Human-level design proposals by an artificial agent in multiple scenarios

Abstract

Many computational tools are usually trained using human-curated data set of design proposals. However, this approach can limit system capabilities to generate creative designs since human knowledge of the solution space is already embedded in the training process. In this paper, we show how by using a flexible design tool that can also be used by humans, an artificial agent can learn to generate creative designs with any prior knowledge of the solution space. Our results show how our agent is able to create human-level design proposals in terms of performance and novelty. Based on these results, we discuss the importance of defining a shared design language and tools in order to support human-AI collaboration in creative scenarios.