Defeated by A.I., a Legend in the Board Game Go Warns: Get Ready
July 12, 2024 9:22 AM Subscribe
“I could no longer enjoy the game,” he said. “So I retired.” Lee Saedol was one of the world’s top Go players, and his shocking loss to an A.I. opponent was a harbinger of a new, unsettling era. “It may not be a happy ending,” he says.
Knowing I'll never be faster than a motorcycle is exactly why I refuse to do cardio.
posted by phunniemee at 9:45 AM on July 12 [80 favorites]
posted by phunniemee at 9:45 AM on July 12 [80 favorites]
The story reminded me of this XKCD strip, which makes a good point that as systems become more fluid in structure, computers struggle with dealing with them. (Also, it seems that AlphaGo's strength was that it identified strategic branches that had been overlooked, and it seems that top players have recognized this and are now using it to study those paths.)
posted by NoxAeternum at 9:49 AM on July 12 [3 favorites]
posted by NoxAeternum at 9:49 AM on July 12 [3 favorites]
Knowing I'll never be faster than a motorcycle is exactly why I refuse to do cardio.
a better analogy is "I am the descendent of 1000 years of runners, with a vast culture of running excellence and study and training, when suddenly a bionically enabled alien transports to earth, trashes us in races,......and now people say us runners, we're slow and useless, and there is nothing more to discover in running. in addition everyone gives money to the aliens and says they are the future in every other discipline too and there is an entire culture of aggressively ignorant alienology that makes outlandish, mostly evidence-free pronouncements about they and their alien's intrinsic worth and superiority compared to traditional approaches"
I normally like this cardio metaphor, but not for this example!
posted by lalochezia at 9:52 AM on July 12 [9 favorites]
a better analogy is "I am the descendent of 1000 years of runners, with a vast culture of running excellence and study and training, when suddenly a bionically enabled alien transports to earth, trashes us in races,......and now people say us runners, we're slow and useless, and there is nothing more to discover in running. in addition everyone gives money to the aliens and says they are the future in every other discipline too and there is an entire culture of aggressively ignorant alienology that makes outlandish, mostly evidence-free pronouncements about they and their alien's intrinsic worth and superiority compared to traditional approaches"
I normally like this cardio metaphor, but not for this example!
posted by lalochezia at 9:52 AM on July 12 [9 favorites]
I think there is an important distinction between "I love Go because I could potentially become the best player in the world" and "I love Go because it is enjoyable to play with people."
aka The Fighting Game Dilemma
posted by star gentle uterus at 10:03 AM on July 12 [4 favorites]
aka The Fighting Game Dilemma
posted by star gentle uterus at 10:03 AM on July 12 [4 favorites]
aka The Fighting Game Dilemma
People see my collection of Street Fighter II merchandise and memorabilia and assume I must be good at the game. They're always a little surprised when I tell them I'm actually rubbish, and that I just really like the lore and the character design.
Anyway, I take comfort in the fact that a computer can never play Dungeons & Dragons better than a human can.
posted by Faint of Butt at 10:10 AM on July 12 [12 favorites]
People see my collection of Street Fighter II merchandise and memorabilia and assume I must be good at the game. They're always a little surprised when I tell them I'm actually rubbish, and that I just really like the lore and the character design.
Anyway, I take comfort in the fact that a computer can never play Dungeons & Dragons better than a human can.
posted by Faint of Butt at 10:10 AM on July 12 [12 favorites]
Let me know when AI can play Uno well enough to keep games under an hour
posted by gottabefunky at 10:20 AM on July 12 [1 favorite]
posted by gottabefunky at 10:20 AM on July 12 [1 favorite]
you guys will love this book on games, including how AI is making us think twice about why humans play them: https://bookshop.org/p/books/seven-games-a-human-history-oliver-roeder/18515442?ean=9781324051022
posted by bobarke at 10:23 AM on July 12 [4 favorites]
posted by bobarke at 10:23 AM on July 12 [4 favorites]
As soon as you take games like these out of average human enjoyment and into a hyper-competitive world, the games change. The entire concept changes.
I like playing Scrabble. I hate playing Scrabble against people who have memorized the entire official dictionary of acceptable 2- and 3- letter words. I want to play with a normal schlub like me who doesn't take the game so goddamned seriously. I have a pretty good vocabulary, but I do not know (nor do I ever care to learn) an exhaustive list of acceptable 2-letter words.
I guess I'm just not very competitive. I enjoy the gameplay, not beating the shit out of everyone.
posted by SoberHighland at 10:27 AM on July 12 [33 favorites]
I like playing Scrabble. I hate playing Scrabble against people who have memorized the entire official dictionary of acceptable 2- and 3- letter words. I want to play with a normal schlub like me who doesn't take the game so goddamned seriously. I have a pretty good vocabulary, but I do not know (nor do I ever care to learn) an exhaustive list of acceptable 2-letter words.
I guess I'm just not very competitive. I enjoy the gameplay, not beating the shit out of everyone.
posted by SoberHighland at 10:27 AM on July 12 [33 favorites]
I like playing Scrabble. I hate playing Scrabble against people who have memorized the entire official dictionary of acceptable 2- and 3- letter words.
