Pieter Spronck Pieter Spronck

Will computergames soon become really intelligent?

Written by  Femke Kools Tuesday, 08 June 2004 00:00

Intelligent opponents make computer games exciting again

Once you know how to beat the opponents in a computer game, you can repeat that trick over and over again to win. After all, characters will not suddenly decide to adjust their attack. They are not programmed that way. At some point, there is no more fun in it for the player. Pieter Spronck of the Faculty of General Sciences therefore investigated the possibility to add intelligent opponents to existing computer games. "I developed a technique, Dynamic Scripting, which allows the computer to anticipate the tactics of the player." Spronck hopes to receive his doctorate in the beginning of 2005 for this research into machine-learning techniques in computer games.

In the Netherlands, he is one of the few people who study the use of this type of artificial intelligence in computer games. Initially, the choice for this research subject came from his personal interest: Spronck is a passionate gamer himself. "Computer games are a hobby of mine, so it was an obvious choice to turn it into a research subject. Moreover, it is a big challenge to apply artificial intelligence to existing games. A lot of difficult practical requirements have to be fulfilled because the technique that I wanted to develop has to fit in with the present structure of computer games. After all, if my method cannot easily be used for existing games, it is not interesting for commercial game developers."

So-called scripts play an important role in directing the virtual characters in a game. They prescribe what a character has to do. "I have tried to connect to these scripts. To that end, I developed a rule base, a small database with rules that determine the actions of computer opponents. A script is then formed by combining arbitrary rules from this rule base. The rules in the rule base have been designed in such a way that they all represent meaningful actions and the script they compose is also meaningful. The idea is that the most effective rules have the biggest opportunity to be chosen so that the opponents' behaviour becomes smarter and smarter. The weight factor of the rules is adapted after evaluating each time the results of the fight with the human player. If the results are good, the weight of the applied rules will be increased. If they turn out bad, they will be decreased again."

The technique was first tested in a computer game simulation. After it was shown that the technique worked well, it was built into an existing game: Neverwinter Nights, a role playing game in which the player himself plays a role with a chosen or created character. It turned out that the learning opponents that started without any knowledge needed about twenty to thirty fights to become systematically better than the standard opponents that were programmed by the developers of the game. They adjust their strategy to that. A human player who finds out that a certain strategy works very well, can no longer rely on these tactics against a learning opponent, because the learning opponent adapts to them. That way, the actions of the computer remain a permanent challenge.

Spronck hopes that his research will ultimately lead to the future development and marketing of computer games with intelligent opponents that adapt during the game to the tactics of the player. For the time being, the producers of games are not very keen on implementing his technique in existing games. "Sometimes it is not very clear to what the use of machine learning techniques can lead in the game. It can go wrong if the human player does not play fair or makes stupid decisions. The computer opponent can learn the wrong things and even turn against himself. That is not good for the game." In spite of the scepticism in the producers of computer games, Spronck expects within a few years lots of games with intelligent opponents will be on the market. After all, the form of these computer games has already developed so much in the last few years that the intellectual level of the characters cannot stay behind.

Ir. Pieter Spronck studied technical computer science at the Technische Universiteit Delft, where he graduated with honours in 1996. Since 2001, he has been attached to the IKAT (Institute for Knowledge and Agent Technology) of the Faculty of General Sciences of Universiteit Maastricht. Since 2002, he has been studying computer games and the changes that take place in these games.
Website IKAT: http://www.cs.unimaas.nl
e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Rate this item
(0 votes)
Read 31934 times
You are here: Home Research Technology Will computergames soon become really intelligent?

Maastricht University Webmagazine

Marketing & Communications
Postbus 616, 6200 MD Maastricht
Minderbroedersberg 4-6, 6211 KL  Maastricht
Tel: +31 43 388 5222
This email address is being protected from spambots. You need JavaScript enabled to view it.

Download UM Magazine

Read a digital version of Maastricht University magazine,
or download the PDF.
UM Webmagazine June 2017

Connect with us