rock-paper-stuff

Rock Paper Stuff

By Lee Spector

This is work in progress. Expect changes.

About

Rock Paper Stuff is a game designed for artificial intelligence experiments and education.

The game itself is intended to be played by computer players, each of which aims to achieve the best balance among five resources: Rock, Paper, Scissors, Fire, and Water.

The goal for students and researchers is to develop computer players that win games and tournaments.

The Game

Any number of players can play.

Players are paired randomly for trades, one pair at a time. Sometimes the trades cause some of the traded things to be lost or other things to be created (details below).

For each trade, each player plays one of: Rock, Paper, Scissors, Fire, or Water.

Players start with five times as many of each of these as there are players. For example, if there are four players, then each player begins with twenty Rock, twenty Paper, twenty Scissors, twenty Fire, and twenty Water.

Players can only play what they have. If they try to play something that they don't have, then one of the kinds of things that they do have will be chosen randomly, and a thing of that kind will be played. If they run out of everything, then they die, participate in no more trades, and can't win. (The dead are perfectly balanced, but losers nonetheless!)

A player's balance is measured as the standard deviation of its amounts of stuff, with a deviation of zero meaning that the player is perfectly balanced, and higher deviations being worse. Intuitively, the more spread out the amounts are, the higher the deviation will be, and the worse the player is doing. Technically, the deviation is calculated by taking all of a player's amounts of stuff, averaging them, looking at how far each is from the average, squaring each of those distances, averaging those squares, and then taking the square root of that average. That's a little complicated, but it's a standard way of measuring how spread out a set of numbers is, and the game does it for you.

The number of trades in a game is the number of possible pairings times 100. For a two-player game, this is 100 trades. For a three-player game, 300 trades. For a four-player game, 600 trades. In general, the number of trades is 100n(n-1)/2 where n is the number of players. Because pairings are indepedently random, different players may participate in somewhat different numbers of trades, and different pairings may occur somewhat different numbers of times.

Although players can't see what each other are going to play before they choose their own play (it's as if they must choose simultaneously), they can see each other's names, inventories, and anything else that a player has chosen to display on its "skin" (which could include promises to behave in a certain way, or information about other players, etc.). Players can also remember whatever they want to record from trade to trade, and they can access a complete history of the trades in which they've participated.

Trades

• If A plays Rock and B plays Paper, then A gets Paper, and B gets Rock (simple exchange).

• If A plays Rock and B plays Scissors, then A gets Fire, and B gets Rock (Rock sparks fire from Scissors)

• If A plays Rock and B plays Fire, then A gets Rock, and B gets Scissors (Fire forges Rock into Scissors)

• If A plays Rock and B plays Water, then A gets nothing, and B gets two Water (Rock sinks and splashes back Water)

• If A plays Paper and B plays Scissors, then A gets two Paper, and B gets Scissors (Scissors cuts paper into two)

• If A plays Paper and B plays Fire, then A gets nothing, and B gets two Fire (Fire lights Paper)

• If A plays Paper and B plays Water, then A gets Paper, and B gets Nothing (Paper absorbs Water)

• If A plays Scissors and B plays Water, then A gets Water and B gets Rock (Water rusts Scissors into Rock)

• If A plays Fire and B plays Water, then A gets nothing and B gets Water (Water extinguishes Fire)

When both players play the same kind of stuff, it sticks together and one of the players (chosen randomly) will get both, while the other gets nothing.

These rules are summarized in the following diagram, in which the circled abbreviations show the overall loss/gain to the ecosystem resulting from the specific trade. The rule about plays of the same kind of stuff sticking together isn't shown in the diagram.

Playing

You can see what it looks like to run the game in this saved Gorilla REPL worksheet.

The code for Rock Paper Stuff is written in Clojure, and intended to be run in a Gorilla REPL worksheet or from the command line, in either case using the Leiningen tool for Clojure project management. The instructions here assume that you have some familiarity with Clojure programming and with your computer's command line interface (terminal).

To play Rock Paper Stuff, download the project folder and then use the terminal to navigate into the project directory.

You can then type lein gorilla, and point your browser at the URL that it prints (after a few seconds).

This will open a fresh worksheet, which you could work in and save... But the first time you do this, you should instead use the menu (upper right), select "Load a worksheet," and load src/rock-paper-stuff/worksheet.clj.

You'll see the evaluated worksheet, and you can click into any code cell and press control-enter to evaluate it again (possibly after changing it). This worksheet will walk you through basic interactions with the Rock Paper Stuff environment.

To run games and tournaments from the terminal, put your code in user.clj in the indicated places. Then typing lein run at the operating system command line will run that function and print the result to the terminal.

Acknowledgments

Based on ideas from conversations with members of the Hampshire College Computational Intelligence Lab and:

• Clifford Bohm
• Ira Fay
• Jim Kan
• Joshua Newman
• Charles Ofria
• Anya Spector

This material is based upon work supported by the National Science Foundation under Grants No. 1617087, 1129139 and 1331283. Any opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of the National Science Foundation.