Oh my god this. I knew someone like this: she always and only wanted to play Scrabble, and she had several editions of the official dictionary memorised. The games were never remotely close because she’d use all sorts of strategic 2 letter words that either I didn’t know or didn’t know counted (like “za.” Short for “pizza.” I know.) Like, why do you want to play at that point? Is it just about refining strategy? God it was no fun.
posted by UltraMorgnus at 10:47 AM on July 12 [5 favorites]
Oh my god this. I knew someone like this: she always and only wanted to play Scrabble, and she had several editions of the official dictionary memorised. The games were never remotely close because she’d use all sorts of strategic 2 letter words that either I didn’t know or didn’t know counted (like “za.” Short for “pizza.” I know.) Like, why do you want to play at that point? Is it just about refining strategy? God it was no fun.
posted by UltraMorgnus at 10:47 AM on July 12 [5 favorites]
If we’re going to treat go as an analogy for the potential impact of AI and to appeal to one of the best players in the world for insight, it seems more than fair to talk about how the perspective of a top competitive player might differ from the perspective of other people who play the game.
posted by atoxyl at 11:04 AM on July 12 [8 favorites]
posted by atoxyl at 11:04 AM on July 12 [8 favorites]
.
posted by HearHere at 11:06 AM on July 12 [2 favorites]
posted by HearHere at 11:06 AM on July 12 [2 favorites]
Chess has had to deal with strong AI for a long time now. The impact is mixed. On the one hand cheating by using these AI’s is rampant, especially online. I used an API to pull down all the opponents I’ve played and about 3% of them get banned for cheating eventually.
On the other hand AI has made chess much more interesting. A lot of new moves early in the game have been found that may not be optimal against a computer but are useful against the lesser calculation capabilities of a human opponent. This has lead to much more exciting games as top level players play these moves as a surprise to unbalance their opponents.
posted by interogative mood at 11:07 AM on July 12 [3 favorites]
On the other hand AI has made chess much more interesting. A lot of new moves early in the game have been found that may not be optimal against a computer but are useful against the lesser calculation capabilities of a human opponent. This has lead to much more exciting games as top level players play these moves as a surprise to unbalance their opponents.
posted by interogative mood at 11:07 AM on July 12 [3 favorites]
Chess is more popular than ever
Also poker, where game theory optimal solvers are better than human players, seems more popular than ever by many metrics.
I don't think Lee Saedol's reaction is truly an indication of the future of Go or any other competitive game disrupted by AI. It says a lot about the emotional reactions of the people caught up in the initial transition.
Have chess and poker entered a "new, unsettling era"? I don't think they have. The complexity of the game tree that initially made it hard for computers to beat humans also makes it hard for humans to learn and use the strategies that computers came up with, so humans are still studying and practicing and trying to get better. They just have new tools to learn from.
Whether the world/economy in general have entered a new, unsettling era is a different question that I don't think can be answered by looking at the impact of AlphaGo on the careers of top competitive players.
posted by allegedly at 11:10 AM on July 12 [2 favorites]
Also poker, where game theory optimal solvers are better than human players, seems more popular than ever by many metrics.
I don't think Lee Saedol's reaction is truly an indication of the future of Go or any other competitive game disrupted by AI. It says a lot about the emotional reactions of the people caught up in the initial transition.
Have chess and poker entered a "new, unsettling era"? I don't think they have. The complexity of the game tree that initially made it hard for computers to beat humans also makes it hard for humans to learn and use the strategies that computers came up with, so humans are still studying and practicing and trying to get better. They just have new tools to learn from.
Whether the world/economy in general have entered a new, unsettling era is a different question that I don't think can be answered by looking at the impact of AlphaGo on the careers of top competitive players.
posted by allegedly at 11:10 AM on July 12 [2 favorites]
One other thing as a chess player who is far from the top of the field; but has improved slowly to the middle. We organize tournaments using an Elo rating system that is designed to match people of equal skill. So even if you are not particularly good at chess, you are able to find events, or sections of events where your chance of winning is only about 50%. It doesn’t matter how much worse I am than current number 1 Magnus Carlson or best computer program, Stockfish. I get to play opponents at my strength and have fun. The reward for getting better at chess isn’t that you win more often, it is mostly that you get to play more complex and difficult games.
posted by interogative mood at 11:12 AM on July 12 [5 favorites]
posted by interogative mood at 11:12 AM on July 12 [5 favorites]
As other people mentioned chess went though this and came out the other side. It came out different, with players looking to the computer as the authority on optimal play, but it certainly survived. Ultimately I think these games are just a limited analogy for the impact of automation because they will always be played as a competition between people, whether amateur or professional. So the comparison hardly speaks to labor issues. It does suggest that other performative human endeavors, like music, will come out the other side as well.
posted by atoxyl at 11:15 AM on July 12 [2 favorites]
posted by atoxyl at 11:15 AM on July 12 [2 favorites]
The reward for getting better at chess isn’t that you win more often, it is mostly that you get to play more complex and difficult games.
I think there's a running analogy for that, too, isn't there? Something like, "The pain doesn't decrease as you get better, you just run faster with the same amount of pain."
