Impact of Go AI on the professional Go world
medium.com
medium.com
1. Professional gameplay changes a lot.
There are no more secret openings/midgames/endgames, you will hardly be able to cheese a top pro (not that it was easy before). Professional gameplay requires memorizing tons of lines and then adding on to add it. There are many chess openings which require memorization till the end result (50 moves) or you risk losing the game. We are going to see a similar trend of memorization in Go.
2. Online casual gameplay changes a lot.
There are lot of cheaters in online forums. Cheat detection has to be built in to the sites or you risk being taken over by cheaters. Easy availability to best moves allows the dedicated player to learn a lot faster than previously where there was a discoverability issue. We see this in the professional chess world where its not uncommon to see 14-16 yr old GMs nowadays.
3. Professional teaching still survives.
Engine can tell you the best move but it doesn't give you a teaching path from beginner to amateur professional to grand master. I am not a master but based on anecdotal evidence people still employ coaches but for much more focussed tasks and more specialized skills (e.g. openings, midgames, endgames).
100k hours is 11.4 years straight.
Just sayin’...
The AI favored lines are likely to get heavily entrenched. I wouldn't expect them to go further because they generally end when becomes better to play elsewhere. There are a few fighting joseki where this isn't the case, but AI doesn't seem to like them much, although there are a few exceptions.
Do we know if those moves are actually weaker overall, or is there some other sort of selection bias at play?
But if I were guessing, the AI probably doesn't like fighting joseki because it knows how the fight will end. AIs are much, much better than humans at working out who won a fight.
Short time controls and live streaming of blitz games by GMs is fun. Maybe this is less “pure” but it’s great entertainment, and doesn’t seem like engine use is a huge problem when the stakes are low (it’s usually pretty obvious to a GM when they start losing to weird moves, they call it out and move on).
As a mid level chump at chess.com, any cheater will be ranked way over me.
It's of course possible that some people only cheats occasionally, for whatever reason. People are weird and do weird things...
Afaik they analyze the deviation (in centipawns) of each played move to the optimal move calculated by popular chess engines (like Stockfish, Houdini, Komodo, etc); and aggregate in a sort of weighted Brier score.
If a player is constantly making optimal moves, they're likely cheating (they can analyze top level professional matches for upper skill benchmarks). There is some lower weighting for opening moves (which is where the database part comes in just knowing when a game moves from theoretical 'book' openings into original territory).
Each of the sites have different implementations of this: chess24 for example even partially gives your accuracy of book moves and then differentiates between Blunder / Mistake / Good / Excellent / Best move based on the deviation from optimal; but it doesn't show you how it's aggregated into a single value.
What is the incentive to do any artistic endeavor, if you are not planning on being one of the world's best pianists, or painters, or singers, or furniture-makers?
Go was probably already solved by some distant species on a planet far away. Did that make it any less enjoyable for us to play? What's the difference?
If I do it to make a living, that’s one thing. If I do it to be “the best,” that’s another thing. If I do it for the pleasure of learning to do a thing, that’s a third thing.
It’s the same with games, but the emphasis on winning skews the culture towards wanting to be “the best.” But for those following the third path, AI beating humans changes nothing for them, just as synth bass and sequencing changes nothing for me.
If suddenly computers can create perfect, beautiful music, I will still play an instrument.
I didn't downvote you because I guess it's a fair question to ask, but it is a rather naive one. Why learn arithmetic if a computer can do it?
Regarding the impact on pro-level teaching:
“The demand for pro-level teaching games and private lessons has plummeted. Professional players used to command a high price for teaching games and lessons, and this has been a critical source of income for many pros.”
I think the community would also be interested in articles which discuss the impact of AI on professionals in other fields, obviously chess, but also more distant ones such as e-sports and language translation.
——
On language, I would guess that the demand for non-certified translation might plummet between many major languages and English since automatic translation of documents is now quite close to human level when there is sufficient training data.
Down the line, might this reduce the demand for advanced foreign language skills and thus change the career paths of language majors as well?
Many players I know learned how to play pretty well on 9x9 by playing a computer program.
I'd argue you can even learn much faster playing a program, because for high level players many sequences were based purely on memorization (including point values) but now seeing different results quickly you can understand it better.
And the level where your play is based on high level abstractions like influence and group strength is not that hard to reach and it wouldn't take much to reason them out just from the games.
That said, it still may be worth having a teacher/trainer I guess. To motivate you, give you that dopamine in person and all that jazz.
I would say it's more like learning to code yourself rather than at the university than learning assembly from gcc.
From the article: "the Go school recruits a professional player, who would agree to play a fixed number of games, like five games, one per week, at a certain price.... Of course there is still room for lower level classes and teaching, but pros are often better at playing teaching games than explaining easy concepts."
