In the Age of Google DeepMind, Do the Young Go Prodigies of Asia Have a Future?
newyorker.comI mean, that's like asking whether marathon runners have a future now that we've got cars. Exercising one's brain/having pleasure through the game could be an end in itself, and is independent of how machines perform.
Take it from someone that walked the path of becoming a professional Go player, being a professional and an amateur are completely different attitudes towards the game. Tic Tac Toe is solved, but can be fun to play when you are a kid. Amateur Go playing can still exist, but the goal of strength is more instilled in the path to pro-ship.
This really suggests that going the path of being the strongest is no longer sensical. Why would a human try to be the best calculator in the world, knowing it will never beat any calculator ever? Just to prove itself to other human caculator wannabes? Senseless.
This is a real paradigm shift and we still need to understand what to do. But obliviously ignore AlphaGo is akin should be unfathomable for a professional aspiring player.
As a professional, the first question to ask is what will AlphaGo bring to Go Theory. We still dont know how much stronger it is than Lee Sedol (or how far it is from "God"). Pushing it to its limits will show us insights we havent found yet and we will update ourselves as players to the most current theory.
The second step is answering the following question: Can human + AlphaGo beat Alpha Go? A human potentiated with AlphaGo's reading power can intuitively pick variations that would give it an edge? If so, we have found that Go still harbors a human secret that is jsut overly compensated by reading.
The last step would be, even if human participation gives negligible results, can human + Alpha Go create better games than Alpha Go?
And yet in chess the Carlsens, Nakamuras, Caruanas, Svidlers etc of the world still compete while knowing they have no chance of beating StockFish running on a modern desktop.
The super GMs of the world - and basically all of the chess loving public with them - seem to have acknowledged it and moved on; why would such a transition be impossible in Go?
It might happen, but I hope it doesn't.
I cannot speak for Chess's mindset but as a Devoted Go Player, we are collectively trying to solve it, and so we have for centuries. We play Go to explore its universe and reach utmost understanding of the game(and a glimpse of ourselves). If we ever find(which eventually we shuold) the exact single pattern that is best for both players, and we solve the game, it becomes something different. Maybe it becomes something senseless, or something artful(I can explain the 'art' part if someone asks) but trying to be competitive is silly.
If I devoted my life to Go today, I would not aim at becoming better competing, I'd have to aim at a more effective way to solve the game. Competing was the only thing we had to figure out what was best, but now we can have a companion that will prevent us from faulty variations and logic, and give us instant validation. We can discover more fuseki with a focus group and AlphaGo in a month than in a decade of tournaments.
Competing for the sake of competing is a petty goal.
> Competing for the sake of competing is a petty goal.
This makes me much more angry than it probably should. Who are you to decide what is or what is not a petty goal? People will always do what they enjoy, and if they enjoy competition then that is what they will do. There is nothing wrong or petty about that.
Its a petty goal because the only thing you want to do is be better than the others. In the end, only 1 person is right, and the only consolation is that others are worse for you. Its also not representative of Go, you can be competitive in eating sausages. If your only reason to play Go is to compete, your contribution to the Go community is marginal
With respect to Checkers (specifically, the 8x8 English Draughts variant), it's been a solved game as of last decade, and people still play in tournaments. I don't think it's that silly, since there are many techniques we've automated away, yet still practice by hand.
"[...]we are collectively trying to solve it, and so we have for centuries. We play Go to explore its universe and reach utmost understanding of the game(and a glimpse of ourselves)."
Is there a literature available in English that explores this attitude/practice? It sounds most interesting. You seem to derive significant meaning from the game and I'm interested in how groups of people find meaning in collective activities.
This is more of an insight you get about Go Life, not something we "All know and agree of". Hikaru No Go was a manga that talked a bit about the purpose of playing Go, about being connected with Go players since time immemorial, trying to get perfect play, and failing but getting better each generation.
This is a professional or go devotee mindset, not a typical amateur player that plays without the intent of changing Go Theory or making a legacy.
Why would you do so? The computer will advance much, much quicker than you will because it doesn't have to memorize or be inefficiently trained to be good at Go. It will be able to solve things better than you will, and one day the game will be solved and truly pointless.
We might get much quicker to the solution of Go Working with the computers as Go players and scientists, than only scientists.
