Creative Together: Human-AI Ensembling for Idea Generation

Abstract

Recent developments in deep learning algorithms have enabled generative artificial intelligence (Gen AI) to produce creative content across various modalities that is often indistinguishable from human output, challenging the long-held belief that creativity is a human prerogative. While such advancements have made human-AI collaboration based on specialization (humans and AI take on different tasks) feasible for creativity, they also pose a risk of diminishing essential human creative skills. To better preserve and develop human creativity, we investigate human-AI ensembling for idea generation, which entails an aggregation of outputs from multiple agents who each tackle the same task. We provide a theoretical framework that reconceptualizes idea generation as a creative search, thereby enabling the exploration of how humans and AI collaborate in idea generation and how ideas generated by humans and AI can be aggregated. In addressing the core trade-off between preserving human skills while producing satisfactory joint outputs, our framework facilitates further theorizing and empirical analysis of collaborative idea generation between humans and Gen AI.

Keywords: Creativity, Collaboration, Search, Human-Centric, Ensembling, Generative AI (Gen AI)