Memories of a Dead Seer: Werewolf at Foocamp08!

Werewolf at Foocamp08
Originally uploaded by nickbilton.

Foo Camp, a.k.a "shangri la for geeks", is, among other things, the premiere laboratory for Werewolf hacks.*

We get to play A LOT of Werewolf. Typically at least 12 solid hours of Werewolf play, divided over two nights. This year, we played 10:oo PM until 3:30 AM (Friday) and then 11:00 PM until 6:30 AM (Saturday). It's the perfect way to interact with people who might otherwise intimidate you with how awesome they are, and to crush your own introverted instincts if you have them. That's why, I think, it's become such a staple at tech and geek events.

But nothing compares to the Werewolf @ FOO. We play it iteratively -- usually more than one circle is going at at time, in nearby rooms, and we move across circles and rooms, ultimately playing many, many games together in different combinations and size groups. We also play with a lot of history -- many of us have played dozens or even hundreds of games with each other, and we have a lot of information accumulated over those games that we use to try to analyze each other.

The end result is an extremely high level (you might say professional level!) of Werewolf gameplay. A lot of people stand around watching the games like a spectator sport (cool) -- at least 'til midnight (And by sunrise, it's only the hard-core players playing). Newbies get brought up to speed VERY quickly and are expected to play quite well (and actively) or get picked off early by the villagers ("Lynch the quiet people! They're not contributing!"). The expectation for elite gameplay can be seen on the Foo schedule in the photograph "new players welcome; experts really welcome!!"

It varies from conference to conference and year to year how well we hit the "sweet spot" of newbie/expert balance, but if you're curious, after this year's Foo, I can say pretty confidently that we've determined the ideal ratio for Werewolf innovation is between 25% and 33% first-time players in any given circle at the start of the night, and working it down to less than 10% first-time players in any given circle in all subsequent rounds. It's good to have some first-time players around because sometimes they think of weird ideas unbiased by previous experience. It also makes it more challenging for experienced players, since the new players' tells haven't been learned yet. (Yes, Werewolf players have tells just like poker players -- and learning to control and misdirect others with your tells is one of the most important meta-stratgies!) Most importantly, to spread the Werewolf phenomenon, you have to let new players play! And a big part of the fun is finding out who is surprisingly awesome. You never know who will turn out to be an amazing player... Last Foo, I thought Kati London was the big surprise star player. And this year, for instance, I was particularly impressed by the gameplay of newbie Nick Bilton (R & D/Future thinker for the NY Times).

As may already be apparent from the discussion in this post, in addition to the gameplay there's tons of theorizing. At previous Foo Camps, we've held extremely useful informal game theory lunches to discuss and create charts and tables of "optimal villager strategies" and "optimal werewolf strategies" and most controversially "optimal seer and healer strategies" (often very dependent on each other). We always say we're going to start a wiki to share these, but we haven't yet. Last year, we held a formal session to discuss Werewolf game strategy and thought just a few people would show up for an intense conversation; instead we had one of the biggest rooms packed and overflowing with people on the floor and out the door. I might be crazy, but I don't think it was JUST curiosity about game theory -- I think there's actually a little bit of social currency and prestige involved with being a good Werewolf players in the tech community. I don't think that's the main motivation of Werewolf players, but you certainly do have an easier time striking up conversations and meetings with fellow players if you do something clever in a game. For me, I usually throw out business cards I get at conferences (whoops, did I admit that?) but I always Google/Facebook friend/Twitter follow people who were interesting Werewolf players.

So, what happened at our Werewolf hacking lab @ FOO this year? Well, last year, at FOO ’07, Avi Bryant and I worked out on paper an Ultimate Optimal Villager Strategy for a Small Village playing “no reveal” (12 or fewer players, with both a seer and a healer). This is basically a PERFECT strategy that would work ruthlessly well to detect and lynch all of the Werewolves every single game, in almost any circumstance. We did all the math, we ran all the scenarios, and then we tested it in a bunch of games with lots of different players. And in ~20 games, the villagers won every time. (Since then, I’ve only seen the Ultimate Optimal Villager strategy fail once, more on that to come).

