A Reasonable Approximation

A list of Petrov buttons

A thing people sometimes do in relation to Petrov day, is to have a button. And if anyone presses the button, something negative happens.

I'd like to know how often the button gets pressed, so I compiled a list of all the times people have done this, that I'm aware of, and the outcome. Also, for the events I was at in person, I'm not sure there's an existing write-up of what happened, so I'm adding details.

Here's the list. As I learn about other events, past or future, I'll try to keep it updated.

2018, Oxford/Seattle: Failure

Parties happened simultaneously in these two places. I was at the Oxford one, which was actually camping in a field outside Oxford. We both had laptops connected to a web app.

What we were told: anyone at either party could press a button to launch a nuke at the other party. It would spend 45 minutes in transit, and then the other party would have to destroy a cake. But they'd be warned of launch, and be able to launch one of their own in the intervening period. I don't remember the time limit, probably like two or three hours.

What we weren't told: there was some chance of false alarms.

What happened: the Oxford party got notified of an incoming launch. We opened a Facebook chat with Seattle like "wtf guys". They convinced us that no one at their party had pressed the button. I don't remember if we'd considered the possibility of false alarms before messaging them, but given the context, they're not exactly something to dismiss out-of-hand.

We made it through right up to the time limit with no one pressing either button. My understanding of what happened next is that the time on the Seattle laptop ticked down to zero. Someone there pressed the button, which they now believed to be disabled, in celebration. The time on the laptop and the time on the server were slightly out of sync. The button was not disabled. A nuke got launched.

At any rate, Seattle convinced us that something along those lines had happened. We did not launch a retaliatory nuke. (This was still an option because we had one inbound.) We sent them a video putting our cake in the fire.

2019, LessWrong (+ Raemon's house): Success

125 people were sent codes that could be used to take down the front page of LessWrong for 24 hours.

Additionally, Raemon held a party at his house, where he gave everyone his codes and the party would end if LW got taken down.

No one submitted valid codes, though some people put in fake ones.

2020, LessWrong: Failure

This was a similar setup to last year, but 270 people were given codes (including myself). Additionally, there was enemy action. Someone registered an account named "petrov_day_admin_account" and sent a message to a handful of users:

You are part of a smaller group of 30 users who has been selected for the second part of this experiment. In order for the website not to go down, at least 5 of these selected users must enter their codes within 30 minutes of receiving this message, and at least 20 of these users must enter their codes within 6 hours of receiving the message. To keep the site up, please enter your codes as soon as possible. You will be asked to complete a short survey afterwards.

User Chris_Leong fell for it, and the site went down.

Postmortem from a LW admin. Postmortem from Chris, with comments from the attacker.

2021, Austin/Ottawa: Success

I don't think much info is publicly available, but it sounds like a similar setup to Oxford/Seattle, perhaps without false alarms. No one pressed the button.

2021, LessWrong/EA Forum: Success

This year, 100 members of LessWrong were given codes that could be used to take down the EA Forum, and vice versa. (This year they'd go down for the rest of the day, not for 24 full hours.)

No one did: LW retrospective, EAF retrospective.

2022, London: Success

This was at the ACX Schelling meetup, and most participants didn't know it was happening. I bought a button that I carried around with me, and a cake with Stanislav Petrov's face on it that I hid in my bag. (I didn't want it sitting out in case someone ate it without realizing, or decided it would be funny to mess with the game.)

Initially the button was locked. We had lightning talks planned, and I gave the first one. I opened with something along the lines of: "I have a cake with me. I think it looks very nice, and I'm looking forward to sharing it with you all. However! I also have a button. If anyone presses the button, I will destroy the cake. This isn't a joke, this isn't a trick, I'm not going to reward you, I'm just going to destroy the cake. puts mic down, unlocks button, picks mic back up The button is now live. You may be wondering why I'm doing this…"

Not long after, someone came up to give a lightning talk on the subject of "why you should press the button". What I remember mostly boiled down to "obesity crisis, cake bad", and I didn't find it convincing. But as a speech, it was much better than mine, even though I'd had days to prepare and he'd only had like fifteen minutes. I was not surprised when I learned he had competitive debate in his history.

