Key Takeaways on the Use of GenAI in Statistical Practice from a JSM Roundtable

Abstract

Generative artificial intelligence (GenAI) is increasingly embedded in the everyday work of statisticians and other quantitative scientists, yet practical experience often advances faster than formal guidance on where these tools add value and where they create new risks. This manuscript synthesizes a moderated discussion among data scientists from pharmaceutical and related health-sciences settings that took place at the 2026 Joint Statistical Meetings (JSM) in Houston, TX. The discussion focused on concrete experiences with GenAI-based tools in coding, scientific writing, literature synthesis, statistical interpretation, workflow automation, and the development of everyday work products such as presentations and spreadsheets. Participants agreed that GenAI appeared most useful for bounded tasks whose outputs could be quickly inspected and checked against authoritative sources. When the outputs required substantial subject-matter expertise to verify, for example, in the design of a complex clinical trial or the generation of a Clinical Study Report (CSR) draft, users were less confident in the value proposition. Source-grounded systems and agents offered promising approaches to organizational knowledge retrieval, but model drift and limited reproducibility raised concerns for regulated workflows. Participants also highlighted implications for training and early-career development. Overall, the discussion suggested that the effective and responsible adoption of GenAI tools should focus less on maximizing automation and more on matching the level of AI autonomy and oversight to the consequences and verifiability of the task.