Here’s how it works. But before I walk you through the logic, please note: this is perfect from a GAME THEORY perspective only. It only works when 1) all of the villager players accept the premise and agree to play according to this strategy and 2) all of the villager players are acting rationally, in the best interest of the village. As we all know from attempting to apply game theory to real life, people are often irrational and don’t follow optimal strategies. So, while villagers win 95% of the time in an ideal mathematical/game theory world, in reality I would say that irrational actors and recalcitrant healers could probably drop the success rate as low as 80%, but only if at least several players were acting like complete and total idiots in combination.

Okay, so the BEST village strategy EVER:

1) The HEALER must heal him or herself on the first night. This ensures the healer is alive on day one, and everyone in the village will know that.

2) The SEER must reveal him or herself on the first day, immediately upon awakening, assuming they have not been killed. (WEREWOLVES have only a 1/10 or 1/9 chance of successfully killing the SEER randomly on the first night, so in the vast majority of games, the SEER is still alive.) There should be no discussion, no thinking, the SEER must simply REVEAL their identity. They should also say what they learned in the night (who they investigated, and what they found out.) At this point, one of two things will happen: Someone else will claim to be the SEER, or no one else will claim to be the SEER. If no one else claims to be the SEER, no problem, the village trusts the SEER. (Keep the other option in mind for a moment, we’ll come back to it)

3) At this point, the OTHER VILLAGERS should accept that this is the real SEER and trust all information the SEER provides for the rest of the game. The SEER becomes a de facto leader of the village. The VILLAGERS can advise the SEER on who to investigate based on their suspicions.

4) The HEALER must heal the SEER every single night, no matter what.

5) If the VILLAGE attempts to lynch the HEALER, the HEALER should out themselves as the healer only as a last resort if it looks like they are going to lose the vote. They should plead not to be lynched, hopefully save themselves from the lynching, and then alternate between healing themselves and the SEER randomly each night, thwarting WEREWOLF efforts to get one of them and prolonging the number of rounds the seer has to investigate.

6) The SEER must reveal what they found out straightaway every single day, no matter what. The VILLAGE should reseat itself, for visual clarity: safe “investigated” players who are proven villagers sit together with the SEER, uninvestigated/unsafe players sit together awaiting their fate.

7) If the WEREWOLVES got the SEER on the first night, then normal Werewolf odds apply. This would be 100% effective if the SEER couldn’t’ be killed on the first night; averaging normal village success rates (~55%) with the perfect success rate at the right weights (90% of the time the SEER isn’t killed the first round) results in a success rate of 95.5% for villagers under this strategy, NOT allowing for awesome and unlikely Werewolf counterstrategy. Which goes as follows…

If a VILLAGE plays this way, the WEREWOLVES have only a few viable strategies to stop their inevitable discovery and lynching.

So here’s the counter-strategy that rarely works, but are the only viable options if a VILLAGE really has its game together:

A) It goes without saying that the Werewolves must always kill “investigated” players so that the pool of “uninvestigated” players remains as big as possible, for them to hide in.

B) After 2-3 nights, they should attempt to kill the seer in the hopes that the healer has died. This is really their only chance to get far enough in the game that enough uninvestigated villagers remain.

C) One of the WEREWOLVES can claim to be the SEER immediately upon waking up, either before the real SEER does, or right after, claiming “WAIT A MINUTE! But I’M the real SEER…” At this point, they have to out-perform each other to earn the village’s trust, and in a best-case scenario, the Werewolf has a 50% chance of winning the village’s trust. The villagers know one is the real SEER and the other is almost certainly a Werewolf. (Well, they could be a drunk, reckless villager – see the note on “irrational players” and game theory above. ^_^)

From my experience, however, this gambit almost always fails. Here’s why. 1) Most Werewolves REALLY don’t want to draw attention to themselves, so they VERY rarely claim falsely to be the Seer, even though it means they are doomed not to claim it. In that first day, they have a powerful instinct to try to “fly under the radar” and an aversion to being called out right away. So you very rarely see two battling Seers. However, if they do take the risk, villagers almost ALWAYS can tell the difference between a real seer and a lying werewolf. You have to trust my observations on this, or try it yourself. It is pretty easy to tell the difference in the first daytime between an honest seer and a lying Werewolf. The Werewolves almost always fail on this gambit.