Could you explain to me (someone with a very naive superficial understanding of Go) how Go relates to exploring and understanding the universe?
Anything that Go has to go through as a result of AlphaGo, Chess has already gone through with Stockfish and its predecessors (such as Deep Blue, though I realize it's not exactly the same).
Is there a particular reason why a chess computer would be any more undefeatable than a Go computer? Even though Kasparov lost, Nakamura destroyed Rybka 10 years later. Now that we have a competitive Go AI, isn't it likely the game of Go will shift and be even more competitive since now more players can get world-class practice and suggestions on their own?
If I understand AlphaGo correctly, I don't think that any human in the future will be able to beat the AlphaGo today. AlphaGo didn't beat Lee Sedol because it played new and marvelous moves we need to understand.
It played better because it knew the exact consequences of the options it was presented, and could calculate it and make better decisions than human intuition. No human can develop that reading power, and its not reasonable to think a human in the future will have intuition that beats the calculation of AlphaGo.
Since reading, the core ability of Go can now be completely replaced by a computer, the question is what others decisions can a player make. Can he make strategic decisions better than AlphaGo? Can intuition still best AlphaGo calculating capacity?
Eventually, we can think that we will have computational power to actually solve Go, and if there is any sense at all to play Go after that, its about finding those beautiful games, from beginning to end, that provoke emotions and turn Go purely into art.
Nakamura may have beaten Rybka, but even with Rybka at his side, he could not stand against a handicapped Stockfish (https://www.chess.com/news/stockfish-outlasts-nakamura-3634).
Eventually everything we do will be done better by technology. Why should anyone do anything? I think the experience is worthwhile in and of itself and I look forward to what AI can teach us about the game.
Hardest heavy weight boxer is a lot weaker than machine. The fastest baseball pitcher cannot throw faster than a throwing machine.
I like that analogy a lot! It's not like "playing Go really well" ever solved any real-world problem, it has always been a competition among humans. What should change, there?
If you like it a lot, let's make sure we credit Michael Redmond who said it last night during the Go match: https://youtu.be/qUAmTYHEyM8?t=3713
I feel like the author completely misunderstood why AlphaGo is exciting. It's about advancements in technology and soft AI, Go has almost nothing to do with it.
Well, we're on hackernews, we like the tech stuff. For most of the world, this news is where people learn about Go and the human player's "humiliation" (whether or not the Go community sees it that way) is the story.
I'd argue, though, that this is merely a proof of deep learning being able to solve "hard" problems. There's headlines everywhere about deep learning solving previously "impossible" image recognition problems, for example. Fields that are much more interesting and relevant in their real-world impact. AlphaGo, in comparison, seems like a PR/pet project. It mostly exists to play Go really well. It's a sub-branch of uses of the technology, not a start.
>Well, we're on hackernews, we like the tech stuff. For most of the world, this news is where people learn about Go and the human player's "humiliation" (whether or not the Go community sees it that way) is the story.
In that regard, Lee Sedol has done the game of Go a great service by being so gracious in defeat (and for once, the media headlines actually mirror his humility and maturity). That has actually been my lasting impression from the match.
Maybe that's why AlphaGo is exciting to you. I don't give a shit about technology and AI compared to how much I want to watch AlphaGo play more games of Go, and I suppose that professionals probably feel the same way.
we're slowly moving towards anticipation in continuous strategy games, and life. progressively better decisions/anticipation in progressively much much larger decision space is how this is relevant
No one will ever collect $200 dollars again
It's worth noting that human brain consuming barely 20 watts of power is competing against giant computing infrastructure gobbling up perhaps few 100 killowatts. It's like one player against the army of 5000 players. Many people write off this energy requirements as non-consequential but the fact is that our major technological bottleneck of 21st century is nothing BUT energy consumption. When we have machine that consumes 20 watts and does everything that human brain does, it would be a day for a new species.
I read somewhere that a single machine still wins 25% against alpha Go. That's still 500 W VS 20 W but a bit closer.
Or in the 1990s in the age of DeepBlue (IBM super computer), do young chess prodigies of countries that formed the former USSR have a future?
>"I mean, that's like asking whether marathon runners have a future now that we've got cars."
