Weekly Monday Discussion #6 - Computers

I do think there’s a pretty big difference between “multi-iso and vote compilation bot” and “word choice analysis”* (or similar things).

It probably wouldn’t be that difficult for me to write word choice analysis, but I’ve not done so because I feel it’s a bit too far into automated analysis**, and it may be very difficult to adapt to as a wolf (I can’t say for sure as I haven’t tried it, but I’d have to imagine so. There’s a reason it’s angleshooting on MU, after all).

Whereas multi-iso and vote compilation are about analyzing** things that are normally analyzed, and it’s certainly possible to adapt one’s play as wolf if a pattern were to be revealed (ie, if compiling votes shows someone rarely votes their partners, then they can distance more).

Graphing post delays is somewhere in-between, as post delays at large are not normally analyzed, but I feel it’s still fairly reasonable, since analyzing post delays in individual interactions is pretty normal, and it’s again possible to adapt one’s play as wolf (ie post faster lol).

Overall though, I feel it makes sense to judge programs case-by-case, rather than allowing or banning them in one group.


*Word choice analysis is collecting data about the frequency words are used as each alignment, and using that data to draw conclusions about the current game.
**Technically yeah this both of these are just compilation, not analysis

1 Like

imo this shouldn’t be allowed (or done) mainly bc you will always have a chance of dying* so if you schedule a psot like that you could be outing that you knew you wouldn’t die or you could be setting yourself up to break the rules and share reads after you die, which would be a huge no-no

*unless the game doesn’t have elimination, such as Combinatorics Mafia, where something like this took place :joy_cat:

4 Likes

ok wait. you’re obviously correct but that would be really funny

7 Likes

oh I mean it’s possible to detect that stuff

scheduled post for [phase change] o’clock: “helo i am back”

8 Likes

(though I agree it shouldn’t be allowed)

I also agree that, like, multi-isos and vote count compilations are very different from detailed word choice analysis. I think the former should be allowed, and no program to do the latter currently exists because unfortunately Zugzwang has too many morals and hasn’t been corrupted by my dark whispering yet. Post delays are, like, on the line to me.

The tools to count votes and multi-iso already exist. They are here. They have been used in games with no significant issue of advantage over other players. No wolves are dying exclusively because somebody could find the list of votes placed in a game faster.

There are minor technical advantages provided by being able to do shit faster, but there are also advantages and disadvantage provided by… being on desktop versus mobile, being a hydra, being an established member of the site, access to previous games, refusing to share your tells publicly after games, refusing to tell other players how you read them, using the same opener every game, being at work (it’s kind of yikes to shade a dude who said he has a job pre-rand for being at work), etc etc et al.

These are all allowed because it is more trouble than it’s worth to disallow them! It makes the game less fun to try to Equalise Everything. Vote compilation and multi-isos are a convenience, and one that is available to everybody who can Use a Computer. Like, I don’t think we’d be having this conversation about Mafia’s Purity if… vote counts and multi-isos were just a Discourse feature that was only available on desktop. It’s literally fine.

It’s not easy to distinguish perfectly between “thing that just saves work that’s reasonable to do within the time of a day phase” (you can just ISO everybody individually, or count up all the votes, the program just saves you an inconvenience and gives you no major play advantage other than BEING LESS ANNOYED) versus “thing nobody could do manually within a mafia game’s time” (super in-depth word choice analysis, which gives an actual advantage to some players over others), but… just do it case by case lol.

I would be sad if “get votecounts fast” was banned because of slippery-slope arguments about The Purity Of Mafia. The two outcomes here are not “ban all programs forever” versus “mafia becomes a programming competition”. It is fine.

7 Likes

Imo just cuz you are on mobile doesn’t mean you can’t go on laptop and get access to laptop tools that are there for everyone if they are open to everyone like some of the suggestions

i don’t get how mobile exclusive for some people is w reason to ban it

Perhaps only tangentially relevant here, but

I am personally very skeptical about the validity of complicated, “sophisticated,” data-analytics-based FM reads.

Complicated multivariate FM models seem doomed to run aground to overfitting and over-parameterization due to the intractable issues of a tiny sample size and very noisy data.

