|
This paper by LI Zhe won first prize at the 5th Hangzhou Qi Culture Conference.
There were several papers on AlphaGo (and much to my surprise FineArt hardly got a mention - even DeepZen got more). The organisers also insisted on referring to AG respectfully as either AlphaGo in western letters or as the equivalent of Alpha Weiqi - no dogs allowed!
Li's paper was judged the best, but he is 3-dan pro after all and so his views carry weight. I haven't read the whole paper yet, just listened to the presentation, but what knocked me for six was the AlphaGo selfplay game he mentioned where there were 17 groups (see the game below - the position he chose was after move 177) and only four had a clear life status. To give a perspective on this level of complexity, there is a reliable proverb that tells us "five groups live, the sixth one dies." That applies to both players, of course, so the maximum number of groups on the board in human pro play would normally be 11 - five each and one dead one which probably decides the game. But that means 10 out of 11 would be alive and so wouldn't need to be juggled in the way AG can mesmerise us with.
In the margins I was able to get a range of views on AG (or AI rather) from China and Korea, and from Japan by reading a magazines. Most important, though, I felt, was the summing up of the conference at the prizegiving.
It seems that a kind of despair is the prevailing mood, with black despondency and sheer delight as the two extremes. Korea seems gloomiest, perhaps because they are behind in the AI race. But the fat part of the Bell curve seems populated by people who have already concluded that it is a waste of time trying to compete with the machines and, instead, we should just accept them and see what we can learn. But no-one has much idea at present what we can learn, how we can learn it, or even whether we can learn it. And so analysis like Li Zhe's is the focus of activity everywhere among pros.
I wasn't giving a paper this year - just a talking-points PowerPoint presentation - and I decided to pass on some views on AG from a western perspective. In particular I passed on L19's suggestion that go is a topological game. This was new to most of the Chinese. In fact they borrow the English word, but I was intrigued by the reaction after I explained (crudely) what it meant. A typical non-mathematical westerner would look cross-eyed at you but the Chinese nodded enthusiastically and said, "Ah yes, that's the way we think!" Maybe they really do see go differently from us. Anyway, I embroidered the theme by showing some of the board-space diagrams from the famous New Fuseki Theory book, and I also mentioned my Go Seigen Group theory which got a surprisingly warm reception. Li Zhe at least was kind enough to say I'd given him some new ideas to chew on.
But what had the most impact on me was the summing up by the (ex-)Mayor of Hangzhou Wang Guoping. He is a quite remarkable man. He doesn't actually play go himself but was the driving force behind building the Tianyuan Tower 34-floor skyscraper where (at least on the floors I've stayed on) the facilities are 6-star and every corridor is a go museum, though there is a proper museum and a huge library downstairs. If you put this building at 10 on a scale of 10, I'd rank the Nihon Ki-in as about 3. The mayor attends the conference every year (even though he is now technically retired) and actually stays to listen to some of the papers. As proof that he has read the papers he gives an extempore summing up, about 40-minutes long. In my former job I've had to listen to many politicians giving speeches, and many were brilliant orators. Mayor Wang's conversational style is up with the best. His key point this year was that AlphaGo is giving us an opportunity. We should concentrate less on the competitive aspect in future and instead try to learn from AG how we can teach ordinary players how to improve. The mayor's main point was that China has a rapidly ageing population (the result of the one-child policy) which needs to be cared for. Physically it can get huge benefits from taiji (yoga is also booming there) but weiqi (go) can act as "taiji for the mind." It was left unspoken but I imagine he also meant it was a healthier alternative to mah-jong!
I expect to see this put into practice soon in China. Partly this is because Mr Wang's track record is so impressive but also because go enjoys access to the highest echelons. For example, China's President (assuming he is voted back in later today) was a classmate of Nie Weiping. We may also see something similar in Korea. I gather Cho Hun-hyeon has recently been elected to Parliament. Although he has effectively had to give up go for the time being as members of parliament are not allowed to have paid part-time work (and have to be especially careful about this given the reasons why the latest elections were called), I can't imagine he wouldn't use his interest in go to inform his decisions.
Some people see AI in chess and go as killing the game, and others often valiantly try to defend the games by pointing out that we still run competitive foot races even though cars are faster. I suspect that future discussion about competition in go will see it more like music - new songs continue to be written and singers vie to be No. 1 in the hit parade, but ordinary people will be content to pick and choose their own favourites from past and present, and the market will stay healthy. In go most of us will have a healthy market and healthy minds while pros continue to explore the limits of the game - or just set new fashions.
I mentioned AI discussions in Japanese magazines. There have been quite a lot. One of the most revealing was by O Meien. He pointed out that working out what AG was doing differently was only a small part of the battle. He found it easy to accept that AG's moves were better but he didn't feel comfortable copying them because he didn't understand them. One reason he accepted the superiority of its moves was his personal experience in working on the computer project GoTrend. When he applied his theories expounded in his book Zone Press Park (or something like that - it's been published in English) to the program a significant one-stone improvement was achieved. These theories relate to the centre in the same nebulous sort of way AG favours. The big difference, O asserts, is that GoTrend was just go "software". Nowadays, with the addition of deep learning, we have go "AI".
He believes pros like him can usually guess AG's move as one of their candidate moves (and more so now they have studied its games) but ranking the candidates with AG-like efficiency is beyond them. And even if they do choose the same move it will be for quite different reasons, which means the follow-up will be totally different.
This may seem a touch despairing, but he adds that he plays go because he enjoys it and the reason he enjoys it is that every move has a meaning. His efforts are devoted to finding that meaning: "A game of go is an accumulation of meanings" AG has expanded the horizons, and so he welcomes it.
I'm with Li Zhe below. The kilt is because I kept a promise to being it from last year.
| Attachments: |
LiZhe+Me_2.jpg [ 35.02 KiB | Viewed 6435 times ]
|
|
Login
Register
FAQ
Search
tchan001
to keep the game interesting.
.