Having played nearly 100 games with the Ultimate Optimal Villager strategy, I have only ONCE seen a Werewolf play this strategy and pull it off. (In games where the village isn't playing by this strategy, it's actually quite common for a Werewolf to successfully claim to be the Seer.) It will probably hurt me in future games to admit that this was a game in which I was the Werewolf and Jimmy Wales was the Seer and investigated me on the first night. So, um, forget that I said that. There is ONE advantage the Werewolves occasionally gain by this gambit, even if the villagers correctly identify the Werewolf as a liar. In an imperfect world, the villagers will decide at this point to abandon the strategy and lynch BOTH the so-called seers, knowing they will get at least one werewolf out, guaranteed. This is not technically a valid move in the optimal strategy, which says you just HAVE to play as if you believe the seer is the right seer. But it happens, and it mildly favors the villagers, so even in an irrational world, it still helps to have the seer come out round one.

D) Alternately, a Werewolf can claim privately to be the HEALER to the SEER, through whispering or eye contact or such. The advantage that could be gained here is that the SEER might forestall investigating the Werewolf because the SEER now trusts that person and concentrates on investigating others. I’ve never seen this work, but it theoretically could improve the Werewolves’ odds of surviving long enough to get the SEER and outplay the rest of the VILLAGERS.

(by the way, how awesome would it be to teach a beginning game theory class using WEREWOLF instead of that idiotically simplistic prisoner’s dilemma?)

So, the cool thing about FOO is that more than half of attendees from year to year are new, and hardly anyone in my Saturday night Werewolf circle had been persuaded of the Ultimate Optimal Villager Strategy the year before. So I had to try to persuade a whole new set of players of how amazingly effective it was. THIS WAS A HUGE CHALLENGE. Mostly because it’s totally conventional wisdom that the seer and healer should keep their identities secret as long as possible, so they don’t get eaten by werewolves.

So, one game, sometime after midnight, I got the Seer card in a small village, was nominated for lynching the first round, and so of course I outed myself and said “But don’t worry! This is fine! This is perfect!” At which point I tried to introduce the other players to the Ultimate Optimal Seer Strategy. It was NOT well received! OMG. They thought I was crazy, crazy wrong. It was SO counter-intuitive. (And hard to walk through all of the game theory and get people to listen and not seem crazy while actually playing a game, lol.) They went through all the arguments: It was too dangerous for a seer to come out in the first round, you couldn’t count on the healer to heal them, the werewolves could be too tricky and the village would fall for their lies, and so on. And naturally, the healer refused to heal me that night, the werewolves got me, and on it went. I was so mad that I scrawled across the whiteboard “THE HEALER NEEDS TO LEARN MATH!!!!!” in the middle of the “night”, and once the healer got picked off, he and I and another dead villager went out in the hall and had a raging argument. REALLY raging. Friendly, but wild. The argument was only settled 5 games later... five games in which we had agreed to test the strategy and saw the villagers win perfect games lasting about 10 minutes each (that’s really short!!!). In fact, after that point, when it was conceded by all that from a game theory perspective this was really ridiculously effective, we had to stop playing with that strategy. It was too boring to be that good of a village! (And really stressful to be a Werewolf in that environment) So we told the Seer to do something crazy, abandon that strategy, and on we went for another few hours…

