How to Construct a Believable Opponent using Cognitive Modeling in the Game of Set

Niels A. Taatgen (niels@ai.rug.nl)
Marcia van Oploo (marcia@ai.rug.nl)
Jos Braaksma (j.braaksma@ai.rug.nl)
Jelle Niemantsverdriet (jelle@niemantsverdriet.nl)

Department of Artificial Intelligence, Grote Kruisstraat 2/1
9712 TS Groningen, Netherlands

Abstract

An interesting domain of application for Cognitive Modeling is the construction of computer opponents in games. We present a model of the game of Set. The model is sensitive to the difficulty of the situation in the game, and can explain the difference between beginners and experts. Furthermore, the model is used in a computer game in which players can play Set against various versions of the model.

Introduction

In computer games, the human player often faces one or more computer opponents. In order to make the game enjoyable, these computer opponents have to be intelligent, otherwise they wouldn’t be much of a challenge. The classical example of playing a game against the computer is chess, and the focus of designing a computer chess player has always been to have a player that plays as good as possible. In fact, since Deep Blue beat Kasparov the interest in computer chess seems to have diminished, but maybe the recent rematch against Kramnik will change matters. Nevertheless, human chess players complain that computer chess programs are no fun to play against. A possible reason for this is that computer chess programs have long left the approach of mimicking human chess players, but instead focused on brute-force techniques.

This leads us to the idea that a computer opponent becomes more interesting and enjoyable to play against as it behaves more like a person. In this paper we will explore this idea and demonstrate how cognitive modeling can help to produce more interesting computer opponents. The basic idea is relatively simple: study how people (preferably at different levels of skill) behave in the game you want a computer opponent of, make a cognitive model of this behavior that closely matches human behavior, and incorporate this in a computer program.

The game we will use is the game of Set. Set is a card game that is quite trivial to play perfect for a computer program, so the challenge is to model an opponent that is interesting to play against. Another interesting aspect of Set is that it combines several aspects of cognition: perception, information processing, strategy, learning and time pressure. We will describe an experiment and several possible models of playing Set, and incorporate those in a program that can one can play against.

Click on the attachment below to read the full paper.