I feel like he's missing the point of why we as humans play games at all. Sure, no human will probably ever be able to beat AlphaGo again, but that isn't why we play or watch games. Being the best...
I feel like he's missing the point of why we as humans play games at all. Sure, no human will probably ever be able to beat AlphaGo again, but that isn't why we play or watch games. Being the best in the world at something isn't always the most interesting thing about watching someone play a game. A lightweight boxer, for example, still enjoys boxing, not because he is the best in the world at boxing as a whole, but because it is interesting to play and watch the narratives that take place in lower levels in boxing. Just because no one can beat AlphaGo doesn't mean that being the best human in the world at Go isn't still incredibly interesting. College basketball is still interesting despite knowing that the top teams in the college leagues couldn't beat the top teams in the NBA.
For me anyway a lot of what makes strategy games interesting is the evolution of the metagame. Players are constantly searching for new counters, optimisations and developing unique personal...
For me anyway a lot of what makes strategy games interesting is the evolution of the metagame. Players are constantly searching for new counters, optimisations and developing unique personal styles. Every now and again there is a era defining seismic shift as someone figures out something completely wild and game breaking that throws all the old rules out.
But once AI exceeds humans the most effective play style will be to imitate the current AI metagame. Even if the game is not solved, it is now the AI doing the solving.
Sure you can choose to play sub-optimally and not copy the AI but I don't think it will have the same magic as when humans where at the forefront pushing the boundaries of what is possible in a game.
If anyone was capable of imitating the AI playstyle they would have already been—by far—the greatest go player in history, with nobody else able to even come close to beating them. The whole point...
If anyone was capable of imitating the AI playstyle they would have already been—by far—the greatest go player in history, with nobody else able to even come close to beating them. The whole point is that it's playing at a level completely beyond even the best human players, so nobody can simply imitate it.
It's like telling someone that if they want to be better at chess they should just play the way Magnus Carlsen does. If they were able to do that, they'd already be one of the best in the world. It's not practical advice.
Would advising someone trying to improve at chess to study Magnus Carlsen be bad advice? I suspect that it isn't. I don't have much of a base in chess or go so I will switch over to starcraft...
Would advising someone trying to improve at chess to study Magnus Carlsen be bad advice?
I suspect that it isn't.
I don't have much of a base in chess or go so I will switch over to starcraft examples.
Around 2012 telling a zerg sc2 player to imitate Stephano's 12min roach max build in protoss vs zerg matchups would absolutely result in them winning nearly all games against protoss players of equivalent skill and experience. They would not become one of the best in the world by becoming Stephano's equal but by direct imitation would be much more formidable players.
An effective counter to this strategy was an unsolved problem. As a protoss player I spent a considerable amount of time experimenting on how to counter this build.
Eventually timings for 2 base immortal all-ins emerged that made Stephano's build extremely risky and the meta moved on.
The meta itself always felt like a community effort to solve a constantly shifting problem that in some ways every player contributed to.
When AI achieves supremacy over humans and initiates all of these shifts in the metagame the human players become non participating observers in a sense.
I also suspect that these dramatic shifts will be less entertaining as a spectator when they are made by incremental versions of neural network rather than by upstart players dethroning the established greats.
You could argue that a strategy game was always a technical problem masquerading as an art. Or that the problem solving aspect was not a real part of the communal experience.
But that is why I described it as being a type of magic, I am lamenting the death of the illusion.
Talking about metagaming in competitive videogames vs tabletop RPGs are comparing two completely different concepts. They don't have anything in common whatsoever besides the fact that they share...
Talking about metagaming in competitive videogames vs tabletop RPGs are comparing two completely different concepts. They don't have anything in common whatsoever besides the fact that they share a term.
Metagaming in D&D is conflating player knowledge with character knowledge. You're not roleplaying a character in a competitive online game, so it's not even something that's possible. Moreover, the goals of each activity are completely different. In a TTRPG you're creating a shared story that everyone at the group can immerse themselves in. In a competitive online game, you're trying to beat other players
It's like comparing cheating in online videogames with cheating in singleplayer games. Completely different concepts.
I think you're trying to compare types of games that aren't comparable. Go and chess are perfect-information, turn-based games. There isn't really a shifting metagame where people are coming up...
I think you're trying to compare types of games that aren't comparable. Go and chess are perfect-information, turn-based games. There isn't really a shifting metagame where people are coming up with strategies that people won't expect or know how to counter.
AlphaGo didn't apply some sort of novel strategy to go that people previously didn't realize was possible and they can imitate, it's just a lot better at the game. It's still playing the game the same way, but the moves it chooses are better.
I sympathize with Lee Se-Dol's point. Your argument makes sense for sports in which the spectacle is the main draw, but games like Chess and Go are almost entirely logic-focused, with well-defined...
I sympathize with Lee Se-Dol's point. Your argument makes sense for sports in which the spectacle is the main draw, but games like Chess and Go are almost entirely logic-focused, with well-defined rules and a narrower field of possibilities (compared with basketball, soccer...) that make them very prone to automation. Most people don't care whether Magnus Carlsen sweats during a game, and the way he touches the pieces is entirely irrelevant. A strong chess engine on a weak laptop can easily defeat GMs every day of the week. This is not true about basketball or any physical sport. Chess and Go are pretty abstract to begin with. Before computers, people had books full of matches using precise notation. Our relation with chess was impersonal already, to the point that, once computers became our personal GMs, their human counterparts became a lot less interesting. I believe Go is the same.
