Has anyone been able to play Mafia using bayesian methods? I have tried and failed due to encountering situations that eluded my attempts to model them mathematically. But since I am not strong at math, I’m hoping others have had success?
And the related question: any mafiascum.net players here?
Edit: I mean specifically using bayesian methods for online forum-based Mafia games. These seem to me to give the player enough time to do conscious calculations.
I wonder if there aren’t any group rationality games that don’t seriously undermined group moral and cohesion. The last time I played Mafia people ended up crying and my relationship with my brother and cousin went through traumatic upheaval. Diplomacy is not a better option.
This seems like an unusual experience to have. I have played Mafia with 3+ non-overlapping groups in person and 4+ non-overlapping groups online, and have yet to encounter any trouble; in fact, in two of the cases we were explicitly playing as a bonding exercise to improve group morale and cohesion, and it seems to have worked both times.
The last time I played Mafia people ended up crying
And what about the times before that?
Playing mafia has never undermined real social relationships in my experience, and I’ve introduced this game to perhaps 20 people in real life, with at least 2 completely non-overlapping groups.
Also, I doubt face-to-face mafia should be considered a game that especially exercises rationality. It seems to me that you get thrown a huge fuckton of cognitive biases with no time to combat them.
(again, my original question should specify “forum based mafia games”...let me edit that now...)
On reflection, I think the problems came from the people in the group being too close. I have certainly had fun before. We may have also taken the game too seriously.
It’s more like it teaches a sort of mini-rationality: “You’re swimming in cognitive biases, but your intuitions can also be helpful. Empirically develop a few techniques to separate good intuitions from bad with decent error probability.”
I’m not sure that doing so would be useful. It seems like normal Mafia techniques already approximate Bayesian reasoning, and formalizing it would be very challenging and IMO unlikely to offer unusual insights. That said, I’m fairly good at online Mafia and I suspect such techniques would better benefit less advanced players.
Certainly. A basic Mafia technique is examining the past play of the person you’re suspicious of, then looking at whether their current play is more similar to their play as scum or their play as town. There is also wide knowledge (at least online) of moves that are generally “scummy,” such as congratulating the doctor after he or she successfully protects, as these moves have been determined to be commonly used by scum. Of course, all of this is constantly evolving, since once something is generally known as a scumtell, advanced scum players avoid it. Further, different things are tells at different levels of play, which tends to make the game much more complicated than my above description might indicate.
That said, I think it’s certainly possible to do better than chance—my own record, at least of games that I can remember, is 4 wins to 1 loss, all as town (I have yet to be scum in my recent games).
Further, there are some situations where certain tactics have been determined, over wide periods of play, to be dominant, and applying these strategies gives you a very high chance to win. For instance, if the town has a doctor and a cop (and knows this) and also knows the scumgroup has no roleblocker, the best strategy is to stop voting to lynch, have the cop claim, and have the cop constantly investigate while protected by the doctor. The scum must then start hitting other targets in hopes of getting the doc. A truly advanced doctor will then, knowing the scum is doing this, not actually protect the cop but instead protect other members of the town in the hopes of blocking the scum’s pseudorandom flailing, but a truly advanced scum player might anticipate this and try to kill the cop instead—so there are mindgames all over the place, but dominant strategies are still known.
Generally, I feel like Mafia—at least online Mafia—is a rather good rationality exercise. I could expand this to a top-level post if there’s interest.
Do make the top-level post please. I think there is use in the making Mafia more well-known in demographics such as the one we have here.
It sounds like online Mafia is a totally different and much better game than what I’ve played at various icebreaker functions, camps, and times when there’s a substitute teacher
In my experience the outcome of face-to-face mafia can be even more dependent on the players’ skill, once you get past the newbie phase. Not just because newbies can’t read others well, but I think they are also less readable due to undeveloped meta and making vastly suboptimal plays that regular scumhunting techniques do not read well. Once there is some standard in the players’ moves and some meta is available, one can read much more accurately in face-to-face games than online due to factors such as tone, moments of hesitation, and body language.
And thus for a given single game, I would rather play mafia face-to-face with groups of regular players than online, though I would prefer playing online to face-to-face with a whole group of newbies.
Face-to-face Mafia is certainly easier to read people in, but this actually (IMO) makes it a worse game. There are other issues as well, such as the inability of the Mafia to communicate articulately at night, but if you’re a good lie detector (or the scum are bad liars) the game becomes almost trivial, and introducing the difficulties of online communication IMO adds an appealing element of challenge. That said, I agree that face-to-face Mafia with a regular group can certainly be fun and even educational in itself.
It sounds like online Mafia is a totally different and much better game than what I’ve played at various icebreaker functions, camps, and times when there’s a substitute teacher. I’ll check it out if I ever have a clear enough schedule. Also, I’d definitely enjoy a top-level post if you made one.