Even for a truly prolific mafia player, you’re not going to have more than a couple to a few dozen wolf games from them. Add in all the noise with varying setups, sites, playerlists, wolfmates, life circumstances, and a literal moving target as the player adjusts their own wolf game…

I personally believe that given the tiny, noisy sample size, anything beyond univariate analysis of highly predictive metrics is going to be at best near useless thanks to over-parameterization. And at worst, actively detrimental as your misplaced confidence in the predictive power of your model makes you conf-biased and worse at reading the game. I think it’d take an AI that can actually semantically understand posts and the progression of the game before you start to see useful results, and like orange mentioned, by the time something like that even exists, it’s going to most likely take zero coding knowledge to abuse.

And as for the actually useful univariate analyses? There are already terms for them - they’re called things like “volume-tells” and being “RWSTFO” You don’t need coding skill to notice them, and if you’ve ever played FM, you’ve surely seen even those tells go wrong as well, lol.

But beyond those variables - I’m actually not personally aware of (but would love to hear about) any others that have held up to repeated scrutiny and weren’t just data-dredged, p-hacked, Baltimore Stockbroker findings.

So, given all that, banning or restricting tools that assist in data analysis in FM feels kind of silly.

Actually… maybe I support banning them because it saves future well-meaning, code-savvy, data-inclined FM players from themselves by sparing them from having to find all this out the hard way :joy_cat:

But I don’t actually play here, I’m just here for Word Match

6 Likes

So, I agree with Lumi broadly that there’s very limited data crunching that can produce limited results. Buuuuuuuut I know I certainly have a word choice tell (my word variety is much, much lower as mafia), and if someone is going to use that in a game against me, that’s not a game I want to play. The problem here is not that someone might have reads that are more accurate for some players, the problem is the potential to make a case that appears to be objective and science-y, and arguing such a case becomes not about the game but fuckin statistics. Mafia is a game where we all agree to get into a game where we lie to each other and participate in mutual torture, and then lie to each other afterwards by saying we had fun doing so. I don’t care if the tools and data to accomplish such analysis are accessible to everyone, I absolutely think discussion of it should be banned in game. Obviously you can’t stop people from using it, but if they can’t say "I think Katze is a wolf because ", and they have to actually case the player based on what they did in the thread, the problem of using data analysis largely goes away because the wolf has more room to maneuver their way out of the noose.

Scheduling posts should just not be allowed by anyone, but hosts. Using it for a mechanical advantage is one of those things that just sucks the fun out of the whole thing to some extent. Scheduling a post to :00 an ITA shot is just kinda slimy lol. Hosts should be cognizant of different timezones and do what they can to make mechanics as accessible as possible with that in mind, but sometimes your wolf partner is not gonna be online to take the shot and then gets domed by the villager you wanted to yeet and thems the breaks. Just completely unjustifiable to explicitly allow post scheduling.

Tools that improve search and/or multi-ISO, and more transparent timestamps are things that are both fine to be surfaced, and should be as accessible as possible. Building it into the site itself via plugins or whatever is ideal, but making it easier to find/read posts is not problematic.

9 Likes

would browser addons technically come under this
scheduling posts is a no go thats for sure but other stuff exists to change the format of site data and stuff
data analytics is kind of against the spirit of forum mafia too imo. like in comparison to irl mafia at least. you dont have statistics with you other than what you can eyeball there

I definitely agree about post scheduling and tool usage,

But I take major grievance when it comes to blanket banning statistics-based arguments. I very strongly disagree. I think statistics-based arguments have as much of a place in mafia as any other type of argument.

Banning them because most people think they’re not fun to play with, while it is a potentially valid decision for a community to make - doing what most people agree with and all that - feels to me more on par with:

  • Banning randing your vote or night actions because “it’s not fun to get caught when you had everyone fooled just because someone got lucky when they randed their night action, they’re supposed to choose, that’s the point of the game, and doing otherwise is against the spirit of the game”

  • Or banning discussing mech at all because “the game is about reading people, not about solving a mechanical puzzle”

(Paraphrasing both real examples from different people/communities, to not just be making up a strawman)

And while I don’t share those view points, they’re not wrong because they’re opinions and preferences about what kinds of things they prefer to see in their games. (Although randing your vote or action is pretty cringe, just don’t think it should be banned)

So looping back, speaking as someone who thinks that statistics arguments in mafia games are fun, the assertion that “mafia isn’t about that” is an opinion not an objective fact.