I could write for hours more about the interesting Werewolf phenomena at this camp, but I’ll close with just one more story. With the help of co-conspirators Cal Henderson and Kati London, I decided to moderate a final game at 6 AM – what we called “The Meta Game”. I would put ZERO Werewolf cards in the deck, and NO Seer and NO Healer. But the PLAYERS would be told that there were 2 werewolves and a seer and a healer. They would each get a VILLAGER card and assume the special roles had been received by others in the group. I would go through the night as usual, “waking” people up, getting their input, and so on. I would just decide who to kill based on who was actively participating (I killed off the least active participant from the previous day each night.) So I faked the whole game that way. I killed Kati first so she could “welcome” each killed/lynched player and convince them to keep quiet while the rest of the game played out. Cal I kept alive a few rounds because he’s funny, but eventually I had to kill him so that the endgame could play out with truly “in the dark” players.

When there were only 5 players left, I decided to announce no deaths in the daytime, thus convincing them the healer was still alive and had saved himself or herself. They were quite flustered when no one would admit to being the healer! (They assumed the seer was long dead). They descended deeper and deeper into truly spectacular confusion, and we played to a final endgame of 3 people, which in normal Werewolf would mean that there was 1 werewolf in the circle and 2 villagers. They went nuts trying to persuade each other that they were really villagers and not werewolves. This was great, because they were all telling the truth! They were so infuriated with each other they were throwing food and such. Literally. Throwing food. At each other. It was awesome beyond awesome. They finally settled on one last person to lynch, at which point I perversely announced that the Werewolves had won. Meaning they would have lynched the wrong person. Their jaws dropped, they stared at each other, frantically trying to figure out, “Who was it? What just happened?” It was pretty priceless. At which point all was revealed. To much celebration. The next day, one of the final 3 players came up to me still excited and said he hadn’t been able to sleep for an hour and a half, he was so keyed up thinking about the final game. AWESOME.

I LOVE WEREWOLF.

* One of the first Werewolf hacks that the FOO/O'Reilly community has been really active in propagating is the fact that we call it "Werewolf" instead of "Mafia", which is the original variant's name but was re-proposed by this guy as Werewolf. I love this, because I spent about a year observing the difference in roleplaying strategies that emerge when people play "Mafia" versus "Werewolf" versus "Vampires" versus "Witchhunt" versus "Zombie Village." I find that Werewolf killing seems to be taken less personally -- players are less offended being eaten by Werewolves in the night than picked off by the mafia in the night. Because there is the potential for a little bit of actual bullying in a game like this and a lot of potential for hurt feelings (why did you kill me??!!!), I think it's important to stick with metaphors that de-personalize the process. Werewolf also seems to evoke the least amount of fictional role playing/posturing and the most game theory and real-person interaction. All of this probably sounds hard to believe -- why would the game metaphor change interaction so much, and is it really consistent across groups? I haven't done a scientific study, these are only ethnographic impressions, but I have a gut sense they are pretty valid. Chat me up about this sometime, I have lots more to say on the subject! At any rate, some of us would like to see wikipedia stop redirecting the Werewolf article to the Mafia article, but we haven't really made a good case for that yet to the wikipedia elite. Maybe at Foo '09 we will stage our wikipedia revolt!

UPDATE: First, in the comments of this post, some EXCELLENT ideas and strategies. Take a look! (And thank you to everyone for adding them.) I am particularly enamored of the "neighbor" strategy, which I have never encountered in Werewolf play. I will DEFINITELY be trying it at the next Werewolf night! Second, this is a really interesting analysis of this post from the angle of "What does it mean to 'solve' a game?" Maybe it's a little meta to link to it, but I'm just such a geek and who could resist a blog called "Geek Out New York"? Third, I just wanted to add that I love playing in different combinations of characters: Seer/No Healer, Reveal/No Specials, No Reveal/No Specials in REALLY small groups, like 5-7 (we call this "Speed Werewolf"), Medical Examiner (my own invention, play with me to find out how it works!) and so on... the great thing about Werewolf/Mafia/etc. is that there are endless variations, and EACH has its own optimal strategy. So the game theory and the geeking out never has to end!