On the conclusions Jjon Diamond doesn't go very deep in the reasons why he thinks Lee Sedol will defeat AlphaGo.
Myungwan Kim points the following weaknesses and strengths in AlphaGo:
AlphaGo's strengths:
- It's not afraid of Ko.
- Reading might be AlphaGo's strength.
AlphaGo's weaknesses:
- Doesn't understand sente, aji and gote [advanced concepts of the game]
- At times too obsessed with following common patterns, when the specific situation might require creative deviation from those patterns..
- AlphaGo has difficulty, or even doesn't at all, evaluating the value of specific stones. It's good at making moves which directly gain territory for itself, but tends to miss moves which reduce the value of the opponent's stones.
-It can make really high level moves at times, but it doesn't understand those moves. Which it displays by making the right moves at the wrong time.
I dismiss Kim's comments on the insularity of AlphaGo's learning method (playing millions of times against itself) as that method includes records from a massive database of past 6 dan+ masters.
Of course those conclusions, both by Jon diamond (he's on Google team's payroll after all :-) ) and Myungwan Kim (an "expert" with all the goods and the bads that come with this condition) must be viewed with some skepticism but both point in the same direction: AlphsGo may not be ready to win against a world champion.
The only doubt is whether alphaGo can learn until match day how to avoid the pitfalls reported above.
I'm a big fan of Kaizen. Have you read about what Toyota did with a NYC nonprofit? Instead of continuing to give them money, they sent in staff to implement Kaizen methods on their operations which improved significantly their services and reduced their expenses.
Upon reading the following articles, i am turning 20 to yes.
https://www.reddit.com/r/baduk/comments/43g2jl/synopsis_of_myungwan_kims_analysis_of_fan_hui_vs/ (an analysis by Myungwan Kim a 9 dan American professional )
https://britgo.org/files/2016/deepmind/BGJ174-AlphaGo.pdf (an analysis by a jon Diamond, the British amateur that served as ref on the Fan Hui - AlphaGo match)
On the conclusions Jjon Diamond doesn't go very deep in the reasons why he thinks Lee Sedol will defeat AlphaGo.
Myungwan Kim points the following weaknesses and strengths in AlphaGo:
AlphaGo's strengths:
- It's not afraid of Ko.
- Reading might be AlphaGo's strength.
AlphaGo's weaknesses:
- Doesn't understand sente, aji and gote [advanced concepts of the game]
- At times too obsessed with following common patterns, when the specific situation might require creative deviation from those patterns..
- AlphaGo has difficulty, or even doesn't at all, evaluating the value of specific stones. It's good at making moves which directly gain territory for itself, but tends to miss moves which reduce the value of the opponent's stones.
-It can make really high level moves at times, but it doesn't understand those moves. Which it displays by making the right moves at the wrong time.
I dismiss Kim's comments on the insularity of AlphaGo's learning method (playing millions of times against itself) as that method includes records from a massive database of past 6 dan+ masters.
Of course those conclusions, both by Jon diamond (he's on Google team's payroll after all :-) ) and Myungwan Kim (an "expert" with all the goods and the bads that come with this condition) must be viewed with some skepticism but both point in the same direction: AlphsGo may not be ready to win against a world champion.
The only doubt is whether alphaGo can learn until match day how to avoid the pitfalls reported above.
Great analysis. Thanks.
Glad you liked it. Feel free to suggest changes.
I practice what Japanese production engineers call Kaizen, continuous improvement
I'm a big fan of Kaizen. Have you read about what Toyota did with a NYC nonprofit? Instead of continuing to give them money, they sent in staff to implement Kaizen methods on their operations which improved significantly their services and reduced their expenses.
Comment deleted on Sep 05, 2025 10:42AM UTC