But pithier. I'm pretty sure the original phrase is pithier.
posted by clawsoon at 11:18 AM on July 12 [4 favorites]
I think there's a running analogy for that, too, isn't there? Something like, "The pain doesn't decrease as you get better, you just run faster with the same amount of pain."
But pithier. I'm pretty sure the original phrase is pithier.
posted by clawsoon at 11:18 AM on July 12 [4 favorites]
Have chess and poker entered a "new, unsettling era"? I don't think they have
I suspect the transition was unsettling to a lot of chess players, but some top players had already been early adopters of engines as tools and at this point a whole generation has grown up with the computer being unbeatable so it’s not a big thing to them.
The weirdest thing is just that the engine is so instantly authoritative now that you’ll see even top players, when analyzing games, look at the evaluation and top lines and kind of work backwards to explain them in human terms.
posted by atoxyl at 11:23 AM on July 12 [1 favorite]
I suspect the transition was unsettling to a lot of chess players, but some top players had already been early adopters of engines as tools and at this point a whole generation has grown up with the computer being unbeatable so it’s not a big thing to them.
The weirdest thing is just that the engine is so instantly authoritative now that you’ll see even top players, when analyzing games, look at the evaluation and top lines and kind of work backwards to explain them in human terms.
posted by atoxyl at 11:23 AM on July 12 [1 favorite]
The story reminded me of this XKCD strip, which makes a good point that as systems become more fluid in structure, computers struggle with dealing with them.
I was going to make a Calvinball joke, but then looked at the strip and realized Munro had already beat me to it.
(so now I'm not going to play Calvinball anymore}
posted by Greg_Ace at 11:24 AM on July 12
I was going to make a Calvinball joke, but then looked at the strip and realized Munro had already beat me to it.
(so now I'm not going to play Calvinball anymore}
posted by Greg_Ace at 11:24 AM on July 12
I mean the match happened in 2016, and it was decisive and everyone understood that a human pro would never beat a top AI opponent again, and now we've been living the future of go for 8 years and what that's looked like is a second revival in the game's international popularity, continuing growth in China and Korea, the youngest pro ever, Sumire Nakamura, in Japan, new young content creators on YouTube and Twitch, and AI enabling new forms of game analysis. The only way you could conclude that AI killed go is if you care a lot about AI and don't really pay attention to go.
posted by jy4m at 11:31 AM on July 12 [2 favorites]
posted by jy4m at 11:31 AM on July 12 [2 favorites]
...comments from people about games that they dabble in and find it annoying when other players are prepared...
Not sure why you got so snappy there? My point was not so much my feelings, but more that the game completely changes depending on players. Two average teenagers sitting down to play Go is a completely different thing that the all-time world champion being defeated by AI and getting their feelings hurt. I guess I feel an infinitesimal amount of sadness for the all-time champion, but... not really. People can still play Go and have fun! Some people will get better and keep playing and have more fun.
Computers are going to outthink humans in games with strict rules, limitations and numbers and I guess I'm not surprised in the least.
posted by SoberHighland at 11:38 AM on July 12 [2 favorites]
Not sure why you got so snappy there? My point was not so much my feelings, but more that the game completely changes depending on players. Two average teenagers sitting down to play Go is a completely different thing that the all-time world champion being defeated by AI and getting their feelings hurt. I guess I feel an infinitesimal amount of sadness for the all-time champion, but... not really. People can still play Go and have fun! Some people will get better and keep playing and have more fun.
Computers are going to outthink humans in games with strict rules, limitations and numbers and I guess I'm not surprised in the least.
posted by SoberHighland at 11:38 AM on July 12 [2 favorites]
But pithier. I'm pretty sure the original phrase is pithier.
"It never gets easier; you just go faster." - Greg LeMond
posted by flod at 12:00 PM on July 12 [4 favorites]
"It never gets easier; you just go faster." - Greg LeMond
posted by flod at 12:00 PM on July 12 [4 favorites]
"It never gets easier; you just go faster." - Greg LeMond
A tangent: I was at a data summit where the keynotr speaker claimed a very close variant of this as her mother's words. I liked it so much I used it and attributed that speaker .
Retconning five years of anecdotes because now I know the correct attribution.
posted by Gorgik at 12:07 PM on July 12 [2 favorites]
A tangent: I was at a data summit where the keynotr speaker claimed a very close variant of this as her mother's words. I liked it so much I used it and attributed that speaker .
Retconning five years of anecdotes because now I know the correct attribution.
posted by Gorgik at 12:07 PM on July 12 [2 favorites]
I used to have a lot of fun playing chess as a kid. I didn't know anything about the game apart from the rules, but no one I ever played against (relatives, kids at school) did either so it didn't bother me. It was like playing any other game and sometimes I won and sometimes I lost.
At one point someone was like "you really like chess. you should join the chess club at school!", so I did and holy shit was I completely outmatched. There was so much about chess that I didn't know, it might as well have been an entirely different game. All of my peers had long left the naive goofing round phase I was still in and they weren't going to indulge my relative inexperience by dialing things back at all.
