Epistemic status: the claim is nothing new[1] for LW readers; the framing is what I think is new, and the list at the end is a work in progress.
In a social deduction game[2], each player receives a role, which often specifies a goal. Some goals can only be achieved by lying, so there's usually a very strong correlation between lying and having a particular goal. In the simplest version of such a game, there are only two roles: the goodies and the baddies, and the baddies have to kill the goodies while the goodies try to identify who the baddies are through discussion. Honestly admitting that you are a baddie makes you lose immediately, so baddies need to lie, whereas goodies don't have to, and often don't benefit from lying. So there is a very strong correlation between "Person X is lying" and "Person X is a baddie".
In other words, there is an object level (who has which role), and there is a social level (what everyone is saying about their and others' role). Being skillful at the social level, primarily by identifying lies, allows you (depending on the simplicity of the specific game) to make educated guesses about the object level. Identifying liars gives you information about facts.
In real life there is nothing analogous. There are some goals that require lying (e.g. "getting someone to trust a promise you don't mean to keep"), but this is pretty uncommon. People lie for all kinds of different reasons. And even more importantly, many people who say false things aren't lying. Ordinary people lie constantly and almost all of it is innocent: politeness, privacy, self-presentation, kindness, face-saving, not wanting to explain. People have all kinds of wrong models of the world. And people sometimes have wrong models of things for which they have plenty of evidence.
So getting information about the world by
Being good at social-level skills like identifying liars is often uninformative about the world: Life isn't a social deduction game.
Claude also informed me of an additional relevant fact: people are really bad at detecting lies[3].
Having a list of heuristics that try to derive object-level information from the social level seems useful, nonetheless, so I plan to list here all I come across:
Sad revolutionaries
When it comes to revolutionaries, trust only the sad ones. The enthusiastic ones are the oppressors of tomorrow.
Rising Up and Rising Down: Some Thoughts on Violence, Freedom and Urgent Means (2003)
The Missing Moods
You can learn a lot by comparing the mood reasonable proponents would hold to the mood actual proponents do hold.
Mafia, also known as Werewolf, is possibly the most famous one, though the much more complex Blood on the Clocktower has also become very famous lately.
Bond & DePaulo's 2006 meta-analysis (206 studies, ~24,500 judges) found average lie-detection accuracy of ~54% against a 50% chance baseline — 61% on truths, 47% on lies, i.e. worse than chance at spotting the actual lies. A follow-up found trained professionals no better than laypeople.
Epistemic status: the claim is nothing new[1] for LW readers; the framing is what I think is new, and the list at the end is a work in progress.
In a social deduction game[2], each player receives a role, which often specifies a goal. Some goals can only be achieved by lying, so there's usually a very strong correlation between lying and having a particular goal. In the simplest version of such a game, there are only two roles: the goodies and the baddies, and the baddies have to kill the goodies while the goodies try to identify who the baddies are through discussion. Honestly admitting that you are a baddie makes you lose immediately, so baddies need to lie, whereas goodies don't have to, and often don't benefit from lying. So there is a very strong correlation between "Person X is lying" and "Person X is a baddie".
In other words, there is an object level (who has which role), and there is a social level (what everyone is saying about their and others' role). Being skillful at the social level, primarily by identifying lies, allows you (depending on the simplicity of the specific game) to make educated guesses about the object level. Identifying liars gives you information about facts.
In real life there is nothing analogous. There are some goals that require lying (e.g. "getting someone to trust a promise you don't mean to keep"), but this is pretty uncommon. People lie for all kinds of different reasons. And even more importantly, many people who say false things aren't lying. Ordinary people lie constantly and almost all of it is innocent: politeness, privacy, self-presentation, kindness, face-saving, not wanting to explain. People have all kinds of wrong models of the world. And people sometimes have wrong models of things for which they have plenty of evidence.
So getting information about the world by
Being good at social-level skills like identifying liars is often uninformative about the world: Life isn't a social deduction game.
Claude also informed me of an additional relevant fact: people are really bad at detecting lies[3].
Having a list of heuristics that try to derive object-level information from the social level seems useful, nonetheless, so I plan to list here all I come across:
Sad revolutionaries
The Missing Moods
The Mellow Heuristic
Written at Inscribe.
There are also adjacent reflections on authority.
Mafia, also known as Werewolf, is possibly the most famous one, though the much more complex Blood on the Clocktower has also become very famous lately.
Bond & DePaulo's 2006 meta-analysis (206 studies, ~24,500 judges) found average lie-detection accuracy of ~54% against a 50% chance baseline — 61% on truths, 47% on lies, i.e. worse than chance at spotting the actual lies. A follow-up found trained professionals no better than laypeople.
See also this and this, and less relatedly this.