I think ai will prove "good enough" for most applications in the near future, but this won't eliminate human translation (or journalism, or copywriting, or whatever you might want to apply similar technology/techniques to). but it may well eliminate the bottom 80%
> Lee seems to combine these two with the silent assumption that the professionals’ goal is to strive for the ‘highest possible level of go’, which is no longer possible because the ai cannot be beat; incidentally, Lee Sedol remarked something similar when he announced his retirement. To me, on the other hand, it seems like nothing has changed because I have always reached for the ‘highest possible personal level of go’, and this should be the same for most players who are not near the top of the world.
Perhaps it has less to do with being at "the top of the world" and more to do with the role of the game in the broader culture. If what attracts you is not just your personal drive to improve, but the sense of overall societal contribution, that will surely wane if training goes towards study of AI rather than active research in play. On the other hand in the US/Europe such a thing hardly exists at all even for professional players.
Maybe a comparison point: What happened to voice and instrument lessons once recorded music became commonplace? In the fairly recent past any social group that wanted to listen to decent music must have had a member who was at least a slightly-above-average musician; and once they didn't need that anymore, that skill may have significantly dropped in social standing. That doesn't mean no one takes lessons anymore, but it seems like the number would have greatly reduced. (On the other hand, the median interest level and drive among people taking lessons probably increased.)
(Does that mean the go world is going to have to find its equivalent of Kurt Cobain to get lessons started again?)
I believe they are saying that Cobain's work (and the popularity of Grunge generally) brought new life to the Rock music scene, leading to an uptick in garage bands forming (and therefore demand for music lessons?) as a second order effect.
I am a double digit kyu player, but I enjoyed clicking through the most recent game with it: http://www.yss-aya.com/cgos/viewer.cgi?19x19/SGF/2020/08/21/...
The battle in the upper left was fascinating!
Ironically, the AI-Explainability problem is, at least for now, an asset. The model can't explain its strategy, only that it judges some positions stronger than others based on a huge amount of experience.
That reduction of aggregate experience to comprehensible language is still (mostly) in the realm of humans and so should provide a bridge for talented, insightful teachers to teach the next generation.
It's actually the same as Douglas Adams'. Now we know the answer is 42 and now we need to solve what the question really was :)
I think it's ironic how he foresaw the problem we are experiencing with AI right now.
For strategic/positional issues, it is not nearly as good a teacher.
Obviously it's mostly pattern recognition, but that pattern recognition is partitioning parts of the board into classifications of what those sections are doing and whether to prioritise them or let it go in favour of something else, the immediate decisions within those sections are fairly logical. The more complex broader pattern recognition is how those sections will interact as they approach each other, which if there is that much inventing explanations where I imagine it comes in.
My biggest complaint is that the jargon is just out of control. For whatever reason, instead of localizing go jargon like the chess community has done, the Go community just uses native japanese terms for everything. This makes it super confusing as a beginner and you basically have to have the wiki[0] open all the time if you have any hope of understanding what people are saying.
For example, when reading a beginner's article:
"In this case you can't move here because of ko, so you would instead look for other pieces that are atari or close to atari - just be careful of seki"
Would a bit of localization killed the go community?
"In this case you can't move here because of the repetition rule, so you would instead look for other groups of pieces that can be captured or are close to being captured - just be careful not to get in a capture-stalemate scenario"
My second biggest complaint is that scoring is not straightforward at all and there are 2 rule sets (Chinese, Japanese). Which ruleset should I use? Are there any subtle strategy changes you need to make when switching between rulesets? When playing I have no idea who is currently winning if it looks close (i.e. roughly equal white/black areas) and I basically just wait for the computer to announce the winner at the end.
Anyway, just venting about my frustrations learning Go. Chess, by comparison, was much easier to learn. I didn't feel swamped with jargon and it was much easier for me to grasp which side was winning (though occasionally good chess players can pull a checkmate out of a seeming disadvantaged situation that I didn't see coming!)
Yah, we never say Zugzwang or J'adoube or Desperado or En Passant or Fianchetto or Zwischenzug.
There's not so many foreign words to learn to understand Go compared to chess. For lots of the fundamental Go words, they'd need a new English word coined anyway or an imperfect metaphor.
This. Even the chess words that are anglicised have special meanings in chess. Being a native speaker of English doesn't help you understand what a "Bishop" is or what is "Checkmate" (OK, that's an anglicised Persian word) or the difference between a "Pin" and a "Skewer".
Also, most of the Japanese terms don't have a meaning outside Go, so it's not like Japanese people get a leg up there.
Japanese counting is more popular but you really don't need to worry about it until some high dan level if you play on 19x19. It takes a long, long time before you will be able to say confidently who is winning on 19x19 if the game is close.
I suck at Chess, but to me Go was much easier to learn. You can have quite a lot of fun without studying any openings.
