Mafia Debate: Behavioral Analysis On D1
== Intro ==
There are objective differences between town and mafia. Town are uninformed; mafia are informed. Town scumhunt; mafia don't. Town push lynches on people they think are mafia; mafia know the alignment of those they push and defend. Town must lynch to win; mafia need not lynch anyone. These differences create "tells." Analyzing behavior allows the town to locate these "tells," which in turn allows town to find and lynch mafia.
bossy's going to argue that tells cannot exist on D1. He'll argue that D1 behaviors aren't "objective information" and he'll argue that "tells" only exist after the town has "objective information." However, the town does have "objective informaton" on D1. That objective information is knowledge about the differences between town and mafia. Town know they must lynch mafia to win; they also know mafia need not lynch anyone. Or another example -- town knows that mafia knows everyone's alignment. This information is "objective information" that all townies possess at the start of every mafia game.
This "objective information" is also all you need to find scum on D1. Even if it's not successful 100% of the time, it certainly increases the likelihood of a successful D1 lynch. Ultimately, the mafia cannot help but post from an informed perspective, and town cannot help but post from an uninformed perspective. Tells are unavoidable. Thus, even the best mafiosos are often caught via behavioral analysis on D1. This is not to say everyone can spot tells, but just becuase you can't spot tells doesn't mean they don't exist. The trick is developing the ability to spot them.
== What does behavioral analysis on D1 look like? ==
The following quote comes from a game on mafiascum that F-16 recommended I use for purposes of illustrating good behavioral analysis on D1. I edited a couple posts together for the sake of readability. The substance of the analysis remains unchanged. Baboon, the subject of the analysis, was a hydra who flipped mafia after being lynched on D2. However, the analysis itself is from D1.
"Baboon's certainty that I'm scum is excessively high for the gamestate/reasoning. Their reasoning was: OMGUS + reactive play + a belief that I stole their idea. That reasoning does not square with their certainty, which is a scum-tell because it shows that their read is fake. How much weight would you give each point? OMGUS? I'd give it none. It's not a scum-tell, and Baboon are experienced enough not to give it much if any weight. Reactive play? Arguably more weight than OMGUS but still, not much, because I tend to be a reactive player sometimes. Stealing their idea? Nothing scummy about it. Does the behavior show that my thought process was contrived? No. Does it show that my behaviors are contrived? No. Does it have any pro-scum effect? No. Does it have any scum motivation? No. I could go down the list of ways to connect a behavior to scum. But it doesn't have any of them. It's a purely null behavior. But Baboon never considered much deeper than surface-level. And the weight Baboon gives these behaviors simply doesn't square with their certainty. Which shows that their thought process is contrived." 
The analysis works by focusing on a major difference between town and mafia: the mafia must fake scumreads because they know who the town are. The result is mafia sometimes reveal a discrepancy between their degree of certaity and their provided reasoning. Town never have a discrepancy between their degree of certainty and their provided reasoning, because their degree of certainty is always a product of their reasoning.
Mafia cannot know what degree of certainty a read should have because they cannot evaluate their reasoning fron an uninformed perspective. That produces a "tell" based on "objective inforation," i.e. mafia will sometimes understate or overstate their degree of certainty vis-a-vis their reasoning. Town can use this fact to find scum on D1 by analyzing behaviors.
== Why analyze behavior on D1? ==
(1) The town should lynch on D1.
I think bossy agrees, but to be safe, here's two simple reasons to lynch on D1. First, town cannot predict ahead-of-time if it'll get to MYLO or LYLO. If you no lynch and then end up at MYLO, the town's down a possible mislynch. To ensure we use every possible mislynch every game, town should lynch on D1. Ultimately, lynching on D1 gives the town more control over who ends up at LYLO.
Second, D1 lynches provide tons of behavioral information, because it's when the informational disparity between town and scum is greatest. That means the informed/uninformed dichotomy is at its strongest on D1, which makes the D1 lynch a goldmine of behavioral information to analyze in retrospect, and even during D1 itself.
