Document Type

Article

Publication Title

Duke Law Journal Online

Abstract

In Part I, we elaborate on the scope of the FDA’s experimentation, which extends beyond the fascinating (albeit limited) case studies of FDA molecular modeling and clinical trial simulation presented by Marks. In particular, we highlight novel pilot projects in which the FDA used Natural Language Processing (NLP) to analyze data collected through its Adverse Event Reporting System for postmarket surveillance of drugs. In Part II, we describe the paradigm shift in resources and efforts at the FDA from stringent ex ante premarket approval to more dynamic and rigorous postmarket surveillance. Whereas Marks places exclusive emphasis on the potential perils from this shift, we fill out the picture by pointing to the potential promise of an AI-enabled postmarket surveillance regime. In Part III, we explore the FDA’s track record in building internal AI capacity and show how the agency’s bold experimentation with the collection of structured “fit-for-purpose” data (as distinguished from unstructured text-based adverse event reports) illustrates its transformation into an “information agency” of the twenty-first century. Given that for most federal agencies the question is not whether the agency will eventually embrace AI technologies, but how and in which domains, the FDA’s experience provides a window into the future promise of AI in the administrative state.

First Page

86

Volume

72

Publication Date

2022

Download

Share

COinS