Kristine Zheng¹ (), Jean-Peïc Chou¹, Junyi Chu¹, Judith Fan¹; ¹Stanford University

Many people choose to spend time watching other people play video games. What makes some gameplay more enjoyable to watch than other gameplay, even in the context of the same game? Here we explored features of gameplay that might influence observers’ self-reported enjoyment, especially factors related to the apparent skill level of the game player. We developed a variant of the Flappy Bird game, in which a player traverses an environment from left to right while trying to minimize the number of collisions with vertical pipes of variable length extending from the top and bottom of the screen. We trained four artificial agents that exhibited different “playstyles,” varying in how close these player agents passed by these pipes and the velocity with which they approached them. One agent consistently traveled just beneath the upper pipes (“top-skimming”); one traveled just above the lower pipes (“bottom-skimming”); one avoided both sets of pipes with a large margin (“centered”); and one agent followed an undulating, swooping trajectory that made it approach obstacles at high velocity (“swing”). Participants (n=80) watched one video clip of each agent navigating a single course, then rated how enjoyable it was to watch. The number of collisions sustained by agents was variable (i.e., between 0-3 per course), but collision frequency was held constant across clips viewed by each participant to isolate the effect of playstyle. We found that enjoyment ratings assigned to these agents were reliably different (χ²(3)=16.55, p<.001; d≈0.05–0.29), with "swing" being most enjoyable, then "centered", “top-skimming”, and “bottom-skimming.” These preliminary findings suggest that what observers find enjoyable in gameplay goes beyond the outcome in each clip (e.g., number of collisions), but also encompasses the way in which a player achieved that outcome.

Acknowledgements: This work was supported by NSF CAREER Award #2436199, NSF DRL #2400471, and awards from the Stanford Human-Centered AI Institute (HAI), Stanford Accelerator for Learning, and the Brown Institute for Media Innovation.