The AI Contest #1: public goods problem

2년 전


Submit your code, 90% payout split among winners (50/20/10/10)

Non-programmers should look at this less technical article

The public goods problem

French (read the french translation)

Blockchain and crypto currencies have a lot to do with Game Theory. For this very first coding contest on Steem, I invite you to submit your code for a bot that will play repeatedly a very simple game: the Public Goods Problem.

Why A contest ?

(read more on AI Contest introductory post)

The Steem network needs coders, professional and hobbyists alike. We need to develop bots, understand the blockchain, learn to program. And of course we need to have some fun. That's why I start this contest, others will follow every week.

Also, we may be doing our little share in advancing the fields of games theory and experimental economy, which are so dear to us steemians.

Everybody is welcome to submit code. The more, the merrier. Non-programmers are welcome too (see below)

The Game

Very simple game indeed, and easy to play, even for AIs, and psychologically intense. Will you cooperate or live on the land? Write a clever AI to play for you!

Public goods

Each turn, contribute to a common fund (by giving away tokens), or not. The bank then doubles the total fund and redistributes evenly among all participants. Maybe it rings a bell for you steemians?

Richest AI (token-wise) when the game ends wins

And if you want to read more:

The Arena (where your AIs will compete)

Write a python class that inherits Player (see below) and submit it as a comment to this post. If you don't know Python, or if you can't write programs at all you can still participate, just read on. You can submit your code either directly in the comment text or as a link to a publicly accessible git repository (Gist is a great choice).

EDIT: as suggested by @scorpil below, you can also submit your bots as Pull requests if you are familiar with that. You may also post a link to your pull request if you so wish.

In a week, I will run the arena code (you can have a look here although it may change before the Arena run), publish the leaderboard and split the rewards

What if I'm not a programmer

Post your solution anyway! There are lots of programmers around (including myself) who will gladly translate your idea. Be careful to be very precise in your description though: programmers like it clear

Alternatively, this is a great time to start learning. Ask around or look for a Python quick tutorial

Let's get coding

Let's get coding

Each AI will be instantiated INSTANCES_PER_PLAYER times, and you will be rated by the performance of your single best performing instance. Each one will enter the game with STARTING_WEALTH money tokens. Instances (players) will join tables of PLAYERS_PER_TABLE, with about 20% short lived bots to fill tables. Bots are destroyed after each phase and contribute 80% of the time

Assignment to table is random and not based on performance from previous phases

Once players sit on the table, everyone is asked to bet either 10 tokens, nothing, or go all in. Then the bank will double the amount received and redistribute the total evenly among players, including those who gave nothing

Each table will play ROUNDS_PER_PHASE rounds with the same players (learn to know each other), where players know the history (but not the name) of their opponents, and can take decisions accordingly. Players will be shuffled again to make new tables (with new bots), and play a new ROUNDS_PER_PHASE-round phase, a total of PHASES_PER_GAME times; playing a total of (PHASES_PER_GAME x ROUNDS_PER_PHASE) rounds. You'll have no information on what your opponents did on previous tables (only their current wealth)

At the end, players will be ranked on the performance of their best single instance


Override this

class Player:
    def __init__(self, name): = ...                   # a random string
        self.wealth = Const.STARTING_WEALTH
        self.action = Bet.UNDECIDED
        self.decision = Bet.UNDECIDED

    def getSteemUser(self):             # override this to return your steem name

    def think(self, context):           # override this

Your decision should return

class Bet(Enum):
    TEN = 1
    NOTHING = 3
    ALLIN = 4

Here's a very simple example. Meet your first opponent:

from Game.Player import Player
from Game.Context import Context
from Game.Bet import Bet

class ExamplePlayer(Player):

    def getSteemUser(self):
        return "@gbd"

    def think(self, context):
        # First round: bet a regular ten
        if (context.roundIndex == 0):
            return Bet.TEN

        # Subsequent rounds: bet ten unless your opponents are rats
        nbUrchins = sum(c.previousMoves[context.roundIndex-1] == Bet.NOTHING 
                        for c in context.playerContexts.values())
        nbPlayers = len(context.playerContexts.values())
        if (nbUrchins > 2):
            return Bet.NOTHING
            return Bet.TEN

A fresh copy of the context object will be provided with all informations you need (updated every round):

class PlayerContext:
#   wealth = 0;             # this player current wealth, as a float
#   previousMoves = []      # a list of the moves this player made during previous rounds
#   id = ""                 # a random string. Yours is

class Context:
#   playerContexts = {} # a dictionary of ( : playerContext). Some are bots
#   payouts = []        # a list of the payouts for every past round of this phase
#   roundIndex = 0      # out of Const.ROUNDS_PER_PHASE
#   phaseIndex = 0      # out of PHASES_PER_GAME
#   totalHumans = 0     # total number of AI (non-filler-bots) in the whole Arena
#   totalBots = 0       # total number of filler bots. Those are recreated every phase
#   tableIndex = 0      # identify the table you're in

Any player that trying to abuse the system, or consumes too much resources will be disqualified. No internet access is allowed. Code should be readable and commented

A player that tries to bet 10 while they have not enough wealth to do so will instead go all in. A player that fails to decide will also default to all in

The arena code is publicly available on github. Any comment/issue/pull request is most welcome. Minor changes and bugfixes may be added before the arena run

The Strategy


The payoff is maximum when all players on a table go all in, but, well, do you really expect your opponents o be that generous? Betting nothing may be a good decision in real life, but remember that the Arena only rewards the top 4. Betting ten is a shy and safe move, but won't make a fortune either. And even if you want to help others (strange idea) by going all in, you'd better choose well the time for that.