(2) Analyzing behavior increases the success of D1 lynches. It's "possible" to find scum through behaviors alone. The quote above from the mafiascum game is an example of this. F-16's WhiteFlag Nightless game is the best example of that on this site. The game had no roles or theme. The players had to rely solely on behaviors. The town steamrolled the scum.  Towns that use behavioral analysis on D1 lynch with a higher percentage success on D1 than if they were to lynch randomly. That makes behavioral analysis a useful tool which increases town's chance of winning.
My opponent is correct in identifying the fact that there are differences between town and scum. The problem is that, if town is uninformed, all of their actions are baseless and random – as such, there is no way to distinguish town from scum, since scum-sided actions are just as likely to come from town as they are to come from scum. If town tries to read people or do anything but end the DP ASAP, they are liable to being misread as scum. Since you can’t distinguish between town and scum, most of DP1 is clutter.
The fact of the matter is that reactions can only be evaluated as reactions to something; some event is needed for there to be legitimate reactions that have motives behind them. My opponent claims that the differences between what town and scum have to do to win the game is enough to induce differences in behavior – the behavioral differences that can be seen on DP1 come from the fact that reads must be fabricated. The problem with this argument is that it assumes that any reads on DP1 can be justified to begin with. If scum are caught because their reads are groundless and they cannot judge the degree of certainty to put behind them, these reads must have less of a rational basis than the reads of town. This begs the question – it assumes that reads can be given on DP1 by town that are grounded in anything but thin air.
If town cannot give rational reads, then there is no reason for assuming that the certainty they put behind their irrational reads are anything but arbitrary. If that is so, then there are no distinctions between scum and town reads, and, as such, this is not a valid foundation for behavioral analysis.
My opponent’s arguments regarding the success rates of behavioral analysis on DP1 do not help his case – they assume a causal link between town wins and behavioral analysis simply from correlation, rather than any rational justification for such a relationship.
Second, the fact that town are uninformed and mafia are informed means town and mafia act differently. So even if bossy's right that town's actions are "baseless and random," that's still a basis for analyzing behavior, because the mafia won't have "baseless and random" actions. The mafia cannot act "randomly" because their votes will reflect an informed perspective. If the town is building up a lynch on town, the mafia KNOWS it's on town. The mafia KNOWS it's a mislynch. The mafia can either support the mislynch or not support the mislynch. Either way, the mafia will ALREADY KNOW whether it's a mislynch. That makes their actions non-random, which means their behavior is DISTINGUISHABLE from town behaviors.
Bossy says "scum-sided actions are just as likely to come from town as they are to come from scum." First, that statement is simply untrue. Mafia will act pro-scum more often than town because everything mafia do is pro-scum (or they wouldn't do it), whereas town do things that are pro-town at least some of the time (otherwise they'd never win). Thus, bossy's statement is false on its face. Second, even if town and scum both perform pro-scum actions, they do it with a different intent, and from a different perspective. Mafia do things from an informed perspective, so when they act pro-scum, they do it intentionally, and they know it'll be pro-scum. When town do pro-scum things, they do it unintentionally, from an uninformed perspective, so they don't realize ahead-of-time it'll be pro-scum. That fact alone allows town to distinguish pro-scum behaviors from town and scum.
Bossy says "if town tries to read people or do anything but end the DP ASAP, they are liable to being misread as scum." This statement has no impact. Town are always liable to being misread. That doesn't mean they should sit around and do nothing. If they do nothing, the mafia wins. Town MUST do something, and anytime they do something, they're liable to being misread. That's simply not a reason against behavioral analysis.
Bossy says "reactions can only be evaluated as reactions to something; some event is needed for there to be legitimate reactions that have motives behind them." I do not disagree. Town and mafia are always reacting to something. They're first reaction begins when they receive their role. Then, when someone posts in the DP, everyone reacts. The reactions continue. Yes, the FIRST vote that someone puts down will probably be random, but that doesn't mean that REACTIONS are random. The reactions themselves are legitimate.
Bossy says "if scum are caught because their reads are groundless and they cannot judge the degree of certainty to put behind them, these reads must have less of a rational basis than the reads of town." I think bossy misunderstands how the degree of certainty tell works. Mafia often provide reasons with better reasoning than town's reads. Even so, the mafia can be caught, because often their degree of certainty won't match that reasoning. If the reasoning is really good, but the mafia is uncertain to an unnatural degree, then it suggests that the read itself is fake, even though the reasoning is good. The relevant issue here is whether the degree of certainty matches the reasoning. Whether the reasoning is good or bad, or whether the reasoning is more or less rational, is only relevant for the tell insofar as it helps evaluate the read's degree of certainty.