Statistical arguments have as much of a place as any other kind of convoluted argument. The problems you listed with them are your reasons for why you personally dislike them, and why many others also dislike them - but they’re not inherently problematic or unfair.

And now going to a more personal level - yes, only some people have the ability to make and understand those kinds of arguments, but similarly only some people have the time to hyperpost and flood the thread - should there be a site-wide ban on posting more than 200 times a day to make things more fair for everyone?
Only some people are speed readers and are able to keep up with thousands of posts every 24 hours, should there be a site-wide postcap of 50 per day phase in large games to keep thread length under control so that speed readers don’t have an unfair advantage?
You say that mafia isn’t about statistical arguments, I say that mafia isn’t about using game threads as chatrooms and flooding it with irrelevant chatter, does that mean that there should be a ban on off-topic discussion in game threads?

As someone that is a slow reader, that prefers more analytical games, lower post counts, more thought-out posts and less chatter - these are all real reasons that I retired from mafia. But given my preferences, if I were in a community where those preferences were widely held, should the community implement those bans that I mentioned?

The answer is - maybe?? It would be heavy-handed, but if it’s something that was widely agreed upon by the users of the site?

But all of those bans, including the use of statistical arguments, would have to be based on user preferences, not some objective “what mafia is about”, nor some “major integrity + fairness issue”

However, a ban on statistical arguments just feels particularly bad and unprecedented because some people are just much more analytically and statistically inclined, that’s just how they naturally approach problems and convey their ideas, it’s how they think.

So going ahead to ban statistical arguments outright would be because of a heavy-handed, relatively-unprecedented collective decision that “we don’t like playing with this kind of argument, so let’s get rid of it, people that are inclined to think that way should just play how we like to play.”

And honestly, given how ineffective statistical arguments are at convincing people that aren’t statistically inclined - having to translate your thoughts and analyses to something that the majority of players will actually connect with and listen to is already a big enough deterrent, no ban necessary.

6 Likes

To add on to this - as a slow reader, trying to get a feel for someone’s game when you’ve never played with them before is a nightmare. Just going back and reading their ISOs is not feasible.

Something I tried at one point leading up to my Champs year (but abandoned because it failed to demonstrate enough predictive power to be statistically significant, much less worth all the time it took, eventually having to settle on simpler metrics) was writing a web scraper to collect all of their past games and then analyzing that dataset, looking for trends I could apply.

So being told that “Your way of addressing your problem of being a slow reader isn’t fair, so we need you to either not do it, or if we allow you to do it, you have to lie about why you believe your reads.” is just suuuuper irksome.

(And ironically, despite how strongly I’m arguing for the validity of this approach, it’s been a very long time since I bothered touching any sort of data-driven mafia analysis)

4 Likes

Someone should run a game where people are only allowed to use LLMs to write posts

3 Likes

On a practicality note, I find myself agreeing with orange - if a player had access to such a tool that can “search or synthesize information accessible to anybody,” how will the hosts or any other players know if they are not voluntarily disclosed? From cyberiad overhaul, wrongboy (I assume the zugzwang half) had a fair assumption on my alignment based on time between posts, revealing a meta I had not known before; I would’ve had 0 idea that they were using this tool and would’ve thought “wow they’re a good player” (which, they are)

Question on rule 13 then:

I figure a better discussion on “AI” and “assist” and a possible clarification of the rules is required here…
Going way far off the deep end with orange’s stance, developing a ML program to predict a particular player’s alignment by some variables would fall under “synthesize information available to anybody” but at a much greater efficiency than that of any human on this site (and probably anywhere)

making tools that have a level of detail unavailable to a regular user is the issue no? a regular user cannot analyse like an ai. they cannot take in that amount of data
similarly with timed posts you have timing unlike any person
i say this bc i think there is a place for some tools but not others. autoformatting your posts would not be the same as automating a last second eod vote

where’s the boundary? for instance if I say “oh you post more as town,” it can be just from my experience, or I could’ve not known any of it until I ran an algorithm

theres a level of trusting people mandatory in this game. i also could’ve called up my buddy sal and asked him for his opinion and say it was my own it would still not be allowed

Shit

1 Like

I’m opposed to automated posting, but if a program were ever allowed to be used, it must have a 1/10000 chance of threatening to post your rolecard and then posting your rolecard for you.

5 Likes