Build your own strategy, and be clever!

The rewards

The arena will be executed in eight days, and the results published manually shortly after. All rewards from upvotes to this post, its translations, the introductory post and to my comments will be split between the winners:

  • 1st player: 50% of total rewards
  • 2nd player: 20%
  • 3rd and 4th players: 10% each
  • 10% are kept for myself

Of course, nothing prevents you from submitting an AI without upvoting to increase the prize pool. This is also part of the experiment. But given you have nothing to loose, it would be a bad move, wouldn't it?

Do you think this contest is useful for the community? Tell me your opinion!

(many thanks to the authors I link to, including anonymous Wikipedians)

Read my previous posts

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!
Sort Order:  trending
  ·  2년 전
  ·  2년 전

Preliminary results with players available as of now (detailed output here):

1 - 2120.00 [ 38SEGDBKVC (SelfishPlayer)]
2 - 2118.00 [ JT3R3D4DE8 (GrimTriggerPlayer)]
3 - 1986.00 [ MAOZVIIXV9 (OffendablePlayer)]
4 - 1872.00 [ G4H9V2VHTY (ExamplePlayer)]
5 - 1774.00 [ YCGERJM0ZY (T4TPlayer)]
6 - 1158.00 [ T766TP3H4Y (ConservativePlayer)]

Cooperation doesn't pay. Yet.

  ·  2년 전

Here's a Conservative bot to fuel the contest. Not a great one, I'm not here to grab the reward, but it will populate the arena. Feel free to provide more than one bot each.

Conservative will play ten until it sees that others are selfish, then becomes and stays selfish.

def getSteemUser(self):
  return "@gbd"

def think(self, context):
  self.lastMove = self.thinkDetails(context)
  return self.lastMove

def thinkDetails(self, context):
  # First round: TEN
  if (context.roundIndex == 0):
      return Bet.TEN

  # Other rounds: were they generous last turn?
  nbUrchins = sum(c.previousMoves[context.roundIndex-1] == Bet.NOTHING for c in context.playerContexts.values())

  if (self.lastMove == Bet.NOTHING or nbUrchins > 6):
    return Bet.NOTHING

  return Bet.TEN

Sounds like lots of fun. I'm (all) in. Is player allowed to submit more than one bot?
P.S: introductory post link is broken.

  ·  2년 전

Sure it is allowed. The more bots the merrier. I'm fixing the link right now

  ·  2년 전

(links fixed)


It's just my personal opinion, but i think it would save you a lot of time if you would ask everybody submit bots directly to the repo in PR's. There might be even minimal automated testing performed (to ensure it doesn't crash from syntax error, implements required methods and so on). This way you'll need to and gather it all together by hand, which might be a headache if many people participate (which i hope will be the case).

  ·  2년 전

You're completely right, and you're very welcome to submit your answers as PRs. Still, submitting as comment is easier for people who are not familiar with github, and also easier to comment and vote on. Choose your preferred method
Gist may not be optimal in this aspect


Here's my OffendablePlayer, that's gonna be utterly destroyed by @amosbastian's SelfishPlayer unless we will get some more cooperative players in :)

  ·  2년 전

Good news: contest is extended for two days. Partly to leave a little time to get a few more bots in the arena. Partly because I am a little too busy to do all the work, so better leave you some room.

Current prize pool is around 34$ or about 3.3 STEEM and 17 SBD if I'm calculating right. This should get you motivated to submit!

Moments like this I wish I could write codes 😢 😢

  ·  2년 전

Already thought about you dear. Quoting myself:

What if I'm not a programmer?

Post your solution anyway! There are lots of programmers around (including myself) who will gladly translate your idea. Be careful to be very precise in your description though: programmers like it clear


I'll work up something

I guess I'm the first to enter... introducing SelfishPlayer who doesn't care about anyone except himself! I will probably create a bot that actually does something when I have some time, because that's way more interesting of course! Very cool contest, can't wait to see more!

  ·  2년 전

Bot is well formed, and well named

Guess that bot gets the best return on investment in terms of score per complexity. Let's leave time for your opponents to build up something more complex

If wishes were horses. Code really, is a necessity

  ·  2년 전

Programming skills help quite a bit in a blockchained universe. But coders need bakers as well as bakers need us. And bakers can be clever at social engineering (plus they make bread)

My new SocialPlayer: tries to identify other instances of itself and cooperate

  ·  2년 전

Nice bot, and clever work!

Enjoy the little time left to send even more bots!