Bossy says "if town cannot give rational reads, then there is no reason for assuming that the certainty they put behind their irrational reads are anything but arbitrary." I'm not sure what bossy's talking about. The degree of certainty will be a product of their reasoning. Townies don't go around calling someone scum unless they have a reason to think that person is scum. Their degree of certainty will reflect the strength or weakness of their reasoning, as well as how much depth their thought process has, plus other situational details and player meta. Behavioral analysis incorporates all these things.
Bossy argues that town cannot "justify" their reads because it's based on nothing, I already addressed this argument multiple times, in this round as well as in the previous round. Bossy does a terrible job addressing the actual arguments I made in the prior round. He keeps saying town have no basis for their reads, but as I explained, town and mafia behave differently.
The town KNOWS three things: (1) it knows that they have a different win condition than mafia; (2) it knows that mafia cannot treat the game like a puzzle because mafia knows everyone's alignments; and (3) it knows that mafia therefore must fabricate reads. Those three facts -- which the Town knows from the beginning of the game -- are enough to begin analyzing behaviors. They provide a "ground," or a "base," or a "non-random" element, to behavioral analysis. Bossy never addresses that argument. He does not ever explain how the town's KNOWLEDGE is not sufficient to provide a "ground." He simply just states -- without ever substantiating -- that town has no basis for its reads.
My opponent correctly identifies my argument - that, since there is nothing to go off of, Town will always play essentially randomly on DP1, and, if this is so, there is no way to distinguish between Town and Scum play. He argues that Town's actions aren't baseless because they're capable of choosing to do things intentionally - the problem is that he never shows that these actions are inherently ratonally justifiable.
My point is not that Scum act randomly. My point is that there is no way of distinguishing between directed and purposeful Scum actions and Town actions that were random. By definition, if Town acts randomly, it is completely possible that they will act exactly as Scum would. Therefore, as long as Town is forced to act randomly (and they are before there are objective results/flips), they can appear to be just as scummy as Scum, and, if this is so, there is no way to read people to distinguish Town from Scum.
My opponent has built his case around the idea that Scum act to further their wincon, and, because of this, all you have to do to find Scum is to find people doing what Scum would do. This completely fails if town does what Scum would do, and, if DP1 Town play is essentially random, Town is likely to do what Scum would.
My opponent claims that you can tell Town from Scum because Townies do pro-Town things, while Scum do not. This presupposes that there is a way to tell what actions are pro-Town on DP1, and, since the only way to do this is to read people to judge who is Town and Scum so you can analyze which actions benefit which factions, my opponent is begging the question. What is a pro-Town action if Town has no way of telling Town from Scum?
Pro's entire case boils down to this: "Behavioral analysis on DP1 is valuable because we can distinguish Town from Scum. We can distinguish Town from Scum because behavioral analysis on DP1 is valid." Again, this just begs the question - how do you validate behavioral analysis without presupposing that it's useful? My opponent has yet to give a plausible solution.
My point wasn't that Town shouldn't read because they're liable to be misread - it's that the fact that Town can be misread shows that Town cannot distinguish between Town and Scum, and, as such, Town should not try. I agree, Town must be active to win the game, but the way to do this is most certainly not by trying to read people on DP1. Intent doesn't matter if Town is unable to determine it.
Re: the certainty behind reads
It's irrelevant whether or not Mafia must fake reads in a certain way. What is relevant is if Town is able to tell that the reads are fake, and this can only be done if Town reads are any different. If Town has no basis for reads, they must basically throw them out randomly, and, if this is so, there is no reason to suppose that any particular reads are intentionally misleading instead of just coming from a Townie with nothing better to post. My opponent says that "Townies don't go around calling someone scum unless they have a reason to think that person is scum", but this assumes that Townies are able to have a reason for thinking that players are scum on DP1, which my arguments deny.
