The parallels between the game of mafia and interrogation

[DkKoba]

Hey, Koba here.

I recently, as many others in the recent past, stumbled upon a channel on Youtube that focuses on showing how interrogations work and the methods they use. As I was watching these videos, I was surprised to hear how similar the interrogators felt to scumhunters in a mafia game. (Particularly my own style of play, I cannot comment on anyone else's). They used methods such as making a suspect feel as though they are safe(reassuring a scumread in order to get them to potentially spew), pressuring them over and over in order to get them to slip up and insistence on restating things are some thing I found interesting. Most of this is most useful and applicable in a video/IRL mafia setting, however there are some limited applications I believe for forum mafia as well.

There was also a video where they displayed what an innocent person looks like. One thing they said during this video was that you could just feel the suspect is innocent in this case as opposed to the other cases where you could feel their guilt. These can be equated to gut scumreads/townreads in a way. A clip they showed was a man who was falsely accused of a robbery and they kept insisting they were confirmed guilty of it, and the man was reacting very aggressively. Aggression in reacting to being accused is something we do typically associate with town.


Anyways before I start typing up a whole ass essay about it now, I'll link the channel here: https://www.youtube.com/channel/UCYwVxW ... u8TML-Te9A



Here's some of the relevant videos I feel:
(the first one makes reference to the latter 3)




I implore you all to watch some of these, if not all, and let me know what you think of the parallel of mafia to interrogation in the real world.

[Korina]

egoing, these videos are on the to-watch list anyways because psychology is interesting af

[KittyTacky]

Mafia is just mutual interrogation with often more than 10 people who are both detectives and suspects.

[nomnomnom]

I've been a long time fan of JCS's work, alongside other people who have since then made similar stuff (I suggest Matt Orchard, he's also very great), and to me there are still many key differences that you need to keep in mind.
First of all, the stakes are very different. Here, you are playing a game. Sure, some people consider the stake a bit higher because there's an incentive of winning, but it doesn't compare to the very real fear of facing an investigator especially if you actually did commit a crime. That affects how one person will react to it and so a lot of the things that could indicate guilt in that situation, won't necessary make someone guilty in a mafia game.
The power dynamics are also different. The detective always has the advantage over the suspect. They also get to control the flow of conversation. They control it using a lot of manipulation tricks designed to make the suspect slip. Most of the time they are also more or less convinced you did it so there's no room for doubt. You simply will not win against a detective. Even if you know all the tricks in the book, the game is simply rigged against you and anything you say can be twisted in a thousand ways in a court of law.
Mafia differs from that greatly, and the most notable aspect is that there is close to little power dynamics. Nobody has an inherent status advantage because anyone can be scum. The bad guys have the advantage here. Paranoia plays a much bigger factor here, and the stakes will always be less high. The threat of being accused is also much less high considering that you need a lot of town members to vote you up for it to have any real impact.
That said, I agree with you that some manipulation techniques can be good! The problem being that you can also easily fall for them, and since you do not know have the information advantage like these detectives have, with evidence that points to the suspect baseline, that is a rather easy thing to fall for And that's really what this game is all about!

[yessiree]

I thought the only thing one should say when being interrogated is "lawyer"

[Psyche]

Cool thread!

Vrij, A., Hartwig, M., & Granhag, P. A. (2019). Reading lies: Nonverbal communication and deception. Annual review of psychology, 70, 295-317.

I've been putting off reading this high-profile review of current research on deception detection. The title and abstract seems a bit deceptive; there's apparently a lot of content about interrogative strategies for driving people to make slips. Haven't read it yet though!

[MaxTheFox]

If I ever get interrogated I will tell the detective he is a mafia goon because he is tunneling on me and completely ignoring what the other detective is saying.

[Jake The Wolfie]

Tell him he's a werewolf trying to execute you for being town.

[northsidegal]

the "OMGUS" defense has gotten me acquitted of multiple felonies

[T3]

VOTE: koba
Tell me why you made this thread... sus, don't you think /j

[DkKoba]

VOTE: koba
Tell me why you made this thread... sus, don't you think /j

I know you did it, and I have the report to prove it. Just come clean and tell us why you randed scum. Don't you want to get it off your chest?

[T3]

VOTE: koba
Tell me why you made this thread... sus, don't you think /j

I know you did it, and I have the report to prove it. Just come clean and tell us why you randed scum. Don't you want to get it off your chest?

I bribed the mod, mafia style.

[Jake The Wolfie]

bites a K-9

[yessiree]

One thing I've noticed is that the innocent gives short and concise answers when recalling something from memory, but the guilty feels the need to overjustify their narratives and pays careful attention to the perception of themselves

[Psyche]

That's probably not true based on my research and the broader literature on deception detection I'm aware of.

For example, here's a tweet about a table in the paper I referred to above:

[Psyche]

Incidentally, if we wanna go out of our way to connect the associations in this table to text-based mafia, we might suspect that stuff like what we've sometimes seen called "flailing" is mildly predictive of scum behavior. A previous small investigation I did found no real relationship between a player getting accused in thread of flailing by town and their alliance, but it was statistically flawed and even if it wasn't it wouldn't necessarily prove that whatever behavior is denoted by "flailing" has nothing to do with someone's alliance. It would be pretty tough to get that evidence, even just observationally, without a better operational definition.

[Psyche]

But then what about this cognitive approach to lie detection in the first tweet? That's probably more material here.

The cognitive lie detection approach outlined in the paper consists of three techniques: (1) imposing cognitive load, (2) encouraging interviewees to say more, and (3) asking unexpected questions. It's apparently quite effective overall - works over 70% of the time if you include truth detection in assessment of it.