You can call me a quitter, but there really is something different about playing a game where everyone is familiar with the rules but hasn't yet honed their skills. This is why most people don't play chess as a party game.
posted by RonButNotStupid at 12:11 PM on July 12 [3 favorites]
At one point someone was like "you really like chess. you should join the chess club at school!", so I did and holy shit was I completely outmatched. There was so much about chess that I didn't know, it might as well have been an entirely different game. All of my peers had long left the naive goofing round phase I was still in and they weren't going to indulge my relative inexperience by dialing things back at all.
You can call me a quitter, but there really is something different about playing a game where everyone is familiar with the rules but hasn't yet honed their skills. This is why most people don't play chess as a party game.
posted by RonButNotStupid at 12:11 PM on July 12 [3 favorites]
Not sure what the point of the article is, despite having read it a couple of times.
Am I supposed to feel sorry for a world championship player who dedicated his life to learning a thousand year old craft?
Lots of people have lost their jobs due to technology. Lots of people no longer enjoy their jobs. (I retired too!)
It sounds like his brother is more accepting. "Like other pros, he now trains with A.I. systems that continue to learn and improve."
You can't turn back the clock. As the Smothers Brothers sang:
John Henry said to the captain, "By God I ain't no fool. Before I'll die with my hammer in my hand, I'm gonna get me a steam drill too, Lord, Lord. Get me a steam drill too. (Whop, whop, whop, whop)"
And as far as Clavinball goes, I see no reason why it's safe from AI.
posted by MtDewd at 12:18 PM on July 12 [3 favorites]
Am I supposed to feel sorry for a world championship player who dedicated his life to learning a thousand year old craft?
Lots of people have lost their jobs due to technology. Lots of people no longer enjoy their jobs. (I retired too!)
It sounds like his brother is more accepting. "Like other pros, he now trains with A.I. systems that continue to learn and improve."
You can't turn back the clock. As the Smothers Brothers sang:
John Henry said to the captain, "By God I ain't no fool. Before I'll die with my hammer in my hand, I'm gonna get me a steam drill too, Lord, Lord. Get me a steam drill too. (Whop, whop, whop, whop)"
And as far as Clavinball goes, I see no reason why it's safe from AI.
posted by MtDewd at 12:18 PM on July 12 [3 favorites]
From what I understand, one of the side effects of the incredible improvements in computer chess is that one of the main bottlenecks in chess development is much less of a bottleneck now: playing strong opponents. In the olden days, it was sometimes non-trivial for chess learners once they've gotten past some level to get sufficient game time against stronger opponents but nowadays everyone has access to grandmaster-level opponents with the download of an app. Furthermore, I believe a lot of chess apps also provide analysis, so there's even a kind of rudimentary (?) coaching that's available. So, it seems to me that there's probably never been a better time to be a chess learner than in this current era of computer chess.
I can't help but wonder if these kinds of phenomena will be (or already have been?) repeated in the go community with the rise of computer go.
posted by mhum at 12:40 PM on July 12 [2 favorites]
I can't help but wonder if these kinds of phenomena will be (or already have been?) repeated in the go community with the rise of computer go.
posted by mhum at 12:40 PM on July 12 [2 favorites]
Am I supposed to feel sorry for a world championship player who dedicated his life to learning a thousand year old craft?
Lots of people have lost their jobs due to technology. Lots of people no longer enjoy their jobs. (I retired too!)
Aren't all of those people deserving of compassion?
posted by bq at 12:48 PM on July 12 [7 favorites]
Lots of people have lost their jobs due to technology. Lots of people no longer enjoy their jobs. (I retired too!)
Aren't all of those people deserving of compassion?
posted by bq at 12:48 PM on July 12 [7 favorites]
I get that it can be tough. I talked with Kasparov a couple times and of course he still plays, but as I understand it just for fun. His loss to Deep Blue was embarrassing in some ways but although he holds onto some anger at IBM for, in his opinion, cheating at that match in particular (they apparently did a last-minute patch to counter a winning strategy he'd hit on), he long ago let go of the anger at the machine.
The way he said he thought of it was (as I remember it): I wasn't the first world champion to be beaten by a computer - I was the last person in history capable of beating a computer. That's a pretty cool way to think about it!
But of course that was 25 years ago or whatever for him, and Lee Saedol was whipped not long ago and in, I believe, rather more soundly. I would not be at all surprised to see him return to the game in a few years once he's grieved the loss of not just his role but anyone's role as the "best in the world."
posted by BlackLeotardFront at 12:49 PM on July 12 [5 favorites]
The way he said he thought of it was (as I remember it): I wasn't the first world champion to be beaten by a computer - I was the last person in history capable of beating a computer. That's a pretty cool way to think about it!
