Practitioner Perspectives on Data Quality Challenges in AI-Driven Systems

Interviews with 16 practitioners reveal that AI-driven data quality requires shifting from modular debugging to attributing model behavior. Key emerging concerns include the circularity of using models as quality assessors, the authenticity of synthetic data, and the legal gatekeeping of foundation model training sets.