Artistic User Expressions in AI-powered Creativity Support Tools
John Joon Young Chung
Abstract
Novel AI algorithms introduce a new generation of AI-powered Creativity Support Tools (AI-CSTs). These tools can inspire and surprise users with algorithmic outputs that the users could not expect. However, users can struggle to align their intentions with unexpected algorithmic behaviors. My dissertation research studies how user expressions in art-making AI-CSTs need to be designed. With an interview study with 14 artists and a literature survey on 111 existing CSTs, I first isolate three requirements: 1) allow users to express under-constrained intentions, 2) enable the tool and the user to co-learn the user expressions and the algorithmic behaviors, and 3) allow easy and expressive iteration. Based on these requirements, I introduce two tools, 1) Artinter, which learns how the users express their visual art concepts within their communication process for art commissions, and 2) TaleBrush, which facilitates the under-constrained and iterative expression of user intents through sketching-based story generation. My research provides guidelines for designing user expression interactions for AI-CSTs while demonstrating how they can suggest new designs of AI-CSTs.