Master(P)

5 posts / 0 new
Last post
Bloemen
Master(P)

Recently a very strong computer program beat top go professionals like Ke Jie and Park Junghwan in about 50 speedy games at go playing internet sites. It is clear now that this program is a new beta testing version of AlphaGo.

I replayed some of these games, they are totally awesome. To me, the first few moves are quite normal, but then some of the moves are so unexpected, and difficult to understand: what do they accomplish, why? It seems that all kinds of strategic concepts are not valid anymore, I wouldn't be surprised if the only valid answer to the "why" is "because it wins". So I guess the next few months or years will be a great adventure to find out how good go looks like, kicking out not valid strategic concepts (which ones?), substituting them by new ones (which ones?), where I think that tactical reading is paramount. According to Ke Jie the basics of go have to be rewritten: quite amazing. I guess moves that were considered sente are not sente anymore, that miai, swaps, probes are what the best go is about, making the games extremely complicated for the humans, but not so for the computers.

I wonder what is going to happen to the games recently played, and the ones that are going to be played, between humans and computers, and by computers against computers. Is go4go going to show them? They may not be part of ordinary tournaments, as they are played on the internet sites, but they are very interesting, they are building the future of go.

I wonder what the opinions are of those visiting go4go, and of go4go: we are living in exciting times when looking at go. Thank you all for your reaction!

Kind regards,
Paul

Jimsquint
What is the strength of

What is the strength of Master P? 13 Dan, 15 Dan? Also, what is the strength of perfection in Go?

Bloemen
It is very, very strong, and improving

Quite probably the next official version will not lose to a human player anymore. The most interesting part of it is that it may show new ways of playing, breaking the conceptual walls we erected for ourselves. That is, we will see new go, new ideas if you may call them so. If the only criterion for a move is, "because it delivers the best result", you cannot pinpoint a strategical idea to it, it may just not be there. That is, strategical ideas are shortcuts for human beings that cannot read it out sufficiently. They may be correct the one time, and wrong the other time, you could call it luck when it happens to work out fine. Still inverting the process, trying to extract ideas from the new moves we see may help us humans to play better. I really hope to see a clear set of new ideas, however difficult, and equally the abolishment of wrong ideas, when the computer signals that some human moves are inferior to other ones. The set of abolished and correct ideas would make for a great book. Analyzing existing human games in search for bad moves that we didn't consider as such could be very enlightening too. We are in for a great few years!

Kind regards,
Paul

AnthonyC
Not entirely sure or

Not entirely sure or convinced that humans can truly mimic AI moves that are already performing at beyond human level, given those moves only made sense (to the AI) after one computes/evaluates the position at a level that is also beyond human.

Probably only the very top Go players like Ke Jie or Park Junghwan can mimic those moves.

Bloemen
New moves and mistakes

When the computers have played a few thousand games against top professionals, I think the type of play of these computers will become quite clear to the professionals. The style of play may be style-less, that is, by searching for the most promising move, it will not be guided, hampered by subconscious preferences like feeling OK with a territory oriented style or a cosmic style. The thin line between sente and gote will become a bit clearer, as will be the successful use of probes, the moment to abandon stones. There may be no fundamental surprises anymore, like the fifth line shoulder hit, all of these kind of moves will have been played. Thus, the kind of play of the computer will become quite clear, people will try to play the same way: we saw that with chess too.

The real difference will be that the human professionals will make, small, mistakes more often than the computer, and the computer will be able to punish these mistakes quite more mercilessly than the professionals. If the computer is able to get rid of "winning by one point is enough" it may win most games by more than five points no doubt, more often we will see bigger wins, like because a professional group gets captured against all human odds, or because swaps always will be in favour of the computer.

Thankfully the possibilities in go are so vast that, while the results will be boringly one-sided, the games will not be boring at all, there will be much to learn from each and every one of them. I am quite sure that in many positions there are a few equally promising moves, thus the variety in games will be virtually endless. To start with the first move: I don't think the computer will ever be able to convincingly state that the 4-3 move is more promising than the 4-4 move. What I hope is that the computer finds no real difference with some other moves to start the game, and will play them. The same goes for the type of enclosures, extentions, jumps to the centre: by showing us this the computer may enrich the game of go in a totally overwhelming way. In fact, this kind of happened in chess too, where innocent looking, obscure openings saw the light of day at top level, encouraged by the human computer counterpart, world champion Carlsen.

I am looking forward to what is going to happen!

Kind regards,
Paul