Cho Chikun vs. Deep Zen Go: 3 Things we learned
chae YG / 2016-11-23
Cho Chikun vs. DeepZen Go: 3 Things we learned
Beomgeun (Evan)Cho 7dan
Cho Chikun took down Deep Zen Go in the 3-gamematch heldNovember 19th, 20th and 23rd, winning thefirst and third games. A great deal of attention was focused on this match by Badukplayers, computer scientists, and the public as well, all wondering if the second AIBaduk program canrepeat victory overhuman beingsinthe game of Baduk inthe wake of Alpha Go, which has astonished the world by defeating Lee Sedolearly this year.
Despite the series victory, Cho struggled in openingstages of all three games, playing too aggressively and testing Zen Go by usinguncommon josekies. However, Cho managed to reverse the first and third games withgreat persistence and finally came up with the 3-Game series victory. Lookingback on the whole series, here are three things that we learned.
Despitethe result, Zen Go showed a huge improvement in a short time
As mentioned earlier, Zen Go dominated theopenings in the three games with great strategies based on wide perspective, andactually finished the Game 2 with an incredible win by killing Cho's bigdragon, which is a very good indicator that Zen Go has significantly improvedsince the 3-stone handicap match with Kobayashi Koich held last March.
Before the shocking outcome of AlphaGo and LeeSedol match, everybody had thought AI was unable to win against human beings inthe game of Baduk at least for the coming decade, due to the infinite complexityand fluidity of Baduk. But now we haveto admit that not only AlphaGo but more AI Baduk programs in professional levelwill come out from all over the world and eventually reach out the top levelalthough it is not the time yet.
AI'sweakness is likely the end game
Contrary to Zen Go's opening skills demonstrated in the whole series,Zen Go's late middle game and end game skills were quite inconsistent anddisappointing. Zen Go was leading in the Game 1, but made a number of mistakes inthe end game stage that were quite basic and got caught by Cho. Also in Game 3,it was very close middle game situation that Zen Go might have been ahead or beenable to make into a half-point game. But a fatal mistake was made by Zen Go andthat cost Zen Go an immediate resignation defeat.
The current number one Baduk player in theworld, Ke Jie commented before this series, via his Weibo:
"Ihave played a number of Deeplearning AI Baduk programs and discussed with my colleaguesin order to research on AI's weakness. I hope that Mr. Cho will make a closesituation in the end game stage or play until the end game regardless how thesituation goes."
Probably,Ke Jie found AI's current weakness in the same way that I did.
A commercialized Go program in professional levelwill be in the market just like Chess
In the AlphaGo and Lee Sedole match, AlphaGoplayed along with 1202 interlocked computers (CPU). During this time, Zen Goonly played with one computer and defeated Cho Chikun who is a living legendBaduk player. This indicates that the AI technology in the game of Baduk hassignificantly improved by the Deep Zen Go Project team along with TokyoUniversity AI Development team even if Cho is less competitive than Lee Sedolnow.
Cho Chikuninterviewed after the Game 3:
"AIhas improved remarkably and human's opening theories and mindset in Baduk mightneed to be reconsidered. Zen Go prefers influence to territory unlike thecurrent trend of Baduk and Zen Go's playing style in center is quite differentthan human's which might actually be superior."
Consideringthe speed that AI is improving at Go, it will be no surprising that acommercialized version of Zen Go or any other AI Baduk programs in professionallevel are ready for customer market sometime soon.