More Than "Means to an End": Supporting Reasoning with Transparently Designed AI Data Science Processes

Researchers analyzed AI data science systems designed for medical settings, finding that success depends on creating transparent intermediate artifacts like readable query languages and concept definitions. These intermediates help users reason about analytical choices and contribute domain expertise, despite opacity in other parts of the AI process.

• →AI data science tools can help non-experts perform complex tasks but end-to-end approaches limit users' ability to evaluate alternatives.
• →Success in medical AI systems was driven by intentionally-designed intermediate artifacts that promote transparency.
• →Readable query languages, concept definitions, and input-output examples help users reason about analytical choices.
• →Transparent intermediates allow users to refine initial questions and contribute unique domain knowledge.
• →The research suggests strategic design of intermediate artifacts is crucial for effective AI-assisted data science.