Towards AI Drawing Agents

Abstract

This paper speculates on the applicability of methods from artificial intelligence in generative design. In particular, Cppn-Neat as a recent concept is implemented and tested on an established platform for procedural design. The method evolves artificial neural networks and has proven successful in control of robotic navigation, cognition, and collaboration, therefore is proposed to control the behaviour of drawing agents for architectural optimization tasks. A projected goal is the development of networks which, once evolved, instantly perform specific tasks on generic geometries, to be re-used on different projects and save run-time in the optimization process. Potentials of the method regarding general encoding and novel modes of formal articulation are examined in a first set of parameterization-studies. Further examples are testing the scalability of results from different problem domains and work towards the evolution of large ANNs to be suitable for the synthesis of higher level concepts. KEYWORDS