A Proposition for Master's of Arts in New Media Thesis:

Design of a Multi-player Video Gambling Game with Artificially Intelligent, Automated Test-Players

Author: Petri Kuittinen

Supervisors: Timo Honkela (Ph.D), professor at University of Arts and Design, Media Lab
Pekka Puomio (M. Sc.), Raha-automaattiyhdistys, Research and Design department

Abstract:
The proposed Master's Thesis includes design and prototype implementation of a multi-player poker game, which will have several different artificially intelligent, automated test-players. A network protocol which supports automated testing of single- and multi-player games will be developed.

Background

In Spring 1999 our Game Design Study Project group (Petri Kuittinen, Janne Frösen, Jan-Mikael Lepistö, Tomi Knuutila, Katerina Antonaki and Miska Natunen) designed and implemented a prototype version of a multi-poker video game using Java programming language. Due lack of time the resulting prototype was unfortunately far from finished. The game did not have a single-player mode, the user-interface was so vague that it made the game very difficult to play and the network code and protocol also had errors making the game very unstable. I estimate that it would take at least the same amount of time, about 500 hours, as Janne and Jan-Mikael used for the GD study project, to fix these problems in Java source code.

I converted the internal logic of the game (handling of cards, comparison of wins, randomness) from Java to Python, which is a free interpreted, interactive, object-oriented programming language. Even thought the Python version is much more generic (e.g. it implements three different kind of poker games instead of one) the resulting source code was shorter (665 lines vs. 429 lines) and the time used was less than 1/5 of the original.

Objectives, Methods and Approaches

At the end of 1998 Raha-automaattiyhdistys (RAY) started a project called "Pelisukupolvi PSI-2002" (= "Game Platform PSI-2002"). The goal of that project is to produce a completely new generation of gambling machines which have better reliability, easier maintenance and enable easier and faster software development. That includes better play-testing of the games. So far the games have been tested by human test-players. Multi-player games with large number of simultaneous players would require a small army of human test-players which isn't feasible. So there have to be automated test-players which simulate the way real human players play the game. Automated test-players can play hundreds of thousands of games without tiring, leading to better-tested and more robust single-player games as well. It would also be useful to be able to record and playback games played by human players. The statistics and observations gathered from human test-playing can be used to create automated test-players which imitate human behavior.

During summer 1999 I came to the conclusion that the network protocol needs to be redesigned by first thinking about different kind of multi-player gambling games and their needs and then design an extendable and robust network protocol which has:

1. Part which is common to all multi-player games:
   • changing game mode between single/multi-player game, query game status
   • starting a multi-player game round, game synchronization
   • handling wins and money
2. Part which has all the necessary information concerning game logic (this part of specific to each game type)
3. Part which has the necessary user-input messages to implement automated testing of single-player game (different kind of user input events e.g. such as "pay wins", "double", select buttons...)

I have don't have any previous experience in designing a network protocol, automated test-players, implementing network communication for a multi-player game nor a graphical user-interface (GUI) using Python so there is certainly a lot to learn.

I would like implement the AI of test-players using conventional hand-coded (static) approach. Neural networks and other more sophisticated results could perhaps lead to better results, but my resources are limited.

Resources

I will do almost all the work by myself. I will consult Pekka Puomio in questions relating to network protocol design and calculating the propabilities. Software engineer Tuomo Hakola (of RAY R&D department) is the principal person responsible for developing better testing procedure for PSI-2002, so we will certainly exchange ideas and I will present my ideas to him before actually implementing them.

I will do my work using the computer facilities of RAY. All (or almost all) the work will be done using free software (Linux, Gnu Emacs etc.).

People at RAY R&D department will help me play-testing the game. Probably some people at Medialab would also be interested in play-testing?

I don't have much experience or knowledge in user-interface design and my time is limited, so I won't pay much attention to the user-interface of the game within my Master's Thesis. It would be nice to find someone interested user-interface design and skilled with GUI programming to collaborate with this project. I am sure there is enough work for another final thesis just to develop a really good GUI for a multi-player poker.

Research Process & Results

Content of the Master's Thesis:

• Design document of the game(s)
• Design, implementation and documentation of the network protocol described above. The network protocol will have all the needed elements for multi-player poker game and it is designed so that it can be easily expanded for other games as well. It will be implemented on top of TCP/IP. At this stage the protocol will not include any kind of crypting or protection against network attacks.
• Implementation of a prototype of multi-player video poker game similar to the one sketched out in Game Design Study Project. The game will have:
  • a single-player mode
  • a multi-player mode with maximum of 4 simultaneous players
  • ability to record user-input
  • two different kind of user-interfaces: a text-based (used mostly for debugging) and a simple graphical one
  The main emphasis will be on the play-ability of the game, not on the fancy look of the user-interface. The game will have no sound.
• Statistical data and experiences from automated play-testing and recorded human test-playing.
• At least two different kind of automated test players. At least one of them includes random behavior and game-play patterns which try to make game behave in a wrong way (to find out bugs). At least one of the automated test-players is "smart". It tries to make good decisions to maximize wins. This artificial intelligence will probably be as good as an average human player.
• The resulting programs will be implemented using Python. The game will run on Linux, possibly also on Windows and Macintosh (if it doesn't require extra work).
• The source code is documented using standard Python documentation strings. The most important program concepts will have additional documentation, including UML-alike graphical representation.
• Experiences from using Python instead of C-alike languages in development of new gambling machines.
• The Master's Thesis will be written in English and resulting documentation will be in HTML 4.0.

If there is too much time (which is unlike) the following things might be included to Master's Thesis. Anyway they are good starting points for further work.

• Several different kind of poker games (Joker Poker, Ace Poker, Deuces Wild)
• Doubling in single-player mode
• Other multi-player games: a number game similar to bingo, keno or lotto, maybe also a jackpot game.
• Sound and music.
• consideration/ideas for Internet games (with large number of players)
• automated test players have personality (e.g. like to take risks), emotions (doesn't like losing, loves to win).

Open Issues:

• publicity? (most likely at least the source code must remain private, is this possible?)