Procedural content generation algorithms require design
knowledge that captures the intuition of human designers about
good content. Human authoring of design knowledge remains a
large roadblock in procedural content generation and in general
video game design for non-experts. We have presented a novel
approach to automated learning of computer game design
knowledge from gameplay video, with a specific focus on level
design knowledge. An initial evaluation of our approach indicates
an ability to produce level sections that are both playable and
close to the original Super Mario Bros. without hand coding any
design criteria. Initial experiments suggest that our approach
extends beyond the Super Mario Bros. platformer game and that
additional design knowledge such as avatar mechanics may be
acquired from gameplay video as well. As gameplay video
becomes more accessible and as open machine vision toolkits
become more advanced, we see gameplay video as a rich source
of design knowledge to be exploited for future procedural content
generation and procedural game generation systems.