Chapter 01 6 joseki

Joseki That AI Killed

For decades, these sequences were taught as correct, memorized by thousands of players, and played in the highest-level professional games. Then AI engines proved they were all mistakes — some by narrow margins, others by embarrassing ones.

Use the Traditional / AI Preferred tabs to switch between the old sequence and what AI recommends instead. Step through each move with the controls or arrow keys.

1950s–2016

Large Avalanche Joseki

One of the most complex joseki in Go history, the Large Avalanche was a test of memorization and reading ability. Professional players spent years studying its dozens of variations. It was considered essential knowledge for any serious player.

Traditional Large Avalanche

MMNNOOPPQQRRSSTT1919181817171616151514141313
AI VERDICT -3% for both sides vs simpler alternatives

AI engines evaluate the Large Avalanche as roughly equal but overly complex — both sides take on unnecessary risk. Simpler alternatives achieve the same strategic goals with far less reading required and fewer chances for mistakes.

The avalanche's complexity was its selling point for humans. For AI, complexity without clear benefit is just unnecessary risk. Why memorize 40+ moves when a 7-move sequence achieves the same result?

1970s–2017

Attach-and-Extend at 3-4

When approaching a 3-4 stone, the attach-and-extend (tsuke-nobi) was a standard technique taught in Go schools worldwide. The idea: force your opponent to choose a direction, then develop on the opposite side. Considered "efficient" for over 40 years.

Classic Attach-Extend

AABBCCDDEEFFGGHHJJ1919181817171616151514141313
AI VERDICT -2.5% for White

AI showed that the attach-and-extend over-concentrates White's stones. The attachment helps Black solidify the corner while White gets a low, flat position. A one-space approach or shoulder hit creates better shape and more flexibility.

The attach helps your opponent. This simple principle was hiding in plain sight for decades, but human intuition valued the "forcing" nature of the attachment over the subtle loss in efficiency.

2000s–2017

Mini-Chinese Opening

The Mini-Chinese (also called the "Micro-Chinese") was one of the most popular fuseki systems in the 2000s and 2010s. Adopted by top professionals like Gu Li and Lee Changho, it aimed for a balanced opening with flexible follow-ups. At its peak, it appeared in roughly 15% of professional games.

Mini-Chinese Formation

AABBCCDDEEFFGGHHJJKKLLMMNNOOPPQQRRSSTT1919181817171616151514141313121211111010998877665544332211
AI VERDICT -1.5% for Black

AI evaluates Black's Mini-Chinese extension (R10) as slightly inefficient. The stone on the right side is too far from both corners to defend either one effectively. It also creates a predictable game plan that White can counter with established patterns. AI prefers more flexible opening moves.

Instead of building a "system" on one side, AI prefers immediate engagement. The era of opening systems — Chinese, Mini-Chinese, Kobayashi — gave way to flexible, contact-heavy play from move 5 onward.

1980s–2017

Star Point Large Knight Enclosure

The large knight enclosure (ogeima shimari) from the star point was considered a natural, flexible way to secure corner influence. Every Go textbook taught it as a solid opening move. It appeared in thousands of professional games alongside the small knight enclosure.

Traditional Large Knight Shimari

AABBCCDDEEFFGGHHJJ19191818171716161515141413131212
AI VERDICT -2% for Black

AI revealed a fundamental flaw: the large knight enclosure leaves a critical weakness at the 3-3 point. White can invade and live unconditionally. Unlike the small knight, the large knight cannot prevent White from taking the corner, making all that "influence" illusory.

The one-space difference between small knight and large knight seemed trivial. AI showed it's the difference between "corner is secure" and "corner is invadable." That single intersection changed everything.

1930s–2016

Orthodox Fuseki (Parallel Opening)

The orthodox (or parallel) fuseki — both players occupying diagonally opposite 3-4 points, then approaching the opponent's corner — was the dominant professional opening for almost a century. It was considered the most balanced, theoretically sound way to begin a game.

Classic Parallel Fuseki

AABBCCDDEEFFGGHHJJKKLLMMNNOOPPQQRRSSTT1919181817171616151514141313121211111010998877665544332211
AI VERDICT -1% for both sides (too passive)

AI plays approach moves much earlier and favors wider spacing. The orthodox pattern of "secure corners first, then approach" is too slow. AI typically approaches by move 3, before the opponent finishes their corner setup. Contact fights in the opening create more opportunities.

Humans valued symmetry and balance in the opening. AI values initiative and disruption. The entire concept of "building a framework before fighting" was replaced by "fight immediately, build from the wreckage."

1960s–2017

High One-Space Approach to 3-4

When approaching a 3-4 stone from the outside, professionals had two main choices: the low approach (keima kakari) and the high approach (one-space high). The high approach was preferred in many situations because it emphasizes influence and center development — values deeply embedded in the Go aesthetic.

Classic High Approach

AABBCCDDEEFFGGHHJJ998877665544332211
AI VERDICT -1.8% for White

AI overwhelmingly prefers the low approach over the high approach. The high stone leaves more room for the defender to counter-attack, and the promised "influence" it generates is less valuable than previously thought. In AI's evaluation, solid territory outweighs vague influence in most positions.

The high vs low debate was one of Go's great aesthetic arguments. AI settled it decisively: low wins. Influence was the most overrated concept in classical Go theory — territory you can count beats influence you can only feel.

More joseki coming soon.

Target: 10-15 entries

Built by a 6-dan Go player who watched AI rewrite the game.

Go × AI Unpacked