I don’t know how well it works in games with only 1 scum player, but with at least two just the fact that there are two players who know they each have a partner changes their behavior enough that the game isn’t random. There’s also some change in what people say just because each side has a different win condition, although again this is less true with just one scum player.
As just a simple example, when you’re playing as the scum it can be really hard (at least for me) to make a good argument that someone I know is a normal villager isn’t, which can be enough for another player to deduce my role.
That’s interesting; I haven’t played enough mafia to really study it. And in all the games I have played, the town always lynches the first player someone bothers to accuse—there aren’t any actual arguments.
I just had 4 games with the same 5 players (setup is 4 town 1 scum) that all ended in scum victory. Random lynching should yield only 53% chance of scum victory. 0.53^4 seems low enough that this is likely a case of better than random.
The players in this case were new to the game with the exception of myself (and after the first couple games I was constantly night killed). I was going to say that this seems to suggest that scum is stronger in newbie games, but then I realized I have no data to draw this comparison with. :-(
Unfortunately I play mostly as a diversion on a private site, not on mafiascum or epicmafia, so they aren’t as out in the open as you’d like. If you want I can link you to a recent newbie game that I was in on mafiascum, but the number of replacements makes it a little hard to follow and it’s not exactly anyone’s best play either.
Trying to update even on just the well-defined data looks impossible for humans, trying to update on what other people are saying would be difficult, even with a computer. Also, it seems like there might be certain disadvantages if you turn out to be Mafia.
Allow me to specify: I am referring to online forum mafia games.
These games are slow enough that one can do some calculations, if one can find the numbers (and that seems to be the hard part, along with deciding how they should be calculated).
I’ve thought and am still thinking that the fact that I’ve never heard of bayesian methods being used in mafia is simply an observation about the failures of players, not that it inherently cannot be done using available tools.
Frankly I’m surprised mafia does not seem to attract more attention from the demographic concerned with rationality. If some set of methods were developed that consistently worked and cut through the jungle of biases that is the nature of the game, then that would be an achievement for the progress of rationality, would it not? I think many methods that may develop would easily transfer to other uses as well.
Has anyone been able to play Mafia using bayesian methods? I have tried and failed due to encountering situations that eluded my attempts to model them mathematically. But since I am not strong at math, I’m hoping others have had success?
And the related question: any mafiascum.net players here?
Edit: I mean specifically using bayesian methods for online forum-based Mafia games. These seem to me to give the player enough time to do conscious calculations.
I wonder if there aren’t any group rationality games that don’t seriously undermined group moral and cohesion. The last time I played Mafia people ended up crying and my relationship with my brother and cousin went through traumatic upheaval. Diplomacy is not a better option.
This seems like an unusual experience to have. I have played Mafia with 3+ non-overlapping groups in person and 4+ non-overlapping groups online, and have yet to encounter any trouble; in fact, in two of the cases we were explicitly playing as a bonding exercise to improve group morale and cohesion, and it seems to have worked both times.
And what about the times before that?
Playing mafia has never undermined real social relationships in my experience, and I’ve introduced this game to perhaps 20 people in real life, with at least 2 completely non-overlapping groups.
Also, I doubt face-to-face mafia should be considered a game that especially exercises rationality. It seems to me that you get thrown a huge fuckton of cognitive biases with no time to combat them.
(again, my original question should specify “forum based mafia games”...let me edit that now...)
On reflection, I think the problems came from the people in the group being too close. I have certainly had fun before. We may have also taken the game too seriously.
It’s more like it teaches a sort of mini-rationality: “You’re swimming in cognitive biases, but your intuitions can also be helpful. Empirically develop a few techniques to separate good intuitions from bad with decent error probability.”
In my experience playing with a rationality crowd (at a meet-up), it was excellent for learning the visceral feeling of motivated cognition.
I can report that playing Mafia at a meetup markedly improved group interaction. What impact this has on your position is unknown.
Balderdash/Fictionary?
I play online Mafia but haven’t attempted to use explicit Bayesian reasoning to do so.
Please attempt and see if you have better results than I did. And if you succeed come back and tell us all about it!
:-)
I’m not sure that doing so would be useful. It seems like normal Mafia techniques already approximate Bayesian reasoning, and formalizing it would be very challenging and IMO unlikely to offer unusual insights. That said, I’m fairly good at online Mafia and I suspect such techniques would better benefit less advanced players.
There are such things as Mafia techniques? I’ve never seen anyone do better than chance. Care to explain?
Certainly. A basic Mafia technique is examining the past play of the person you’re suspicious of, then looking at whether their current play is more similar to their play as scum or their play as town. There is also wide knowledge (at least online) of moves that are generally “scummy,” such as congratulating the doctor after he or she successfully protects, as these moves have been determined to be commonly used by scum. Of course, all of this is constantly evolving, since once something is generally known as a scumtell, advanced scum players avoid it. Further, different things are tells at different levels of play, which tends to make the game much more complicated than my above description might indicate.