I do not object to the fact that the Town has some certain knowledge at the start of the game. It's obvious that Town knows that they are Town and that Mafia is not and that each faction must play for themselves. This is totally irrelevant, however, if it cannot be shown that this knowledge helps ground any reads in reality. Like I have already argued, even if Scum must play a certain way to win, if this play is indistinguishable from Town play then there is no way to distinguish Scum from Town. I accept that Scum has to fabricate reads to win the game. What my opponent is missing is that, on DP1, Town also has to fabricate reads. Like my opponent said, "the FIRST vote that someone puts down will probably be random". If this is so, there is no correct way to react. If something is random and arbitrary, any responses to it will also be so. As of yet, my opponent hasn't been able to give a substantial defence of his idea of Townie play and why it is different from Scum play outside of just saying that it is. He has not shown that Town is even able to non-randomly act to fulfill their win-con on DP1.
I'll start with definitions. "Random" means acting "without definite aim, direction, rule, or method"; "lacking a definite plan, purpose, or pattern."  "Rational" means "based on facts or reason and not on emotions or feelings." 
Town act with a definite aim, direction, and purpose. Their goal is to find the mafia and lynch them. Even when town act irrationally (i.e. based on emotions or feelings), they still act with that specific intent. The idea that town always act randomly is simply untrue.
The idea that behavioral analysis cannot be rational is also untrue. The differences between town and mafia are "facts." Votes, reads, and claims are "facts." Themes are "facts." These "facts" provide a "rational" basis for analyzing the game. The behavioral analysis I cited earlier is an example. It refers to the specific facts in the game, and provides reasons (not emotions or feelings) why those facts lead to a certain conclusion about alignment.
NOTE: Bossy's argument rests on a logical fallacy. He conflates the words "random" and "irrational" (he says town plays randomly because they cannot justify anything they do rationally). But irrational acts can be and often are non-random. Bossy gives no reason to think otherwise.
== What if town always play randomly? ==
For the sake of argument, let's assume town always play randomly. Sometimes, these random actions will look scummy. Sometimes, these random actions will not look scummy. That's the nature of randomness, right?
So, sometimes, town will do random things that mafia would not do (because mafia do not act randomly). That means, at the very least, town can use behavioral analysis to locate other townies via town-tells. Under this logic, there's no such thing as a scum-tell, but there is such thing as a town-tell.
If you can find all the townies, you've effectively found all the mafia through process of elimination. The net result of analyzing behavior is thus positive even if we assume town always act randomly.
NOTE: I'm not conceding that town always play randomly. I'm just testing out the hypo to show that bossy's argument is wrong and his framework incoherent. I say the framework is incoherent because the moment town start locating town-tells, town aren't acting randomly anymore, so the framework collapses.
== Begging The Question ==
"Begging the question" means "assuming the conclusion of an argument" (i.e. circular reasoning). Bossy seems to love the phrase. For almost everything I argue, he claims I'm begging the question.
The irony of bossy's position is that he's begging the question, not me. This is bossy's argument: "Town and mafia behave the same because town play randomly. Town play randomly because they cannot distinguish town behaviors from mafia behaviors. Town cannot distinguish town behaviors from mafia behaviors because town and mafia play the same." That's circular reasoning.
The argument I made is not circular. I begin with the facts. Town do not know alignments; mafia know alignments. Town must lynch; mafia need not lynch. Town need not fake reads; mafia must fake reads. These facts are not in dispute.
From these facts, I argue that town and mafia act differently. I apply common sense. When town read the game, they're looking for the mafia. When mafia read the game, they're looking for townies to mislynch. That means town and mafia will interpret the game differently. The result is that town respond to new facts (votes, reads, and claims) differently than mafia.
I'm assuming that anyone reading this debate has played mafia. Just recall your experience as town and your experience as mafia; the two experiences are entirely different. The way you think about the game is different. I argue that town and mafia manifest these differences in their behaviors. This is common sense. I support the argument with a specific example (e.g. the degree of certainty tell). I could cite many more examples.
Finally, I argue that if town and mafia manifest different behaviors, then town should be able to locate these differences. If town and mafia play differently, then those differences should theoretically be identifiable. The data supports that thesis, since town's that use behavioral analysis do better than those that don't use it (bossy concedes this point earlier in the debate). There are countless examples of games where town destroys the mafia on D1 (and the rest of the game) using only behavioral analysis. That suggests that town can indeed locate town behaviors and mafia behaviors with better success than random.
