Design Games and Machine Learning: A Method for the Application of Machine Learning in Design Environments.

Machine learning techniques have been applied to an increasing variety of problems, of increasing complexity that were once considered infeasible for artificial intelligence solutions of any kind (Sliver et al. 2016). We propose a solution to this by creating a shared, abstract, environment that is computable but also provides enough user-freedom to exercise a design problem through exploring solutions. We use a game-like environment to model not only complex simulations of systems, but also allow both human and AI actors to “play” in a shared space to exploring design alternatives. To address the openness of the design problem, we close the problem for the AI by providing user-controlled weights on game performance metrics. Thus allowing the AI system to optimize in favorable ways with respect to the bespoke demands and goals of the designer.