I don't think they are the same things. Marathon is about the physical limit of human beings, while Go is about mental limit.
Before there are cars, or trains, we all know we are not the fastest in the world. A rabbit can easily overrun an adult human.
However, we never thought that a dog or a bird will beat us in a game like GO. Thinking a machine beating the human champion in the game of GO, is like admitting that we are intelligently inferior to machines.
Not really, speaking specifically about marathon, humans should rank pretty high in the list of the best marathon species. We're up there with horses and I guess wolves.
Well, arguably we are machines.
That's a better statement of my own version, which has been: Now that we have bulldozers and cranes, is there any point to getting physically stronger?
Read the first paragraph. Sounds like the students in the story are forgoing most other learning to be great Go players. There's a big difference in their level of dedication. Does Go still remain as exciting of a sport?
If Chess is any indicator, the ebb and flow of popularity of such games are more dependent on the persona and narratives of their champions and championship matches rather than any existence of a machine superior in skill to its human counterparts.s
Man it would have been great to live at the time of horse races.
???
Is this Poe's Law in action?
We are living in the times of horse races.
Right. I could have just said that. But I wasn't sure how to read the statement. And so the reference to Poe's Law. Just to be clear :)
Does that mean that marathon runners will no longer be in regular use in industry, and be relegated to special events justified in part by aesthetics, and also by audience spectacle?
Likewise, if machines can do quality accounting, does that mean that accounts will be relegated to special accounting events, just like there are marathon events, where people come for aesthetic enjoyment or audience spectacle?
IMO the analogy is correct for Go players as that's mostly a "sport".
There's real cases of similar technology putting people out of jobs, for example I recently read that more and more finance firms more or less completely automate a lot of processes, even those where you'd traditionally would have counted on people's "gut feelings". Deep learning is eerily good at simulating "gut feelings".
Yea, the technology to make marathon-distance couriers obsolete has existed since...before the Scythians?
There are certainly mathematics competitions, where humans (usually high-school or college-age students) participate and cheer on other humans who are artificially restricted from the use of a calculator. And these competitions are considered useful because the skills required to do well on them are very valuable in certain industries, even if the artificial restriction on calculators is removed. In fact, some of the companies that most aggressively hire people who perform well at these competitions are also well-known for automating accounting tasks that were done by humans until recently, such as determining the expected price of a stock.
From the article: "It is estimated that, of South Korea’s three hundred and twenty pros, only around fifty are able to earn a living on tournament winnings." This isn't going to result in massive unemployment.
However, cheating with computer assistance is likely to become a problem, as it is in chess.[1] (The state of the art in computer chess is now roughly at "laptop with off the shelf program can curb-stomp human world champion.")
[1] http://en.chessbase.com/post/yet-another-case-of-cheating-in...
What kind of configuration would you need to run the DeepMind setup? Looking at the wikipedia page if we assume the setup [1] there then it seems like at this time it would be out of the realm of feasibility at this time of setting up this software for cheating.
Google's infrastructure gives it an extra boost, but even the single machine version can beat the distributed version 25% of the time.[1]
[1] https://twitter.com/demishassabis/status/708489093676568576
The nice thing about most of this deep learning stuff is that you can use a million machine hours to train your model, and almost no time to make an evaluation. So the single machine version has all of the pattern recognition given to it by the cluster, but a few ply less tree search depth.
So to me this underscores the relative importance of the deep learning model vs the tree search.
Kind of cool to know (if you're Lee) that your thinking ability is on par with 1920 CPUs and 280 GPUs!
Well, the whole point of this match has been to demonstrate that it is not...
Flip it around: it took 1920 CPUs and 280 GPUs to beat him!
I'd be ecstatic if I could keep GNU Go on 1 CPU from wiping the floor with me... :)
It wouldn't be expensive to rent that much "cloud" hardware for a few hours.
But given the strength of the distributed program, the desktop version is already professional strength because of diminishing returns.
The game play among human competitors is sure to change just as in chess, as new players will train on AlphaGo, etc. whose gameplay will differ significantly from accepted norms in a purely human playing field. For instance, chess programs can plan and optimize to a much further depth (number of moves) in terms of strategy. So, newer players like Carlsen, etc. play differently these days than older ones like Anand, etc. [1]
[1] http://noenthuda.com/blog/2016/03/11/how-computers-have-chan...
