hai]

Good to see that Leela changed from 20x256 to 40x256 net, because at the end bigger is better .

[Laskos]

Good to see that Leela changed from 20x256 to 40x256 net, because at the end bigger is better .

Depends on hardware balance. They don't seem to use CUDA (cuDNN) anywhere, only OpenCL, my speeds are not impressive on my RTX 2070. Might be due to batch sizes used not helping much with cuDNN, I don't know. But Leela's 40x256 latest net beats conclusively FaceBook latest 20x256 net at fixed nodes. Not sure at fixed time, especially LTC (I expect 40b to scale better than 20b).

[Laskos]

Why are super top professional Go players so stupid?

Because it's much more difficult to find the right moves over the board. In chess, it's been for a long time that every patzer can start up Stockfish and, lo and behold, find much better moves than the champions played in the actual match.

The only thing that's new here: software is so advanced now that patzers can do the same with Go.

Are you implying that I am weaker than Ke Jie?

[Ras]

Are you implying that I am weaker than Ke Jie?

Of course, that's why you need a computer to find the blunders.

The same happened in chess: "oohhhh Stockfish is swinging form +3 to 0, he threw away his win, it's soooo obvious." Yeah, sure.