search

carykh / PrisonersDilemmaTournament

Watch This Place's awesome video about iterated Prisoner's Dilemma for context! https://www.youtube.com/watch?v=BOvAbjfJ0x0
MIT License
205 stars 160 forks source link

[NOT A COPY OF #34] We need an official statement about "colluding" or "handshake" strategies. #40

Closed shadowleafy closed 3 years ago

shadowleafy commented 3 years ago

The focus of the prisoner's dilemma is that both sides should be attempting to get the highest score, with no prior communication. Colluding strategies completely break this premise, allowing people to get ~4.5 scores with essentially no effort. Ban collusion.

nobody5050 commented 3 years ago

See #34

shadowleafy commented 3 years ago

See #34

It's not that hard to make ~100-ish google alts. There's nothing stopping someone from doing it. It has to be banned or the contest will be ruined.

nekiwo commented 3 years ago

Yeah we need some official word on this

shadowleafy commented 3 years ago

Yeah we need some official word on this

Agreed. Should I rename the ticket so it doesn't just look like a duplicate of #34?

cai-lw commented 3 years ago

I have a feeling that the rules were made intentionally vague. Maybe tempting people to cheat is also part of the research.

Also it really isn't hard to prevent cheaters from actually claiming the money. They can just read the code of the top performers and require them to explain their code. Even in case of uninterpretable code like a neural network they can still ask for training data and training code.

duckboycool commented 3 years ago

There's this from the FAQ on the announcement video in terms of an official statement.

Q: Can't people flood the pool with dumb strategies that boost their own true strategy? A: Perhaps! But, my Google submission Form only allows for one submission per Gmail account. So, if you have multiple Gmail accounts, I think you could submit multiple, but I'm hoping Google has been hampering down on bot accounts over the last decade or so!

shadowleafy commented 3 years ago

I have a feeling that the rules were made intentionally vague. Maybe tempting people to cheat is also part of the research.

Also it really isn't hard to prevent cheaters from actually claiming the money. They can just read the code of the top performers and require them to explain their code. Even in case of uninterpretable code like a neural network they can still ask for training data and training code.

Fair enough. I still would like an official response that isn't "it's discouraged" for winning the competition.

@duckboycool Saw that, but the way it's worded is vauge and seems to suggest that it's plausible but cary thinks it can't be abused. This thread is about how it could be abused.

nobody5050 commented 3 years ago

I believe Cary made a comment in 34 discouraging it

nobody5050 commented 3 years ago

Never mind I was mistaken

carykh commented 3 years ago

Yeah, colluding strategies are not allowed! When I'm reading through the strategies, if I see strategies putting arbitrary codes to check for friendly opponents who are in the same collusion, then I'll disqualify those strategies.

I'll put a QnA entry about that in the YouTube description now.

shadowleafy commented 3 years ago

Yeah, colluding strategies are not allowed! When I'm reading through the strategies, if I see strategies putting arbitrary codes to check for friendly opponents who are in the same collusion, then I'll disqualify those strategies.

I'll put a QnA entry about that in the YouTube description now.

Thank you for the response! I will be closing this now.