I was looking at this story earlier today and saw another story noting that he's actually still going to play 3 games against a Korean baduk AI in December: Final challenge: Go grandmaster Lee...
I posted this in ~games prior to being moved because I think the AI supremacy over human players angle has implications for all games not just go specifically.
I posted this in ~games prior to being moved because I think the AI supremacy over human players angle has implications for all games not just go specifically.
I feel like he's missing the point of why we as humans play games at all. Sure, no human will probably ever be able to beat AlphaGo again, but that isn't why we play or watch games. Being the best in the world at something isn't always the most interesting thing about watching someone play a game. A lightweight boxer, for example, still enjoys boxing, not because he is the best in the world at boxing as a whole, but because it is interesting to play and watch the narratives that take place in lower levels in boxing. Just because no one can beat AlphaGo doesn't mean that being the best human in the world at Go isn't still incredibly interesting. College basketball is still interesting despite knowing that the top teams in the college leagues couldn't beat the top teams in the NBA.
For me anyway a lot of what makes strategy games interesting is the evolution of the metagame. Players are constantly searching for new counters, optimisations and developing unique personal styles. Every now and again there is a era defining seismic shift as someone figures out something completely wild and game breaking that throws all the old rules out.
But once AI exceeds humans the most effective play style will be to imitate the current AI metagame. Even if the game is not solved, it is now the AI doing the solving.
Sure you can choose to play sub-optimally and not copy the AI but I don't think it will have the same magic as when humans where at the forefront pushing the boundaries of what is possible in a game.
If anyone was capable of imitating the AI playstyle they would have already been—by far—the greatest go player in history, with nobody else able to even come close to beating them. The whole point is that it's playing at a level completely beyond even the best human players, so nobody can simply imitate it.
It's like telling someone that if they want to be better at chess they should just play the way Magnus Carlsen does. If they were able to do that, they'd already be one of the best in the world. It's not practical advice.
Would advising someone trying to improve at chess to study Magnus Carlsen be bad advice?
I suspect that it isn't.
I don't have much of a base in chess or go so I will switch over to starcraft examples.
Around 2012 telling a zerg sc2 player to imitate Stephano's 12min roach max build in protoss vs zerg matchups would absolutely result in them winning nearly all games against protoss players of equivalent skill and experience. They would not become one of the best in the world by becoming Stephano's equal but by direct imitation would be much more formidable players.
An effective counter to this strategy was an unsolved problem. As a protoss player I spent a considerable amount of time experimenting on how to counter this build.
Eventually timings for 2 base immortal all-ins emerged that made Stephano's build extremely risky and the meta moved on.
The meta itself always felt like a community effort to solve a constantly shifting problem that in some ways every player contributed to.
When AI achieves supremacy over humans and initiates all of these shifts in the metagame the human players become non participating observers in a sense.
I also suspect that these dramatic shifts will be less entertaining as a spectator when they are made by incremental versions of neural network rather than by upstart players dethroning the established greats.
You could argue that a strategy game was always a technical problem masquerading as an art. Or that the problem solving aspect was not a real part of the communal experience.
But that is why I described it as being a type of magic, I am lamenting the death of the illusion.
Talking about metagaming in competitive videogames vs tabletop RPGs are comparing two completely different concepts. They don't have anything in common whatsoever besides the fact that they share a term.
Metagaming in D&D is conflating player knowledge with character knowledge. You're not roleplaying a character in a competitive online game, so it's not even something that's possible. Moreover, the goals of each activity are completely different. In a TTRPG you're creating a shared story that everyone at the group can immerse themselves in. In a competitive online game, you're trying to beat other players
It's like comparing cheating in online videogames with cheating in singleplayer games. Completely different concepts.
I think you're trying to compare types of games that aren't comparable. Go and chess are perfect-information, turn-based games. There isn't really a shifting metagame where people are coming up with strategies that people won't expect or know how to counter.
AlphaGo didn't apply some sort of novel strategy to go that people previously didn't realize was possible and they can imitate, it's just a lot better at the game. It's still playing the game the same way, but the moves it chooses are better.
I sympathize with Lee Se-Dol's point. Your argument makes sense for sports in which the spectacle is the main draw, but games like Chess and Go are almost entirely logic-focused, with well-defined rules and a narrower field of possibilities (compared with basketball, soccer...) that make them very prone to automation. Most people don't care whether Magnus Carlsen sweats during a game, and the way he touches the pieces is entirely irrelevant. A strong chess engine on a weak laptop can easily defeat GMs every day of the week. This is not true about basketball or any physical sport. Chess and Go are pretty abstract to begin with. Before computers, people had books full of matches using precise notation. Our relation with chess was impersonal already, to the point that, once computers became our personal GMs, their human counterparts became a lot less interesting. I believe Go is the same.
I was looking at this story earlier today and saw another story noting that he's actually still going to play 3 games against a Korean baduk AI in December: Final challenge: Go grandmaster Lee Sedol to face domestic AI on Dec. 18-20
Edit: this seems to be the original interview/article that this story is based on: https://en.yna.co.kr/view/AEN20191127004800315
I posted this in ~games prior to being moved because I think the AI supremacy over human players angle has implications for all games not just go specifically.
That seems reasonable, I'll move it back.