EMA: Guiding Principles on LLMs, a Category of Generative AI

On 5 September 2024 the EMA and the Heads of Medicine Agencies HMA, published guiding principles that set how EU medicine regulators can use large language models (LLMs), a form of generative AI, focusing on text generation in regulatory science and for regulatory activities. The 10-page document provides high-level recommendations to facilitate LLMs' safe, responsible, and effective use, being of interest to anyone who uses or would like to use such AI models in the future.

"LLMs have enormous transformative potential," says EMA, “but also present challenges: variability in results, returning of irrelevant or inaccurate responses (so-called hallucinations), and potential data security risks." On the other hand, LLMs could be applied to query the extensive documentation regulators receive, to automate knowledge/data mining processes, or as virtual AI assistants in everyday administrative tasks.

Therefore, the document aims to build an understanding of the capabilities and limitations of these applications among staff at regulatory agencies across the EU to harness the potential of LLMs effectively and avoid pitfalls and risks. General ethical considerations, e.g. the rule of law, accountability, security, safety, transparency, non-discrimination, misinformation harms, and malicious uses or copyright issues are included.

Being part of the multiannual AI work plan to 2028 set out by EMA, the guiding principles are considered a living document that will be regularly updated. An EMA factsheet on LLMs can be found here.

Source:

EMA: Guiding principles on the use of LLMs in regulatory science and for medicines regulatory activities

EMA: Factsheet on the 4 principles for safe and responsible use of LLMs