This is already happening. Down at the end of a recent Wired article about Fan Hui, the European champion who lost to AlphaGo recently, it mentions that his close interactions with the software has changed his view of the game, and his ranking has also moved, from 633 "into the 300s".
The relevant paragraph is the one with the heading "Machine changes human", at the end.
http://www.wired.com/2016/03/sadness-beauty-watching-googles...
The safest best for sectors of the economy that will continue to prosper under a hard-AI regime are those that involve making humans, especially the rich (eg, owners of AI-based companies) feel happy via human interactions. Go instructors are such a niche.
Actually I would expect game-players and game-instructors to do better than the median profession under such a scenario, because playing games against other humans for entertainment & pure enjoyment of competition is a very human pursuit.
If that's the future, then I'd say AI companies should have to be at least 49% owned by a trust fund in which every citizen automatically has a share.
In the limit near every company becomes an AI company.
I wouldn't write those professionals off so easily. People barely knows AlphaGo until now. Once they start learning about AlphaGo's behaviors, they may be able to come out with new tactics. No doubt it will be a collective effort from the Go community, no one can beat the machine alone.
It's true that humans should be allowed a rematch. And all professionals Go players could be playing in concert.
Hell, Google should organize a "AlphaGo against the world" on internet.
Can't wait for Twitch Plays Go.
Impressive. You wrote this comment, well before the 4th match results, which Lee Sedol won.
In the age of Deep Blue, do young chess prodigies have a future?
Someone should ask Magnus Carlsen.
Edit: fortunately the fine article brings this up and goes further to point out that go is fundamental to self improvement in some cultures and will likely have a long life after AlphaGo.
[13th century] In the age of gunpowder, do young archery prodigies have a future?
This is a good one. It was still many many centuries later before it stopped being a tool in war and from a competitive prospect it is still kickin (although not pervasive)
In the age of Watson, does Jeopardy have a future?
Did Jeopardy ever really have a purpose?
Jeopardy has many purposes. It provides employment, entertainment, and a cultural touchstone, as well as a benchmark of our quest to replicate our nature. Relatively few things have more purpose.
In the age of high tech robot replacements, do unskilled factory workers have a future?
This one unfortunately isn't analogous with the others.
They actually do not.
The analogy only works if people do the particular activity for pleasure, so the question becomes... is it possible to find pleasure in factory work? One could argue that it's possible, but it may look different to what it does today. Some people may enjoy the feeling of refined manual dexterity, so games could emerge where activities like packing boxes at speed becomes a sport, in a similar vein to how cup stacking is a sport now.
In general I'd suggest reading "The player of Games" by Banks. A Culture novel that explores why would someone care about being good at "boardgames" in a universe where even the dumbest appliance could easily outperform any human.
One thing that Banks never addressed in his Culture novels was humans wanting to become Minds. I mean if I live in a society where there are humanoids and Minds, then I want to be a Mind, with a stream-of-consciousness that includes the transition. Wouldn't this be true of many people who are in to science?
Didn't the Minds manipulate culture to channel their charges in various directions (ie, against war-like tendencies)? Maybe ascension was one of the thought-patterns they engineered out.
They did. It isn't explicitly stated, to the best of my knowledge, but the Minds are smart enough that they could easily have done that—and were shown to engineer other parts of the society, such as their deathism.
Humanity has very limited purpose in the Culture. On the surface, it looks nice, but they're effectively pets. Their willingness to ascend was cut off right along with their willingness to live.
All of which is done by social engineering. If you want to go the intelligence-enhancement route, they'll help. They just make sure that very few people do.
That's true, but it's also rather deceptive. Humans are the fundamental core of the Culture; its basic purpose is to look after human-scale people, and this shapes everything it does --- the Culture's a bit suspicious of ascension because they feel it's an abandonment of their responsibilities. Humans are, in effect, the Minds' religion.
And dying is just one of several end-of-life events you can pick, if you want to (including opting out entirely). Naturally it's the one we tend to focus on, because it's the only one we currently have, but the Culture's got lots.
Right, this is where things break down. It's fictional universe, after all.
What are the negative effects of having more Minds? Resource depletion?