But of course that was 25 years ago or whatever for him, and Lee Saedol was whipped not long ago and in, I believe, rather more soundly. I would not be at all surprised to see him return to the game in a few years once he's grieved the loss of not just his role but anyone's role as the "best in the world."
posted by BlackLeotardFront at 12:49 PM on July 12 [5 favorites]
nowadays everyone has access to grandmaster-level opponents with the download of an app
I don’t think anyone particularly recommends playing against engines to learn, because at full strength they are just going to crush you and to make them beatable you have to handicap them such that they will fail in weird inhuman ways. But everyone having a GM++ level analysis partner to go over their games is a big deal.
posted by atoxyl at 12:59 PM on July 12 [2 favorites]
I don’t think anyone particularly recommends playing against engines to learn, because at full strength they are just going to crush you and to make them beatable you have to handicap them such that they will fail in weird inhuman ways. But everyone having a GM++ level analysis partner to go over their games is a big deal.
posted by atoxyl at 12:59 PM on July 12 [2 favorites]
Also it’s made possible things like automatically extracting an endless supply of puzzles from games played online.
posted by atoxyl at 1:02 PM on July 12
posted by atoxyl at 1:02 PM on July 12
I feel sad for him, I remember watching that match and how alien and shocking some of the moves were to the commentators. I hope he comes around to being able to enjoy the game again.
The idea of a game becoming almost an entirely different game at different levels is interesting. Scrabble with a sheet of the two letter words printed out is fun, and you start being able to make really cool shapes. I think that for a lot of people, rather than games becoming more strictly competitive as you get deeper into them, they become almost more collaborative. It's a joy to see your opponent pull something crazy off, when you can't help but appreciate it knowing what it takes. Being on the receiving end of that from an implacable box must be eerie.
posted by lucidium at 1:33 PM on July 12 [3 favorites]
The idea of a game becoming almost an entirely different game at different levels is interesting. Scrabble with a sheet of the two letter words printed out is fun, and you start being able to make really cool shapes. I think that for a lot of people, rather than games becoming more strictly competitive as you get deeper into them, they become almost more collaborative. It's a joy to see your opponent pull something crazy off, when you can't help but appreciate it knowing what it takes. Being on the receiving end of that from an implacable box must be eerie.
posted by lucidium at 1:33 PM on July 12 [3 favorites]
Why doesn't any AI jerk ever making a laundry washing and hanging robot, or maybe at least improve the actual washer? It's got understandable rules, like "throw all game pieces on the floor until someone yells at you", involves game theory like "did I put too much in such that the machine won't work?" and punishments like the "whoops forgot about that stain" penalty card. And I'd bet that Gary Kasparov and Lee Seadol aren't even that great at it. And I'd bet the best person wouldn't even be that John Henry butthurt by being beat, they'd be like "Ok, I'll go do something else like watch tv".
They finally made a Sonic Ice machine though (tiny cubes of ice), so they are learning. Thank you for that.
posted by The_Vegetables at 1:37 PM on July 12 [3 favorites]
They finally made a Sonic Ice machine though (tiny cubes of ice), so they are learning. Thank you for that.
posted by The_Vegetables at 1:37 PM on July 12 [3 favorites]
I think the world could do with fewer smart appliances, not more.
posted by No One Ever Does at 1:57 PM on July 12 [1 favorite]
posted by No One Ever Does at 1:57 PM on July 12 [1 favorite]
The idea of a game becoming almost an entirely different game at different levels is interesting. Scrabble with a sheet of the two letter words printed out is fun, and you start being able to make really cool shapes.
Yeah the dynamics here are interesting because playing against someone who knows those word lists can feel unfair if you don’t, for pretty obvious reasons. But knowing them arguably improves the game, because it opens up more possibilities to actually play the words you’re holding. And then on the other end, memorizing the whole dictionary like pros do is just too much dull effort for it to be fun for the vast majority of people.
posted by atoxyl at 2:01 PM on July 12 [1 favorite]
Yeah the dynamics here are interesting because playing against someone who knows those word lists can feel unfair if you don’t, for pretty obvious reasons. But knowing them arguably improves the game, because it opens up more possibilities to actually play the words you’re holding. And then on the other end, memorizing the whole dictionary like pros do is just too much dull effort for it to be fun for the vast majority of people.
posted by atoxyl at 2:01 PM on July 12 [1 favorite]
These tools are not the end of the sport. People play these games against other people. Just because a car can go faster than a person, hasn't meant the end of track and field.
posted by interogative mood at 2:02 PM on July 12
posted by interogative mood at 2:02 PM on July 12
Aren't all of those people deserving of compassion?
Lots of them are. Let's profile them all in the NYT. The buggy whip manufacturers, the shirt collar makers...
I was one of perhaps 150 folks who were trained in-depth to repair the IBM 3033. I spent 14 weeks of my life just in the initial training. Three years later it was effectively obsolete. Five years after that it was withdrawn from maintenance. Do you feel compassion for me? Or should I have adjusted to the new reality and applied what I learned there to my new situation?
I've always had some lack of compassion for people who think that because they are good at something, they deserve to be well-paid for it, regardless of its worth to anyone else.
(and I realize that doesn't show me in the best light)
I prefer those who see this as a new opportunity to learn even more about the possibilities in go. It's working for chess.
posted by MtDewd at 2:06 PM on July 12 [2 favorites]
Lots of them are. Let's profile them all in the NYT. The buggy whip manufacturers, the shirt collar makers...
I was one of perhaps 150 folks who were trained in-depth to repair the IBM 3033. I spent 14 weeks of my life just in the initial training. Three years later it was effectively obsolete. Five years after that it was withdrawn from maintenance. Do you feel compassion for me? Or should I have adjusted to the new reality and applied what I learned there to my new situation?
