AlphaGo: Teach your stones to dance!!
HTML-код
- Опубликовано: 18 сен 2024
- Get rewards for support on Patreon!
/ insente
Check out my website: www.goinsente.com
Catch me on Twitch!
galluccinator_insente
Twitter: @in_sente
The video Clip I used for Mohammad Ali was from here:
• Muhammad Ali dodges 21...
+In Sente
OH MY GOD! Did you have to use the Hikaru No Go music? Now I got all emotional and want to see the anime again... For the 8th time!
Excellent video. I felt I learn more.
go seigen used to shoulder hit enclosures and now alpha go uses the tactic. it makes sense in the context of the five phases of the Game. the first being enclosures/approaches and the shoulder hit is like an approach move
What happened to the anime Hikaru no Go, and where did I find the sequel to the Winter Cup?
Honestly, I really dislike those hyperbolic statements that AlphaGo is inspiring.
"Every school of though challenged!", "Humanity knows nothing about playing Go after 3000 years!", when there is no real evidence to that fact in my opinion. AlphaGo isn't revealing some great unknown mysteries, or tactics to us, it is just really good at playing go, specifically at assessing the security of its group.
Take for example AlphaGo's play against Lee Donghun (professional 8th Dan), you know why Lee lost? Not because AlphaGo knows some secrets about go unobtainable by us mere mortals, it is because it judged 2-3 attacks on Donghuns end to be weak enough that it can ignore them for the time being, allowing it to make huge sente carrying moves disturbing Lee's group and limiting his territory.
Yes, a great achievement, but not some unknown mystery how it achieved that, give a professional a week after every move and a 1000 blank Kifu's so he can play out all possibilities, and he will get the same result AlphaGo gets. Because this is what ALphaGo is doing, in its allocated time it plays out a 1000 scenarios and assess the outcome, choosing if it needs to response or not, and if not grabs Sente and makes big moves for free.
But this is all an effect of it being a machine, it can do thousand of calculations per millisecond, it doesn't have emotions playing with it judgements, if the maths show it that it doesn't need to defend yet, there won't be a nagging voice in the back of its head, it will simply play. But this is hardware we are talking about, no great Go Mysteries from above, Blobs of Fat loosing to Blocks of Silicon.
I can see where you're coming from, but you're sort of making a strawman argument out of the whole situation. No one thinks there's some cosmic truth that is unattainable by humans, it's just that AlphaGo is showing us a drastically different style than contemporary play, and that's inspiring and would have never happened without AI. Go is like martial arts, or painting, or music. You can take as much or as little inspiration out of it as you wish :)
Ke JIe (professional 9th Dan) said: “After humanity spent thousands of years improving our tactics, computers tell us that humans are completely wrong. I would go as far as to say not a single human has touched the edge of the truth of Go."
That's the kind of hyperbole I'm talking about. Will AlphaGo be useful? Of course, I myself can't wait when I'll be able to get tutoring go plays from a 12th Dan on my laptop whenever I want to. Not to mention that it will be incredible useful in analysis, after all it can do a job a professional needs a week for in seconds.
But I resent that sentiment what it is somehow magical, divine in its abilities, which is the sentiment I see in the go community. Its Blocks of Silicon being better at calculating then Blobs of Fat.
If we're defining 'truth' as the ability to achieve maximum possible efficiency, which is the essence of this game (so I think that's a reasonable definition), then Ke Jie's absolutely correct. Sure he was using romantic language, but no one actually believes there's anything magical going on lol. The professional go players aren't idiots that think there's a magic wizard in the computer, they're just inspired to improve. Us' blobs of fat' will likely keep geeking out to our hearts' content. XD That's probably the last I'll say about it, but I'd suggest trying to not be annoyed at the excitement, cause it's probably not gonna go away for a while hahaha
"But this is all an effect of it being a machine, it can do thousand of calculations per millisecond"
Totally, man. The _only_ reason AlphaGo is so much better than everyone is because it's a computer; that's why computers were beating humans years ago, even before AlphaGo was created, right? Not like AlphaGo was created using novel techniques or anything; nope, all DeepMind did was throw more computing power at the problem, and voila, AlphaGo started beating everybody.
Great logic there, buddy.
What's new about AlphaGo are its two NeuralNetworks, the Policy Network and the Value Network.
