Member-only story
Automated Content Creation and A/B testing with Reinforcement Learning
Beauty may be in the eye of the beholder, but it’s creation will soon be in the hands of an algorithm
No attribution required
Accompanying Video: https://youtu.be/dmxrN7uUUhs
This article introduces the burgeoning field of automated content creation with a player generated reinforcement learning signal. In the Unreal Engine example presented, a user demonstrates their preferences concerning the components of a game while interacting with it, and the software in turn automatically adjusts or creates components based on those revealed preferences. At its extreme, this technique could be used to completely re-pattern the script or goals of a piece of software based upon subconscious user feedback.
While not possessing a crystal ball, it seems likely that such methods represent the next frontier in game development and perhaps art, entertainment and user interfaces more generally. In such a scenario, none of the content is fixed but rather continually adjusts to the user as they interact with it. While this may sound like science fiction, we are already familiar with the concept of this through things like Netflix and Pandora recommender algorithms. The tools and methods presented here simply extend this, allowing an AI to continuously…