I don't see how this is true unless you are memorizing the wiki like a dictionary. I can only think of 9 words that I use commonly in games (sente, gote, seki, ko, atari, hane, aji, tenuki, komi). Sure, I think the English Go community could try (and probably has tried) to reinvent translations for each of these, but they probably wouldn't catch on. "Seki" is easier to say than "mutual life", "sente" is more precise and easier to say than "initiative", "ko" is just ko.
I just don't feel like learning new words is any more difficult than keeping track of algebraic notation in chess, and I think it's just part of the commentating culture of both games.
> Which ruleset should I use?
As a beginner, you shouldn't worry about which one you use. They're almost the same except on some edge cases.
Uhhh... give this Japanese page a looksie: https://docs.oracle.com/javase/jp/8/docs/api/index.html
I think we got the better trade. Just my opinion though. Games which are more popular in the far-east will naturally use far-east terminology.
Similarly, things Americans do will get American-terms associated with them (IE: Programming Java). Its the nature of international trade. Even a Japanese-centric language, like Ruby, is largely composed of English terms to continue to appeal to the greater programming community. (Japanese programmers would learn "for" and "while" loops anyway, despite it being foreign terms).
Semaphores are "P" and "V" canonically meaning something in Dutch. Someone tried to explain it to me once, but you pretty much have to learn Dutch for the meaning of it all.
------------
Anyway, its the nature of working internationally. Gotta learn tidbits of the other people's language to really understand something.
"P" stands for "Prolaag", which is a word Dijkstra made up. Both Raise (Verhoog) and lower (Verlaag) start with a V in dutch. So knowing dutch doesn't really help with that one.
About localization, I started playing in a go club and the people I played with told me the Japanese names and their meaning. I already knew all the words when I started reading books. I understand that if you started reading books and watching videos from day one, that's a problem.
And about learning Go, the very first months are a very dense fog. Somebody never emerge from the fog.
The main difference that Japanese rules make on scoring is, essentially, codifying a matter of etiquette directly into the scoring: they aggressively penalize extending the game with unnecessary moves, which to be sure has some distortionary effects, but again, they don't essentially change what people considered the meat of the game; you'd play 90% the same game regardless of counting method, with differences cropping up basically once a player starts passing. This has far more ... pedagogical value outside the game, than the impact it has on the game itself.
By far the most important rules in a ruleset, in terms of how they affect core gameplay, are the repetition rules, and even the choice of suicide rule has a more substantial impact than the differences between scoring systems.
Free to watch on YouTube: https://www.youtube.com/watch?v=WXuK6gekU1Y
If it seems ridiculous to you, that's a good signal that you'd get a lot out of digging into philosophy of mind.
Where do our thoughts come from? Creativity? Is it so preposterous that these too arise from probabilistic electrical storms - but in wet matter as opposed to hard?
Where thoughts come from may be too complex for us to model now, maybe ever. But if you look deep enough, I think you'll find at heart deterministic processes driven by networks that were trained through nature + nurture.
Maybe creativity is less ethereal than we perceive. Maybe it would benefit us to start considering how non-sentient intelligence can achieve it.
Reminds me of a quip about Richard Feynman that he'd spend many hours honing a new way of doing a physics problem, but carefully present it as if he just did it off handed. So we just need to link human brains with cybernetic implants to compete. ;)
Of course I still made something up.
Furthermore, just because we don’t have the computational capability yet to replicate human thought processes using, for example probabilistic search, does not mean that probabilistic search is not part of replicating human thought processes.
There are constraints and limits and rules: rhyme scheme, meter, key, chord progression, genre, structure, and whether or not those things will support the consistency and coherence of the content... that's barely scratching the surface. The "choices" and the "thinking" are highly constrained. Few things are going to work. It's often either "this way" or "wo, back it up".
Even in something like Jazz there are still intricate rules at play and the true greats were eccentrics and/or taking loads of drugs in order to push the boundaries where they could.
[1] And then that one time where I finally cracked the melody at 3am, only to realize in the morning that it was Neil Young's "Heart of Gold". Even my creation was not my creation!
I opened this thread with absolutely no thought about the _game_ Go, and read the top comment about how chess changed in the last 10-20 years thinking it was tangentially related because of Watson and whatnot playing AI chess.
I came here thinking I was going to read comments about how the Go language has been impacted with the introduction of ML and AI packages that attract data science types from Python and R. I thought it would be interesting to see how a newer/smaller language was impacted by a group co-opting the the development and platform.
Whoops.
func PredictNextCodingMove(ctx context.Context, source string, cursor int) (string, error) {
return "if err != nil {", nil
}Nowadays we just make a side-bet about what XG mobile will think about the move.
In the end, there were still drummers, and it is generally recognized that live drumming is, aesthetically speaking, a different thing than sequenced, sampled drum tracks.
I've never heard a drum machine that can truly capture the sound and feel of John Bonham, Keith Moon, Stuart Copeland, or Ringo Starr. It's just a different thing.
The only way to establish that would be through blind listening tests.