I've always had some lack of compassion for people who think that because they are good at something, they deserve to be well-paid for it, regardless of its worth to anyone else.
(and I realize that doesn't show me in the best light)
I prefer those who see this as a new opportunity to learn even more about the possibilities in go. It's working for chess.
posted by MtDewd at 2:06 PM on July 12 [2 favorites]
You can call me a quitter, but there really is something different about playing a game where everyone is familiar with the rules but hasn't yet honed their skills. This is why most people don't play chess as a party game.
That is really unfortunate At the club level we want people to have fun and have fair games. This should be true for clubs at school and other clubs that have adults / kids playing. This is why we generally start players at a low rating and pair them against other low rated players. This is also how opponents are matched on online sites like lichess.org / chess.com.
posted by interogative mood at 2:14 PM on July 12
That is really unfortunate At the club level we want people to have fun and have fair games. This should be true for clubs at school and other clubs that have adults / kids playing. This is why we generally start players at a low rating and pair them against other low rated players. This is also how opponents are matched on online sites like lichess.org / chess.com.
posted by interogative mood at 2:14 PM on July 12
I've often thought that AlphaGo and DeepBlue were interesting not for beating people at their respective games as much as for showing that there was far more to the game. There is the game of Go and there is the smaller subset that is the game of Go that people actually play with each other and, it turns out, those are not the same game.
There was an article this week in nature making the case that the various AIs that beat people at Go are not superhuman, they are simply hyper-optimized for beating humans who are playing the game. It is more than possible to train an AI to beat those AI, but that doesn't grant some greater mastery either. The AI-beating-AI, it turns out, are pretty weak against human opponents.
posted by selenized at 2:20 PM on July 12 [6 favorites]
There was an article this week in nature making the case that the various AIs that beat people at Go are not superhuman, they are simply hyper-optimized for beating humans who are playing the game. It is more than possible to train an AI to beat those AI, but that doesn't grant some greater mastery either. The AI-beating-AI, it turns out, are pretty weak against human opponents.
posted by selenized at 2:20 PM on July 12 [6 favorites]
> atoxyl: "I don’t think anyone particularly recommends playing against engines to learn, because at full strength they are just going to crush you and to make them beatable you have to handicap them such that they will fail in weird inhuman ways."
Oh for sure. The bottleneck I'm thinking of doesn't happen right at the beginning of learning. From what I understand (from my friend who does youth chess coaching), this is the kind of thing that's helpful once you've already started to be competitive in tournaments, i.e.: you're good enough to beat a number of your peers but not quite good enough to climb up into the next ranks. At a certain point in chess development, the most effective thing to progress further isn't puzzles or theory or whatever but just straight-up playing against stronger opponents (and, of course, analyzing the results).
posted by mhum at 2:35 PM on July 12
Oh for sure. The bottleneck I'm thinking of doesn't happen right at the beginning of learning. From what I understand (from my friend who does youth chess coaching), this is the kind of thing that's helpful once you've already started to be competitive in tournaments, i.e.: you're good enough to beat a number of your peers but not quite good enough to climb up into the next ranks. At a certain point in chess development, the most effective thing to progress further isn't puzzles or theory or whatever but just straight-up playing against stronger opponents (and, of course, analyzing the results).
posted by mhum at 2:35 PM on July 12
Back when I was a kid at some county fair like event, I played Tic Tac Toe with a chicken. The chicken either won or it was a draw. I don’t remember. In college I learned that it was just Skinner Behaviorism in action. Looking back, I think it was more beneficial for me to think I just lost to a chicken.
posted by njohnson23 at 2:58 PM on July 12 [4 favorites]
posted by njohnson23 at 2:58 PM on July 12 [4 favorites]
Do you feel compassion for me? Or should I have adjusted to the new reality and applied what I learned there to my new situation?
These are not mutually exclusive propositions. It seems like you see compassion as an attack or an anchor. This confuses me.
posted by bq at 3:28 PM on July 12 [4 favorites]
These are not mutually exclusive propositions. It seems like you see compassion as an attack or an anchor. This confuses me.
posted by bq at 3:28 PM on July 12 [4 favorites]
Yeah, some strange hostility in this thread towards people who don’t enjoy the game the “right” way.
Lots of people enjoy games by testing the limits and becoming “the best” at something and there’s literally nothing wrong with that. Speedrunners and minmaxxers are not “enjoying video games wrong” because they enjoy it differently from me. For a lot of people, this competitive and professional approach to the game scratches a problem-solving, limit stretching itch that’s highly enjoyable. Enjoying that more than or even instead of other aspects of the game is not a sin.
As for “people should suck it up and adapt”… if no one has told you this, you’re allowed to grieve. Even if what has to happen is you move on, it’s natural and healthy to experience grief over loss, including in your professional life. His reaction is very human and I would welcome more pieces talking about the grief people experience when their field is disrupted. Including the shirt collar workers and the people trained on the IBM 3033. Learning about those experiences, as learning about this one, broadens my horizons and helps me build empathy for my fellow humans.
posted by brook horse at 5:38 PM on July 12 [8 favorites]
Lots of people enjoy games by testing the limits and becoming “the best” at something and there’s literally nothing wrong with that. Speedrunners and minmaxxers are not “enjoying video games wrong” because they enjoy it differently from me. For a lot of people, this competitive and professional approach to the game scratches a problem-solving, limit stretching itch that’s highly enjoyable. Enjoying that more than or even instead of other aspects of the game is not a sin.