That said, I think it’s certainly possible to do better than chance—my own record, at least of games that I can remember, is 4 wins to 1 loss, all as town (I have yet to be scum in my recent games).
Further, there are some situations where certain tactics have been determined, over wide periods of play, to be dominant, and applying these strategies gives you a very high chance to win. For instance, if the town has a doctor and a cop (and knows this) and also knows the scumgroup has no roleblocker, the best strategy is to stop voting to lynch, have the cop claim, and have the cop constantly investigate while protected by the doctor. The scum must then start hitting other targets in hopes of getting the doc. A truly advanced doctor will then, knowing the scum is doing this, not actually protect the cop but instead protect other members of the town in the hopes of blocking the scum’s pseudorandom flailing, but a truly advanced scum player might anticipate this and try to kill the cop instead—so there are mindgames all over the place, but dominant strategies are still known.
Generally, I feel like Mafia—at least online Mafia—is a rather good rationality exercise. I could expand this to a top-level post if there’s interest.
Do make the top-level post please. I think there is use in the making Mafia more well-known in demographics such as the one we have here.
In my experience the outcome of face-to-face mafia can be even more dependent on the players’ skill, once you get past the newbie phase. Not just because newbies can’t read others well, but I think they are also less readable due to undeveloped meta and making vastly suboptimal plays that regular scumhunting techniques do not read well. Once there is some standard in the players’ moves and some meta is available, one can read much more accurately in face-to-face games than online due to factors such as tone, moments of hesitation, and body language.
And thus for a given single game, I would rather play mafia face-to-face with groups of regular players than online, though I would prefer playing online to face-to-face with a whole group of newbies.
Face-to-face Mafia is certainly easier to read people in, but this actually (IMO) makes it a worse game. There are other issues as well, such as the inability of the Mafia to communicate articulately at night, but if you’re a good lie detector (or the scum are bad liars) the game becomes almost trivial, and introducing the difficulties of online communication IMO adds an appealing element of challenge. That said, I agree that face-to-face Mafia with a regular group can certainly be fun and even educational in itself.
It sounds like online Mafia is a totally different and much better game than what I’ve played at various icebreaker functions, camps, and times when there’s a substitute teacher. I’ll check it out if I ever have a clear enough schedule. Also, I’d definitely enjoy a top-level post if you made one.
I don’t know how well it works in games with only 1 scum player, but with at least two just the fact that there are two players who know they each have a partner changes their behavior enough that the game isn’t random. There’s also some change in what people say just because each side has a different win condition, although again this is less true with just one scum player.
As just a simple example, when you’re playing as the scum it can be really hard (at least for me) to make a good argument that someone I know is a normal villager isn’t, which can be enough for another player to deduce my role.
That’s interesting; I haven’t played enough mafia to really study it. And in all the games I have played, the town always lynches the first player someone bothers to accuse—there aren’t any actual arguments.
I just had 4 games with the same 5 players (setup is 4 town 1 scum) that all ended in scum victory. Random lynching should yield only 53% chance of scum victory. 0.53^4 seems low enough that this is likely a case of better than random.
The players in this case were new to the game with the exception of myself (and after the first couple games I was constantly night killed). I was going to say that this seems to suggest that scum is stronger in newbie games, but then I realized I have no data to draw this comparison with. :-(
Were you the scum in any of the games?
I was scum in none of the games.
I want to read some games of mafia players who browse this site. Do you mind pointing me to some of your games?
Unfortunately I play mostly as a diversion on a private site, not on mafiascum or epicmafia, so they aren’t as out in the open as you’d like. If you want I can link you to a recent newbie game that I was in on mafiascum, but the number of replacements makes it a little hard to follow and it’s not exactly anyone’s best play either.
Sure, link to it.
Here it is. I’m “Fetterkey.”
Trying to update even on just the well-defined data looks impossible for humans, trying to update on what other people are saying would be difficult, even with a computer. Also, it seems like there might be certain disadvantages if you turn out to be Mafia.
Allow me to specify: I am referring to online forum mafia games.
These games are slow enough that one can do some calculations, if one can find the numbers (and that seems to be the hard part, along with deciding how they should be calculated).
I’ve thought and am still thinking that the fact that I’ve never heard of bayesian methods being used in mafia is simply an observation about the failures of players, not that it inherently cannot be done using available tools.
Frankly I’m surprised mafia does not seem to attract more attention from the demographic concerned with rationality. If some set of methods were developed that consistently worked and cut through the jungle of biases that is the nature of the game, then that would be an achievement for the progress of rationality, would it not? I think many methods that may develop would easily transfer to other uses as well.