The Policy Network chooses "interesting spots" on the board, this allows to limit the search tree to a few places on the board instead of checking all 361 spots a stone can be placed.
The Value Network limits the depth of the search to 15-20 moves, again limiting the amount of hardware AlphaGo will use to analyse a possible branch.
This are the two new things about AlphaGo and in both this areas it is about as strong as an average player. Where AlphaGo is amazing at, and gives it the edge over professionals is that in its allocated time it can brute force with perfect recall all the limited options those two neural networks created and choose the best play.
What's new in AlphaGo makes it as strong as average players, what brings it above all professionals is its non-human hardware that can make thousands of calculations per miliseconds. So my initial statement is true, although I admit it needed more explanation what I meant by it.
After looking through several reviews of AlphaGo's games I feel like the lessons people take depend on the games they choose and, ultimately, their prior preferences.
I think you are correct, and it goes further; I suspect that at this high level of art the same person could review the same game twice and notice totally different things. That is why I liked this review, not because it says anything about AlphaGo, but about the point of view of the reviewer.
1st game: The "M4" Cut is not a good move , Because your corner's group is so strong ,If you catch the balck one stone on "M4" move,It will make your strong group more strong, It's waste it, It's redundant construction your corner.So alphago move is the right move or right cuting.
In your later game, You should be in this way to think about the problem:Did I repeat construction? Did I make my strong group too strong ? If you do that way ,That means your stone's performance will be very low.
31:22 whaaaaaaat? :D
People in China dub AlphaGo "Doggie" because the English pronunciation of Go sounds like dog in Chinese. Though very strong when playing Go, AlphaGo is still a preliminary AI thus the animal character suit it well. So how about calling AlphaGo "it".
Preliminary AI? I heard that it was a general purpose convoluted neural network. It's certainly no Jellyfish #backgammon
Reckless Roges maybe you mean Convolutional Neural Network (CNN). but what makes AlphaGo work so well is a technique called Reinforcement Learning.
Trust me, researchers still have a long way to go before they can rouse consciousness within circuits.
InSente where are you from and what rank are you?
Hello! I'm from America and I'm currently 2 Dan on Tygem
Despite the fact that you are not very strong (no offense), this was really nicely done and useful even for me even though I am currently 6d on tygem and 4d kgs. Awesome video and good points about its play. If you look at it closely, it is following basic common sense alot of the time, and its moves are logical. Sometimes its logic is just beyond what humans consider, for example the shoulder hits. They have been played by humans before, but never really investigated in depth. Now, according to haylees video about korean pros studying alphago like crazy, pros are starting to expand their horizons somewhat by learning from the alphago moves, but they aren't revamping the fundamentals.
It's not that alphago is revolutionizing go and telling us everything we thought we knew was wrong. It isn't invalidating our ways of playing. It's more like it is confirming most of our go theory, showing new ways and also in some cases bringing back old moves that were thought to be sub-optimal.
It is sort of like the shin-fuseki era where it was shown that 4-4 and heavy influence was also viable, and that the 4th and center was important too. Alphago is a bit similar in that it sometimes points out that center is more important than humans thought, and sometimes also that territory is more important (in cases where it invades the 3-3 early). But overall, it isn't telling us that the way we have been playing since shin-fuseki is fundamentally wrong or flawed, like old go was. What I mean is that old go was very slow, and flat/low. Another thing I like is that alphago's game is more balanced than most modern pros. Old Japanese honte styles after the shin-fuseki era had many good points, like keeping shape and so on. Alphago seems to agree that that way of playing is correct in many instances. But it also partly validates modern chinese/korean fighting styles, because it sometimes pushes for a wall and then pincers a pincer and it likes to fight in certain situations too.
It's sort of like Alphago's horizon encompasses more than a single style. It sort of brings it all together in harmony/balance and plays very flexibly. That may lead to another style shift in the future similar to what happened in the 90's where Korean/Chinese early fighting styles started to emerge and then become the norm, replacing the more honte oriented japanese style from the 70's and 80's.
haha hey, none taken XD. On the contrary, I'm overjoyed one of my videos can be worthwhile for stronger players! I started playing 1.5 years ago, and made this channel to originally be a vlog of my progress. Eventually I dedicated it to beginner/intermediate players as I improved.
hey I'm from America to and trying to learn how to play
2.5 years ago. Nice haircut. Commanding voice. Two good game analyses. In my opinion, your best video.