Click here to participate. Entries must be submitted on October 18th, 2020 or earlier.
Entry is now closed.
In 2017, Zvi posted an exciting story about The Darwin Game, a variation of iterated prisoner's dilemma.
I will run my own version of the game in the week following October 18th, 2020. You do not have to know how to program in order to participate. I will code simple bots for non-programmers. If you do know how to program then you may create your own complicated bot.
Here are the rules. Changes from Zvi's original game are in brackets [like this].
For the first round, each player gets 100 copies of their program in the pool, and the pool pairs those programs at random. You can and often will play against yourself.
Each pair now plays an iterated prisoner’s dilemma variation, as follows. Each turn, each player simultaneously submits [an integer] from 0 to 5. If the two numbers add up to 5 or less, both players earn points equal to their number. If the two numbers add up to 6 or more, neither player gets points. This game then lasts for a large but unknown number of turns, so no one knows when the game is about to end; [I guarantee it will be at least 100 turns per iterated prisoner's dilemma].
Each pairing is independent of every other pairing. [You do know what round of the game it is and that you are facing an opponent. If you face a copy of yourself you are automatically awarded the maximum 5 points per round (2.5 points per bot). You otherwise do not know any history of the game to this point.] Your decision algorithm does the same thing each pairing.
At the end of the round, all of the points scored by all of your copies are combined. Your percentage of all the points scored by all programs becomes the percentage of the pool your program gets in the next round. So if you score 10% more points, you get 10% more copies next round, and over time successful programs will displace less successful programs. Hence the name, The Darwin Game.
Your goal is to have as many copies in the pool at the end of the last round as possible, or failing that, to survive as many rounds as possible with at least one copy.
[I will attempt to iterate until there is a stable equilibrium.]
I will add some silly bots of my own to make the early game more interesting.
Instructions for non-programmers
Please give a simple explanation of what you want your bot to do. Write it with mathematical precision. If your specification is even slightly ambiguous then you will be disqualified.
Instructions for programmers
Write a program of the following format.
class TitForTatBot():
def __init__(self, round=0): # Edit: Changed "1" to "0"
self.turn = 0
self.round = round
self.previous = None
def move(self, previous=None):
self.turn += 1
if self.previous:
output = self.previous
self.previous = previous
return output
else:
return 2
# Edit 2020-10-11: The above code is wrong. The properly-implemented TFT `move` method looks like this.
# def move(self, previous=None):
# self.turn += 1
# if previous == None:
# return 2
# else:
# return previous
Your class must have an __init__(self, round=1) intializer and a move(self, previous=None) method. You may write your class in Python 3 or Hy.
Unlike Zvi's original game, you do get to know what round it is. Rounds are indexed starting at 0. The previous parameter of the move method is an integer indicating what your opponent did last iteration. If it is the first iteration then previous equals None.
A new instance of your class will be initialized in each round. You may save whatever information you want into the class instance's fields but you may not save information between rounds or between class instantiations. The move method must always return an integer from 0 to 5 inclusive.
You may import standard libraries like random, scikit and numpy.
Coordinating with other players
Anyone can play, but only players with a Less Wrong account that existed before I declared this tournament will be allowed to coordinate out-of-game. This rule exists to prevent players from submitting multiple entries to this contest and self-coordinating. Coordinating with other people is encouraged. Coordinating with yourself between multiple separate entries is cheating.
Click here to participate. Entries must be submitted on October 18th, 2020 or earlier.
Entry is now closed.
What do you mean by information restrictions? What exactly is restricted? If I were to use some side-channel to determine that I was being simulated by an opponent and forkbomb if so, would I be eliminated for hacking? Would they be eliminated for overusing compute? (after all, it wasn't my code per say that forkbombed, but their code simulating my code) Would I be eliminated for being a dick? (Which would be self-evident, at any rate)
You say in another comment, "Simulator bots are legal for programmers to write provided they never crash, slow the game to a crawl, etc. Any bot (even a non-simulator) that does will be disqualified." I'm just concerned about the edge case where someone writes the troll code described above, and suddenly every person who submitted a simulator bot (i.e. Zak M. Davis) gets DQ'd. The rules as written suggest that such a bot (presuming it does not preserve information across rounds) would be perfectly legal and it's Zak's own fault for executing untrusted code, but that seems to effectively prohibit simulators as there's an infinite plurality of side-channels through which to do this.
A few possible resolutions would be:
Resolution 2 seems the most reasonable to me, but if you choose it I humbly request a way to locally overwrite get_opponent_source so you simulate your opponent's response without risking infinite recursion. (Such an overwrite is probably standard python, but being ignorant of your implementation I'd be concerned about screwing something up and not being able to detect the mistake in testing)
EDIT: Upon further consideration it seems inevitable that Zak will get DQ'd for infinite recursion. So I must be missing something.