If your cat suddenly tells you that what it wants more than anything is to learn General Relativity, then either you're having a psychotic episode, or you have to question how benevolent and ethical your coddling is for your fuzzy wuzzy slave animal.
Intolerable.
It's mentioned a few times - a common end-of-life event is to have your consciousness merged into a Mind. (Possibly with your friends.)
It's also mentioned that you can have your mind transferred into a drone busy, and vice versa, but it's seen as terribly gauche.
Forgot about this. I guess it's more about the relative frequencies. The books, of course, aren't a representative cross-section of the society.
The manners as laws thing is cute, but I don't see it working without invasive non-consensual modification of humans.
E.g. from my own perspective, if I live in a world where there are humans and human-created demigods, then obviously I want to be one of the demigods. Not to enslave the humans, but to look inside of black holes, to travel between stars, to fully understand how biology works, etc.
So not everyone wants to be a scientist or engineer, but wouldn't almost all engineers and scientists want to wield the very best mental and technological tools of their civilization?
I think there's some mention at one point of humans who've augmented themselves to the point of being Mind-like, but by that stage they aren't really human any more.
The only way for that to happen would be to build a mind, and then allow a human conscious control over it. In effect, slavery of a mind by a human. The minds would not allow that, so I'd say you'd be out of luck.
I would think this has to be a serious inflection point... As far learning how to think, strategize, competing with friends, etc go will remain. However for the extremely few say 100 really top level players, this has to in a way be disheartening... at that level its about competition, and the will, discipline and ability to advance... it has to be disheartening to know you will never be the best.
Its kind of like climbing mount Everest 'because it was there'. Its just not 'there' anymore.
It'd be like climbing Everest on foot in the shadow of a cable car which runs to the top. Still a clallenge, but one that feels pointless.
Go has still much more to offer as it is not even close to be a solved game. Additionally, there's handicap Go, and additionally handicaps can be applied to the machine. Go offers the possibility for players of different levels to enjoy a game.
It seems like the authors are trying to use the latest story in the Go world to write about how all the jobs will be replaced by neural networks or something. I mean, people still throw javelins even though it is fairly trivial to build a machine that shoots a projectile farther than a human can throw.
We still enjoy trivia games even though it is easy to google the answers. I dunno, I think it is a bit of a stretch to think AlphaGo has ended serious competitive human Go.
I'm pretty sure humans will not stop playing go at intensely high levels just because a computer can beat them.
The analogy to chess is an interesting one, though, not quite as straightforward as it may seem. Chess, when it was first conquered by computers a couple of decades ago, was a triumph of computer vs human, sure, but in such a different way from the way humans play it. Chess is amenable to brute force search in a way that go isn't (though I understand the chess programs really aren't pure brute force), but human chess players don't (as far as I know) really don't play chess in a brute force way, they rely in intuition, experience, and even a bit of gambling and hedging whether their opponent will "see" or "realize" the strategy in time.
As a result, the chess programs were winning through a "reasoning" process that was very different from what you experience watching people play the game. Something very different is going on when humans play, which makes it interesting - in that sense you can sort of dismiss the machine as playing a different game, albeit one with the same board, pieces, and rules. Instead, it's a giant calculation that happens to beat the more intuitive approach once you can search and score X positions per second through an entirely alternate approach to the game.
This current breakthrough with go sounds different, in that it may mean that computers now play go in a way that is much more similar to the way humans play it (it would be interesting to see if a chess program designed more like the go program would have a huge edge over the brute force search approach). Or, if not the same, perhaps a way that is equally if not more interesting.
I'm kind of bummed that I'm out of my depth on this one (I don't know go or chess well enough to really say), but it's an interesting question.
Good writing.
I also expect that more people will start playing Go, or like me, get a renewed interest in the game.
I read that Lee Sidol is planning on retiring from active play in a few years and move to the USA to evangelize the game in the West.
I played the South Korean national champion and the women's world champion in handicapped exhibition games in the 1970s. It would be awesome to get to do the same with Lee Sidol!
This reminds me of the 1947 story "With Folded Hands" - https://en.m.wikipedia.org/wiki/With_Folded_Hands - where a post apocalyptic future is the consequence of full automation and robotics . people have no purpose left.