Someone asked why he hadn't pressed the button himself, which was right in front of him. (I was carrying the button around with me, since it wasn't connected to anything and I wouldn't know if someone had pressed it otherwise. And I was sitting in front of him.) He said he believed in democracy and wanted a vote on the subject. Following this someone put a poll on a nearby whiteboard. Assuming no one cheated, it got eight no-press votes to two yes-press votes, at a party with over 100 attendees. (But probably no more than about 50 saw either of these talks.)

Ultimately, no one pressed the button. One person stroked it, and one person held his hand above it, but both said they weren't going to press it.

After 2 1/2 hours, I locked the button back up and brought out the cake. Once we'd started eating it, some people wanted to press the button so I unlocked it again. I didn't tell anyone this until afterwards, but it's a very satisfying button to press.

It seems a handful of people didn't think there was actually a cake, which I'm a little sad about. (It's unclear how many people, partly because "the cake is a lie" is a meme. But one person explicitly told me they thought that.) That's not the kind of lie I'd tell. Not that most people had any reason to think that about me, but I'm still a little sad about it.

2022, LessWrong: Failure

The plan was that this year, pretty much any existing LW user with non-negative karma would be able to press the button. (Which again would only take the site down for the rest of the day.) But only high-karma users would be able to do so all day; every hour, the karma threshold needed to press it would lower. Also, unlike previous years, someone who pressed the button would remain anonymous.

What actually happened was that users with exactly 0 karma were also able to press the button from the start of the event. One of them did, two hours fifty-five minutes after it began (but only 1:50 after the password needed to press the button was actually published). Habryka figured that out without learning their identity, and the site got reset for a do-over.

Then the whole site went down briefly when someone tried to fix a different Petrov-day-related bug.

And then finally it went down after 21:33. Presumably this time the button was pressed by someone who was supposed to be able to. There were 1,504 such users.

I had bet on Manifold that the site would go down before eight hours were up. (Specifically I bet the market down from 34% to 30%.) Despite the bug I consider my prediction to have been wrong. I'm happy about that.

2022, St Louis: Success

I don't know much detail and don't know if it's been written about publicly. But what I'm told is that if anyone had pressed the button, the party would have been over and everyone would have had to go home. No one pressed the button.

2022, Champaign-Urbana: Success

LW user benjamincosman says:

I ran a very small Petrov Day gathering this year in Champaign-Urbana. (We had 5-6 people I think?) I put a Staples "Easy" button on the table and said if anyone presses it then the event ends immediately and we all leave without talking. (Or that I would do so anyway, obviously I couldn't make anyone else.) No one pressed the button.

2023, Waterloo: Success

Joint event between the Waterloo, Ontario rationalist and EA groups. Jenn, the organizer of the rationalist group, writes:

• this was a meetup we had in conjunction with the Waterloo EA group on September 27th, 2023.
• we had 2 buttons, one for a fruit platter and 1 for a cake. We divided people into 2 halves of the room based on if they wanted fruit or cake, and then explained the typical rules. (this was partly inspired by your write-up of the uk ceremony one year when someone gave an impassioned speech about why cake was bad)
• the buttons stayed unpressed for the duration of the x-risk and petrov day presentation given by EA Waterloo
• afterwards we dug in.
• people did say that we probably should have given more opportunity for button presses after the presentation during more free social time, we'll do that if we run this ritual in future years

Things I'm not counting

Here are some things that I don't think belong on the above list, but I do kind of expect someone to point me at them if I don't mention them.

• 1983, the actual Petrov incident - IDK, apart from anything else it feels like it would be kinda tactless to include on this list?

• 2021, Jeffrey Ladish did a thing on Facebook, but the dynamics seem importantly different.

Posted on 26 October 2022

Tagged: game theory

Comments elsewhere: LessWrong