At no point does my argument beg the question. Only bossy's argument uses circular reasoning. My argument flows logically from the initial facts I started with, which themselves are not in dispute. By contrast, bossy's argument never starts with any facts. Instead, he starts with conclusions ("town always play randomly," or "town and mafia play the same"), which he never justifies except by reference to other conclusions.
== How to start D1? ==
When the game first starts, the town needs to do something to get the game moving. They can do this in different ways, but the most popular way is a random voting stage ("RVS"). We describe the votes at this stage as "random," but usually players attach some "reason" for their vote. For example, someone might say, "FourTrouble sounds like trouble, so I'm voting him." Or they might refer to a player's history: "Khaos is always mafia, so let's start there." Or they might apply a rule (e.g. the inactive rule or noob rule).
These votes are not based in anything from the game yet, but they do flow from players having different alignments and goals. Sometimes, these RVS-votes reveal stuff about alignments because mafia are trying to blend in, whereas town are trying to get the game moving. These different goals can and do create differences.
Even if no differences emerge during RVS itself, responses begin to create meaningful information. Town and mafia will interpret RVS-votes differently simply because their perspective on the game is different. The mafia know which votes are wrong and which votes are right; the town don't. The perspective on these initial random votes is different, so the responses to them are (at least sometimes) different.
This eventually leads to a meaningful discussion about behaviors. That discussion itself produces more meaningful discussion. Eventually, RVS is the further thing from your mind. The players are pushing votes on people they ACTUALLY think are mafia, or mafia are pushing votes on people they're ACTUALLY trying to mislynch. At that point, nothing is randomly anymore. Town and mafia each have specific goals they're trying to accomplish, and that leads the town and mafia to manifest differences in their behavior.
Bossy says all responses are random, but like I explained earlier, that's simply not true. Town will respond to RVS votes by looking for scum. They look for something they can start meaningful discussion about. Examples of RVS moving into meaningful discussion are everywhere. You can look at almost every game where town lynched on D1. Or any game on mafiascum. Or F-16's game cited earlier. The town starts with RVS and eventually moves into meaningful discussion. At some point, folks push lynches they believe in. And that's the point where nothing is random anymore, because that's when you can actually analyze the motives behind those pushes.
Re: Definitions, etc.
This is my argument summed up:
P.1 That which is random cannot be known to be aimed neither for town nor scum
P.2 Town DP1 actions are random
C.1 Town DP1 actions cannot be known to be aimed neither for town nor scum
P.3 For behavioral analysis of actions to be useful, there must be a distinction between factions – it must be knowable that town must aim its actions for town and scum for scum
P.4 It is not knowable that town aims its actions for town on DP1
C.2 Behavioral analysis is not useful on DP1
Finding which actions are town-sided can only be done if the players that make up town are already known. From the perspective of a town player on DP1, no action can be known to be pro-town or pro-scum prior to a flip. As such, you cannot base reads off of this – to base reads off of pro-town actions, you would have to first be able to read players to determine which actions are pro-town, and to do that you would have to have already established a standard to read people by, etc. to infinity. This is clearly absurd.
It is a truth that, because of the very nature of town (being uninformed players), there is no solid basis for reads. This is not circular – my conclusion is based on this near-a-priori fact about the game. There is no justification for the idea that an individual townie having information regarding himself is enough for him to pass any judgements on the alignments of other players, and, if this is so, any such judgements must be random.
My opponent’s argument is predicated on the idea that “Town need not fake reads”. I have shown this to beg the question – for town to do anything other than “fake” (pull out of thin air) reads, they must first be able to read in the first place, which is absurd if the entire justification for reads even existing is that townies do not fake reads. It’s entirely circular. It doesn’t matter what town and mafia have as goals – what matters is if you’re able to distinguish between the actions each take to reach those goals. Since I have shown that Town acts randomly, distinguishment of this sort is impossible, and, as such, my opponent’s case collapses.
My opponent’s examples are irrelevant if he is not able to show a causal link between behavioral analysis and winning. He has only shown a correlation, which is not sufficient to rebut my arguments.
== Re: How to start D1? ==
It doesn’t matter who people “think” are mafia – what matters is what they actually post, and the reasons for posting what they post are essentially random.
|Who won the debate:||-|