Well, absolutely, if they're talking about Golang that is. In all seriousness, a computer did fairly well on Jeopardy some years back, and we're still watching the game show as it's still quite relevant.
And a trained human + computer with internet access (Google search!) could beat the best humans in Jeopardy for quite a while now.
Not really. Jeopardy is all about buzzer timing[0]. With Google search, you're always going to be slow on the buzzer. That's okay, though. The actual trivia is pretty easy. It's designed this way to be accessible to the average TV viewer.
Buzz then search. I think players tend to work this way too, buzzing as soon as they feel like they'll know the answer, but before they actually consciously have it.
Maybe if they learn to collaborate with AI. Or eventually, get implants. As all serious professionals will need to do. That may seem far-fetched, but how many runners compete with bicyclists?
Well you don't see pro runners get bicycle implants (that would hurt)
True. But people do get surrogate limbs.
TL;DR: several personal stories about people who invested a lot of time in the game, general information about go in general as well as what's going on right now, and finally, a semi-answer with a quote:
“A dolphin swims faster than Michael Phelps, but we still want to see how fast he can go,” Lockhart said. “We’re humans and we care about other humans and what they can do.”
Too many words, too little information for one article.
Here's a related and possibly more interesting question nobody seems to be asking -- in the age of robotics and artificial intelligence, does exploitation of workers and environmental catastrophe at the point of extraction of the materials we use to build computers and robots have a future?
And if the answer is yes, why isn't anyone trying to use robots for that purpose?
People are automating resource gathering all the time. See eg http://www.abc.net.au/news/2015-10-18/rio-tinto-opens-worlds...
All AI does is remove the ability to progress through the ranks accurately via online play. Unregulated online competition will suddenly become a bumpy road full of Elo-breaking presences.
I gather this may be a Big Deal, but except insofar as it kills the sport by a thousand cuts, 'Young Go Prodigies' have nothing to worry about.
The computer will be able to give you an Elo number just by looking at all the moves you did in a few games.
Since the computer can tell for every single move whether you played perfectly or if not, by how much you decreased your chances of winning against perfect competition, you'll be able to get a hundred signals out of a game into your elo calculation, instead of just one win/loss condition.
They already do that for catching cheaters in chess. In essence, you treat the positions that occur in a game not as a logical sequence, but as a series of multiple-choice questions.
Color me skeptical. The value of things like Elo is they provide a ramping scale with some degree of statistical significance, because they cover play over many games. I don't think a computer is going to be able to extrapolate with a high degree of confidence, because human play is variable from game to game and long-run strategies are non-obvious constructs for the computer.
You're thinking in terms of 'The AI has solved Go mathematically', but that's not the case; Just because you can run a Monte Carlo best-choice-picker/guesser algorithm doesn't mean you can meaningfully rank how deliberate choices compare with each other more than a few plays away.
Long-run strategies are a human crutch. It's easiest to see when you solved a game mathematically, that you can just value positions independently.
Go hasn't been solved to that level, but it's apparently been solved to higher level than humans ever reached.
I am just parroting http://www.uschess.org/content/view/12677/763 here, so I might as well quote:
"To catch an alleged cheater, Regan takes a set of chess positions played by a single player—ideally 200 or more but his analysis can work with as few as 20—and treats each position like a question on a multiple-choice exam. The score on this exam translates to an Elo rating, a score Regan calls an Intrinsic Performance Rating (IPR)."
This approach also allows to score historic players absolutely, instead of only relatively and trying to find sets of overlapping lifetimes until we reach the modern age.
This article is trying to answer a question no one asked. Of course Go prodigies will always have a future regardless of how well computers become at playing Go. When we start handling out championships, officially ranking computer programs and awarding prize money to them then we can have this conversation.
Why should I really be concerned about Go prodigies? Aren't there a lot more ordinary people being affected by technological unemployment who have much fewer resources to fall back on?
Such as lawyers. And truck drivers, of course.
But there's so much more to this than human obsolescence. This is the cusp of a new stage in evolution.
I would love to see AlphaGo vs itself.
In the age of computers and pocket calculators, does mental calculation have a future?
Forgive me, but the headline, all by itself, left me thinking:
Why are there so many young programmers adopting the Go
programming language, throughout Asia, exclusively?...Have Chess players had a future...?!
This is stupid.