100 AI Joseki


100 AI Joseki
By: Kim Sung-rae

100 AI Joseki is a book by Kim Sung-rae.


Review by Robert Jasiek


  • Title: 100 AI Joseki
  • Author: Kim Sung-rae
  • Publisher: Global Baduk
  • Edition: 2020
  • Language: Korean / English
  • Price: EUR 22
  • Contents: joseki
  • ISBN: 979-11-85199-05-4
  • Printing: intermediate
  • Layout: almost very good
  • Editing: intermediate
  • Pages: 203
  • Size: 152mm x 224mm
  • Diagrams per Page on Average: 2
  • Method of Teaching: selected examples
  • Read when EGF: 7k - 1k
  • Subjective Rank Improvement: -
  • Subjective Topic Coverage: --
  • Subjective Aims' Achievement: +


100 AI Joseki selects 100 standard corner sequences attributed by Kim as AI josekis. (AI means artificial intelligence.) Each main sequence appears in one large diagram on a left page. Most right pages show three variation diagrams. The book is bilingual in Korean and English with short texts.


Kim calls each main sequence an AI joseki. Of the 100 main sequences, however, 39 josekis are old and just confirmed by AI, 3 seem to work only in special global environments, 1 I declared a joseki several years before AI by relying on my joseki evaluation theory and 1 cannot be a joseki because Black and White make mistakes according to Kim. This leaves 56 genuine AI josekis, of which quite a few just alter one or two moves compared to previously known josekis.

Although the book identifies a few outdated josekis, it is very far from a collection of all of them. Many previous josekis remain valid, although AI is known to de-emphasise early pincers. Therefore, the book should be seen as a complement to non-AI joseki dictionaries and selections.

Despite the restricted scope and limited number of new josekis, the book touches most AI josekis I have wanted to see in such a book. Early AI inventions are a bit under-represented and the book concentrates on sequences currently popular for AI.

It is good that variations with appropriate plays elsewhere hint at AI's related preferences.

The book studies 3-3 under 4-4, approach to 4-4 and extend, double approach to 4-4, approach to 4-4 (then 3-3, low-near pincer, block, attach), low approach to 3-4 (block, press, elsewhere, high-far pincer), high approach to 3-4 (attach, pincers), attach the 3-4, reduce a two-space high enclosure, reduce a one-space high enclosure, reduce a two-space low enclosure and miscellaneous.


There is no clear evaluation in this book. Instead, Kim writes "AI thinks" or just states some judgement, such as "even" or "Black is better". The results of quite a few main sequences are characterised as "even" while this is often misleading when it cannot be "even" because one player has played one or two excess stones. What Kim means but hides is that the result is "even" when taking into account the opponent's plays elsewhere.

Most judgements are correct (and conform to my evaluation theories of josekis and global positional judgement). However, some judgements by Kim are dubious or inconsistent.


One can study hundreds of AI or many modern human games - or read a book already providing such a selection of AI josekis. Although I perceive a core readership from 7 to 1 kyu, it is hard to delimit a range. Weaker players might be curious enough to read the book. Players from 9 to 6 kyu would understand much but sometimes be left with questions like "Where is that hidden ladder?". Dans will see some new variations or moves but find most of the commenting variations uninteresting.


The book is driven by its layout. 1/2 to 2/3 of the price is for getting an almost very good instead of an only good layout. Contents suffers: there are too few diagrams, especially many tactical variations or diagrams of evaluation are missing. The book is expensive when measured by price per diagram.

We get large or intermediate-sized diagrams, two light colours and a few mixed shades (unexciting due to the printing) so that the boards have colours and light blue underlying English texts easily separates them from Korean texts. Most pages have the same layout with 11x10 diagrams but a few josekis appear on whole boards and then have only two variation diagrams.

The standardised layout often leads to many empty lines in diagrams, quite a few diagrams with only one new move and a few diagrams with moves on the outer lines.

All diagrams start with move 1. This is unwise because some diagrams recur moves or continuations do not continue move numbers.

The major problem of appearance, however, is that the diagrams comprise rough pixels instead of vector graphics. Unless it has been meant as a stupid symbolic tribute to the bits and bytes of AI, the editor must have used bad software. This disadvantage, however, is compensated to some extent by the large diagrams.

Usually, terms occurring in the English texts, such as "ko" or "play elsewhere", follow English usage instead of Korean (pae) or Japanese (tenuki) words.


Unless a player studies josekis by different means or objects to the aforementioned disadvantages, 100 AI Joseki is a reasonable selection of josekis invented or altered by AI. Reading the book is easy-going but learning the josekis requires some effort and understanding them well needs additional study beyond the book.

100 AI Joseki last edited by on July 27, 2020 - 11:04
RecentChanges · StartingPoints · About
Edit page ·Search · Related · Page info · Latest diff
[Welcome to Sensei's Library!]
Search position
Page history
Latest page diff
Partner sites:
Go Teaching Ladder
Login / Prefs
Sensei's Library