As for “people should suck it up and adapt”… if no one has told you this, you’re allowed to grieve. Even if what has to happen is you move on, it’s natural and healthy to experience grief over loss, including in your professional life. His reaction is very human and I would welcome more pieces talking about the grief people experience when their field is disrupted. Including the shirt collar workers and the people trained on the IBM 3033. Learning about those experiences, as learning about this one, broadens my horizons and helps me build empathy for my fellow humans.
posted by brook horse at 5:38 PM on July 12 [8 favorites]
I was going to comment, but brook horse said everything I needed to say (and probably better than I would have).
posted by biogeo at 5:48 PM on July 12 [1 favorite]
posted by biogeo at 5:48 PM on July 12 [1 favorite]
Well, this is timely: https://arstechnica.com/ai/2024/07/superhuman-go-ais-still-have-trouble-defending-against-these-simple-exploits/
“But in the last few years, researchers have discovered flaws in these top-level AI Go algorithms that give humans a fighting chance. (…) Those exploitable holes highlight the importance of evaluating "worst-case" performance in AI systems, even when the "average-case" performance can seem downright superhuman. On average, KataGo can dominate even high-level human players using traditional strategies. But in the worst case, otherwise "weak" adversaries can find holes in the system that make it fall apart.”
posted by bq at 7:04 PM on July 12
“But in the last few years, researchers have discovered flaws in these top-level AI Go algorithms that give humans a fighting chance. (…) Those exploitable holes highlight the importance of evaluating "worst-case" performance in AI systems, even when the "average-case" performance can seem downright superhuman. On average, KataGo can dominate even high-level human players using traditional strategies. But in the worst case, otherwise "weak" adversaries can find holes in the system that make it fall apart.”
posted by bq at 7:04 PM on July 12
I was one of perhaps 150 folks who were trained in-depth to repair the IBM 3033. I spent 14 weeks of my life just in the initial training. Three years later it was effectively obsolete. Five years after that it was withdrawn from maintenance. Do you feel compassion for me?
You're goddamn right I do. In fact, not only do I feel compassion for you, but I'm also furious on your behalf.
Or should I have adjusted to the new reality and applied what I learned there to my new situation?
Absolutely not. Nobody should have to put up with that kind of treadmill bullshit. Skills should be for life.
posted by Faint of Butt at 8:50 PM on July 12
You're goddamn right I do. In fact, not only do I feel compassion for you, but I'm also furious on your behalf.
Or should I have adjusted to the new reality and applied what I learned there to my new situation?
Absolutely not. Nobody should have to put up with that kind of treadmill bullshit. Skills should be for life.
posted by Faint of Butt at 8:50 PM on July 12
Perhaps all is not lost - - between...
posted by fairmettle at 10:10 PM on July 12
it seems that AlphaGo's strength was that it identified strategic branches that had been overlooked, and it seems that top players have recognized this and are now using it to study those paths....and...
“Superhuman” Go AIs still have trouble defending against these simple exploits...it seems we may still have a lot to learn from each other.
posted by fairmettle at 10:10 PM on July 12
Mod note: A couple deleted; don't attack fellow members. Content policy.
posted by taz (staff) at 12:13 AM on July 13 [1 favorite]
posted by taz (staff) at 12:13 AM on July 13 [1 favorite]
I think a little hallucination might make Calvinball somewhat trivial.
I also think robots are bad at fabric, so I don’t want them doing laundry my laundry.
posted by nat at 6:46 AM on July 13
I also think robots are bad at fabric, so I don’t want them doing laundry my laundry.
posted by nat at 6:46 AM on July 13
I hate playing Scrabble against people who have memorized the entire official dictionary of acceptable 2- and 3- letter words.
Someone makes this comment in pretty much every Scrabble thread. To me, this just means that Scrabble is a bad game, because there’s no functional difference between accumulated hard-won experience and intentional memorization, only one’s motive. And there’s no fair way to ask someone to ignore what they already know.
Of course there are ways to compensate for knowledge differences and still have fun, as others have pointed out, like playing semi-collaboratively, or open-book, etc.
posted by mubba at 7:32 AM on July 13
Someone makes this comment in pretty much every Scrabble thread. To me, this just means that Scrabble is a bad game, because there’s no functional difference between accumulated hard-won experience and intentional memorization, only one’s motive. And there’s no fair way to ask someone to ignore what they already know.
Of course there are ways to compensate for knowledge differences and still have fun, as others have pointed out, like playing semi-collaboratively, or open-book, etc.
posted by mubba at 7:32 AM on July 13
It is more than possible to train an AI to beat those AI, but that doesn't grant some greater mastery either.
It absolutely does. AlphaZero is the most famous of the machine-learning chess engines, and it got to the point of being able to beat grandmasters not by beating grandmasters but exactly by playing games against itself.
