Applying generative AI in analytics: failure modes and opportunities
Published in YouTube / INFORMS, 2026
This webinar details how Generative AI can be integrated into data science workflows while navigating its technical “fault lines.” Framing Gen AI as an “overeager, naïve, new graduate hire,” Dr. Louis Luangkesorn demonstrates its application using the INFORMS Analytics Framework on the classic 1982 RAND Health Insurance Experiment, showing how AI assists with problem framing, model development, and assumption validation.
He highlights examples of AI naivete, emphasizing that Generative AI serves as an analyst’s aid rather than a replacement. The session explores mitigation strategies for common AI mistakes, such as providing domain-specific context, enforcing software engineering best practices (e.g., separation of concerns), and creating inspection artifacts for verification.