Evaluation of a Multi-agent “Human-in-the-loop” Game Design System

Abstract

Designing games is a complicated and time-consuming process, where developing new levels for existing games can take weeks. Procedural content generation offers the potential to shorten this timeframe, however, automated design tools are not adopted widely in the game industry. This article presents an expert evaluation of a human-in-the-loop generative design approach for commercial game maps that incorporates multiple computational agents. The evaluation aims to gauge the extent to which such an approach could support and be accepted by human game designers and to determine whether the computational agents improve the overall design. To evaluate the approach, 11 game designers utilized the approach to design game levels with the computational agents both active and inactive. Eye-tracking, observational, and think-aloud data was collected to determine whether designers favored levels suggested by the computational agents. This data was triangulated with qualitative data from semi-structured interviews that were used to gather overall opinions of the approach. The eye-tracking data indicates that the participating game level designers showed a clear preference for levels suggested by the computational agents, however, expert designers in particular appeared to reject the idea that the computational agents are helpful. The perception of computational tools not being useful needs to be addressed if procedural content generation approaches are to fulfill their potential for the game industry.