The current version of Stockfish, by the way, incorporates a self-trained neural network for positional evaluation, and has as much claim to being AI as any of the *Zero engines at this point.
posted by flabdablet at 7:55 AM on July 13
It absolutely does. AlphaZero is the most famous of the machine-learning chess engines, and it got to the point of being able to beat grandmasters not by beating grandmasters but exactly by playing games against itself.
The current version of Stockfish, by the way, incorporates a self-trained neural network for positional evaluation, and has as much claim to being AI as any of the *Zero engines at this point.
posted by flabdablet at 7:55 AM on July 13
I suppose I should have said that it didn't, as in "it didn't grant some greater mastery in the case of the specific article I linked to about Go playing AIs". The same strategy of playing oneself over and over again is pretty bog standard at this point and yet the researchers were able to, repeatably, create an adversary that could beat the best bot at Go (bots that used that strategy of self-play). Those adversaries, however, were not themselves particularly skilled and were beaten by humans who were recreational players.
I think that shows that there is still more to learn about the game of Go than even our best bots have figured out. Which, I would think, would be seen as exciting and good news.
posted by selenized at 8:09 AM on July 13 [2 favorites]
I think that shows that there is still more to learn about the game of Go than even our best bots have figured out. Which, I would think, would be seen as exciting and good news.
posted by selenized at 8:09 AM on July 13 [2 favorites]
Well the core reason it took so much longer for Go programs to surpass the best human players than chess programs is that the space of game states explodes much faster, right? So you can’t rely as much on deep search. The big innovation with AlphaGo was that it leans more heavily on advanced position evaluation done via neural networks, in combination with a more limited Monte Carlo search. So it kind of seems plausible that it’s still not really exploring the full possible space of ways to play.
The current version of Stockfish, by the way, incorporates a self-trained neural network for positional evaluation, and has as much claim to being AI as any of the *Zero engines at this point
It does but it’s slotting the NN evaluation function into a traditional chess engine framework in place of the prior hand-tuned heuristics. For chess this hybrid approach seems to be holding onto a lead over the *Zero approach (Leela Chess Zero being the standard-bearer in the chess world).
posted by atoxyl at 2:04 PM on July 13
The current version of Stockfish, by the way, incorporates a self-trained neural network for positional evaluation, and has as much claim to being AI as any of the *Zero engines at this point
It does but it’s slotting the NN evaluation function into a traditional chess engine framework in place of the prior hand-tuned heuristics. For chess this hybrid approach seems to be holding onto a lead over the *Zero approach (Leela Chess Zero being the standard-bearer in the chess world).
posted by atoxyl at 2:04 PM on July 13
the NN evaluation function
a much smaller model than what the Alpha/Leela engines use, as I understand - it’s designed to run very efficiently on CPU as part of the brute force minimax machine that is Stockfish, not to stand on its own
posted by atoxyl at 2:17 PM on July 13 [1 favorite]
a much smaller model than what the Alpha/Leela engines use, as I understand - it’s designed to run very efficiently on CPU as part of the brute force minimax machine that is Stockfish, not to stand on its own
posted by atoxyl at 2:17 PM on July 13 [1 favorite]
I think that shows that there is still more to learn about the game of Go than even our best bots have figured out
True of games in general. The strategy being employed by the anti-bot bots is the same as that employed by human players against pre Deep Blue chess bots: make radical early sacrifices in order to enter branches of the game tree that the bot has no idea what to do with due to having alpha-beta pruned them on the basis of poor positional evaluation scores.
These exploitable weaknesses reflect faults not so much in the engines per se as in the simplifying assumption that their designers make in order to keep the game trees tractable: that an opponent will always seek to play the best move. Once one side adopts deliberately playing objectively poor moves as a matter of strategy, the game effectively morphs from Go or Chess into something more resembling Chicken.
Bots are just not good at psychological games because for the time being they lack both minds and theories thereof.
posted by flabdablet at 9:57 PM on July 13
True of games in general. The strategy being employed by the anti-bot bots is the same as that employed by human players against pre Deep Blue chess bots: make radical early sacrifices in order to enter branches of the game tree that the bot has no idea what to do with due to having alpha-beta pruned them on the basis of poor positional evaluation scores.
These exploitable weaknesses reflect faults not so much in the engines per se as in the simplifying assumption that their designers make in order to keep the game trees tractable: that an opponent will always seek to play the best move. Once one side adopts deliberately playing objectively poor moves as a matter of strategy, the game effectively morphs from Go or Chess into something more resembling Chicken.
Bots are just not good at psychological games because for the time being they lack both minds and theories thereof.
posted by flabdablet at 9:57 PM on July 13
« Older "I don't want nothing anymore; I want everything." | You're No Steve Rogers! Newer »
I think there is an important distinction between "I love Go because I could potentially become the best player in the world" and "I love Go because it is enjoyable to play with people." I feel like Saedol is in the first category and -- to be honest -- I don't know if that's a healthy attitude.
Every sport or competitive hobby has to face the fact that calculation and engineering will always beat human genetics (Usain Bolt will never run as fast as a motorcycle) but that doesn't mean humans can't continue to enjoy it.
posted by riotnrrd at 9:44 AM on July 12 [34 favorites]