Skip to main content

The KataGo Era (2019-Present)

After AlphaGo retired in 2017, Go AI development didn't stop. On the contrary, the open source community took up the torch, making top-tier Go AI no longer the exclusive property of tech giants but a tool everyone can use. This "AI democratization" movement completely changed how Go is learned and trained.

The Rise of Open Source AI​

Leela Zero: The Power of Community​

In 2017, inspired by the AlphaZero paper, Belgian programmer Gian-Carlo Pascutto launched the Leela Zero project. This was a completely open-source Go AI with the goal of replicating AlphaZero's achievements.

Leela Zero's features:

  • Completely open source: Code is public, anyone can participate in development
  • Distributed training: Uses computers from volunteers worldwide for training
  • Starting from zero: Uses no human games, learns entirely through self-play

After years of community effort, Leela Zero reached extremely strong playing levels, becoming one of the earliest popularized open-source Go AIs.

ELF OpenGo: Facebook's Contribution​

In 2018, Facebook AI Research released ELF OpenGo, another open-source Go AI based on AlphaZero architecture.

ELF OpenGo's significance:

  • Provided complete training code
  • Published pre-trained neural network weights
  • Allowed researchers to make improvements on this foundation

KataGo: The New Standard​

In 2019, American programmer David Wu released KataGo, which quickly became the most popular open-source Go AI.

KataGo's innovations:

  • More efficient training: Achieves stronger levels with fewer computational resources
  • More features: Supports handicap games, different rules, win rate and score estimation
  • Better human-computer interaction: Provides detailed analysis output
  • Continuous updates: Regularly releases stronger neural network weights

Currently, KataGo has become the de facto Go AI standard, widely used for teaching, training, and research.

The Significance of Open Source AI​

Democratization of AI Tools​

In the AlphaGo era, only Google had top-tier Go AI. Ordinary players could only see AI's performance in news, unable to experience it personally.

Open source AI changed everything:

  • Everyone can use it: Only needs an ordinary computer
  • Free access: No fees required
  • Transparent algorithms: Can understand how AI thinks

This is like going from "only professional photographers can take good photos" to "everyone has a high-quality camera in their pocket."

Revolution in Go Education​

Open source AI brought revolutionary changes to Go education:

Instant feedback

  • For every move, AI can tell you if it's good or bad
  • Can see the gap from "best play"
  • Can explore different variations during review

Personalized learning

  • Choose appropriate AI strength based on your level
  • Targeted training on weaknesses
  • Practice anytime, anywhere

Objective evaluation

  • No longer dependent on subjective evaluations
  • Can quantify your progress
  • Understand the advantages and disadvantages of different moves

AI's Influence on Human Playing Style​

The Joseki Revolution​

After AI appeared, many traditional joseki were proven suboptimal, and new joseki constantly emerged:

The 3-3 Revival

  • Traditional view: Invading 3-3 directly in the opening is "vulgar"
  • AI view: In many situations, 3-3 is the most efficient choice
  • Result: In modern games, opening 3-3 invasions have become normal

Re-evaluation of Komoku Joseki

  • Many joseki passed down for hundreds of years were found not to be optimal
  • New variations were developed
  • Players need to constantly learn new variations

Changes in Whole-Board Concepts​

AI's playing style influenced humans' overall understanding of Go:

Efficiency First

  • AI highly values efficiency of each move
  • Traditional "thick" playing style is considered less efficient
  • Modern players focus more on capturing territory

Not Afraid of Complex Positions

  • AI isn't afraid of calculation, often enters complex fighting
  • Human players also began more actively creating complex positions
  • Intensity of games increased

Precision in Late Game

  • AI's endgame and closing is very precise
  • With AI help, human players' late-game ability significantly improved
  • Half-point, quarter-point wins became more common

New Forms of Creativity​

Some worried AI would kill creativity in Go, but facts proved the opposite:

  • AI showed many moves humans never thought of
  • Players developed new tactics inspired by AI
  • Human-AI "hybrid intelligence" created new possibilities

How Modern Professional Players Use AI Training​

Daily Training​

Modern professional players' training is inseparable from AI:

Game Training

  • Play handicap games against AI
  • Adjust AI strength to match own level
  • Focus on specific position type practice

Review Analysis

  • Use AI to review after every game
  • Find mistakes and best moves
  • Analyze win rate changes of different choices

Joseki Research

  • Use AI to explore new joseki variations
  • Verify if traditional joseki is still valid
  • Develop preparation against specific opponents

Match Preparation​

AI has also changed how professional players prepare for matches:

Opponent Research

  • Use AI to analyze opponent's game records
  • Find opponent's weaknesses and habits
  • Prepare targeted strategies

Opening Preparation

  • Study AI recommended opening moves
  • Prepare variations unfamiliar to opponents
  • Gain advantages in opening phase

Shin Jinseo's Example​

Korean player Shin Jinseo represents the "AI generation" players. He has used AI for training since childhood, his playing style combining human creativity with AI precision.

Shin Jinseo's success shows:

  • Early AI exposure can accelerate growth
  • AI training doesn't replace creativity but enhances it
  • The "AI generation" players' overall level is indeed higher

Human-AI Coexistence​

AI as Tool, Not Opponent​

After the initial shock, the Go world gradually found a way to coexist with AI:

  • AI is a teacher: Teaching us better moves
  • AI is an assistant: Helping us analyze and research
  • AI is a partner: Showing us more possibilities in Go

Games between humans still have their unique value - emotion, psychological warfare, creativity, and things that can't be quantified.

Go's Continued Development​

AI didn't end Go but instead injected new vitality into this ancient game:

Enhanced Viewing Experience

  • Viewers can see AI evaluation in real-time
  • Easier to understand professional players' thinking
  • Increased entertainment value of matches

Lowered Entry Barrier

  • Beginners can get high-quality guidance
  • Learning efficiency greatly improved
  • More people can enjoy the fun of Go

Overall Level Improvement

  • Average level of professional players improved
  • Amateur players can also reach higher levels
  • Go's "ceiling" keeps being pushed higher

Looking to the Future​

Continuous Technological Progress​

KataGo continues to update, each new version stronger than before. Possible future directions:

  • More efficient training methods
  • Better interpretability (telling humans "why" it plays this way)
  • Combination with other AI technologies

New Modes of Human-Machine Collaboration​

In the future, the relationship between humans and AI may further evolve:

  • Human-machine pair Go: Humans team up with AI to play
  • AI-assisted teaching: More intelligent teaching systems
  • Go creation: AI helps design interesting games or puzzles

Continuation of Go Culture​

No matter how technology develops, Go's core values won't change:

  • It is training for thinking
  • It is an aesthetic experience
  • It is a way for people to connect

AI just lets us see this ancient game with new eyes, discovering beauty we never imagined.

"AI is not Go's terminator, but Go's new starting point."

From AlphaGo's shock to KataGo's popularization, Go AI development has let every Go enthusiast enjoy the benefits of technological progress. This isn't a story of humans losing to machines, but a story of humans using tools to transcend themselves.