There's no way to know if it works as well in text-based mafia, where people get to prepare their posts, don't give off non-verbal cues, and aren't under the same sort of pressure as a lot of research conditions surrounding this technique foster.

And it's not super glamorous? I mean overall site meta already encourages people to use votes to "pressure" people into producing content and giving off tells, etc. At the same time, I think a lot of players take a more passive approach to mafia than this research suggests is optimal. They read threads, produces some reads, place down votes and then sort of check-out or press their case.

A better approach to sorting players along with generally putting pressure on them is probably to think up (ideally odd/unanticipated) questions that force them to elaborate

further

on stuff they've said/done in the thread. So like, rather than just making an accusation and watching someone squirm about it, actually interrogate them (along with making them squirm, yeah).

Still some issues with this tough. Biggest one imo people's ability to workshop their answers probably makes this technique less effective. Maybe time taken to produce a response becomes more relevant here, but it's a very noisy variable and other research finds that hesitation isn't a good marker of lying anyway. So idk. Maybe there's some way to pressure players to respond promptly to questions so this factor isn't so influential.

More broadly, something missing from the "cognitive approach to lie detection" meta-analysis referenced above though is an account of what people should be looking for when they try the technique? What is it that people do when you apply this technique that differentiates truth-tellers and liars? There are some explanations of some of this in the cited paper, but they're pretty terse. And what I do see seems to depend on noticing that someone is

struggling

- i.e. to promptly generate accurate details about an event they should have familiarity with if it actually happened. I can see how that's a useful cue in a face-to-face interview where someone's gotta answer your questions immediately and you have the authority to impose cognitive load in odd ways like having them tell a story backwards, but tons of factors make this super hard in text-based games.

So atm I feel like we have the best the broader literature can tell us about one piece of the equation (how to treat suspects so they give off tells) but not others: what tells actually look like, how to generalize this stuff to a meaningfully different format. I doubt there's much research out there about it given how artificial these conditions are.

[yessiree]

I’m curious to know if these studies would produce different results, were they done in a different setting, with different power dynamics and stakes, say detective-suspect power dynamics, and the prospect of getting freedom and conviction of a crime. (I’m assuming they were with volunteers and at fairly low stakes)

So going back to the topic of interrogation, I think it’s still worthwhile to observe the “suspect” in a mafia game if this specific power dynamic (interrogator-suspect) is created. Say for example, a high-profile named townie questioning someone that was being wagoned and in danger of being eliminated, how they respond could be AI

pedit: wall

[Psyche]

paper noted here critiques the cognitive lie detection approach paper.

The Vrij et al. (2017) meta-analysis aggregated two different sorts of outcomes.
Some of the primary studies looked at how human subjects perform in lie detection tasks.
Other primary studies involved ‘objective criteria, using, for example, a discriminant
analysis algorithm. (Vrij et al., p. 8). In this second type of study, researchers coded behaviours
(cues) of subjects who lied or told the truth and used a statistical algorithm to classify
subjects’ honesty–deceit based on the researcher-coded behaviours.

Human judgement of the honesty is different from statistical classification based on
researcher-coded behaviours and statistical analysis. In human detection studies, the
subjects are human receivers who make decisions about senders. In the statistical
classification studies, the subjects are senders and the outcomes come from computer
programs. The computer findings tell us about behavioural cues that differentiate
between truths and lies. The human judgement studies tell us about how well humans can
differentiate between truths and lies – presumably based on behavioural cues. Human
judges do not have the benefit of ground truth hindsight or lots of data for computing fit,
much less the cognitive capacity and know-how to perform statistically optimized
classification in their heads in real time

I think I'm reading that if you don't gloss over this difference (i.e. you just use human judgment data), accuracy is more like 60%. But then again, in a long text-based mafia game we might have the time and energy to achieve something closer to the ceiling set by formal approaches! If you choose that framing, then meta-analysis across studies finds a success rate approaching 80%!

The studies using a formal approach are more interesting too because they'll identify concrete features in participants' responses that contribute to these high success rates. Reviewing those might provide actionable advice to people about how to sort players in their games. And they might give us a firmer basis for statistical projects people in this community might be interested in trying. So I'll look into those next - though I'm maybe crowding out the thread by now.

[yessiree]

it doesn't necessarily have to be based purely on techniques to detect truths or lies, because sometimes they could still be telling truths or half-truths. But we can still focus on various aspects on how the question was answered, and the questions don't even have to be incriminating. So going back to the over-justification part, for example, in one of the videos, one question asked by the detectives was "how long have you known her? (the victim)", to which an innocent man could answer in simple and short sentences, but not the guilty

starts at around 8:36

[Awoo]

Might be worthwhile to cross-reference this thread with the mafiascum dataset.
viewtopic.php?f=5&t=77850

[Psyche]

aw man did i mess up the thread i liked this one

[Gamma Emerald]

There was also a video where they displayed what an innocent person looks like. One thing they said during this video was that you could just feel the suspect is innocent in this case as opposed to the other cases where you could feel their guilt. These can be equated to gut scumreads/townreads in a way. A clip they showed was a man who was falsely accused of a robbery and they kept insisting they were confirmed guilty of it, and the man was reacting very aggressively. Aggression in reacting to being accused is something we do typically associate with town.

I really like this part specifically
Though, are there exceptional cases they go over? I feel like there’s some people you just can’t read in that gut manner and I’m also pretty sure you have some people irl who would give off innocent or guilty vibes constantly regardless of whether they actually are either one.