CoAutoML: User Interface Framework for Machine Learning Novices using LLM-based AutoML and Test-Driven Machine Teaching

An explosion in automated machine learning (AutoML) tools has led to numerous back-end frameworks enabling users with machine learning expertise to leverage the power of Machine Learning (ML). However, to ensure that ML tools are openly accessible to all who stand to benefit from their predictive and analytical powers, we must examine how true novices without ML knowledge interact with AutoML tools, perceive ML, and form their mental models of ML processes. We achieve this goal with our user-facing framework that combines the understandability of conversation with Large Language Models (LLMs) and the interface scaffolding necessary to support true machine learning novices in building their own models. We then evaluate the effectiveness of our framework in a user study. Results show that our ML-novice participants felt confident performing ML tasks independently, citing the tool's ease of use and its ability to help them formalize their ML goals.