Toward an Evaluation Science for Generative AI Systems
This work tackles the problem of ensuring performance and safety in generative AI deployments, which is crucial for developers and regulators, but it is incremental as it builds on existing evaluation concepts.
The paper addresses the inadequacy of current evaluation methods for generative AI systems, advocating for a more rigorous evaluation science by drawing lessons from safety practices in fields like transportation and aerospace, and outlining a path forward.
There is an increasing imperative to anticipate and understand the performance and safety of generative AI systems in real-world deployment contexts. However, the current evaluation ecosystem is insufficient: Commonly used static benchmarks face validity challenges, and ad hoc case-by-case audits rarely scale. In this piece, we advocate for maturing an evaluation science for generative AI systems. While generative AI creates unique challenges for system safety engineering and measurement science, the field can draw valuable insights from the development of safety evaluation practices in other fields, including transportation, aerospace, and pharmaceutical engineering. In particular, we present three key lessons: Evaluation metrics must be applicable to real-world performance, metrics must be iteratively refined, and evaluation institutions and norms must be established. Applying these insights, we outline a concrete path toward a more rigorous approach for evaluating generative AI systems.