Artificial intelligence (AI) tools can be useful to caregivers but they must be used with great care, in particular by systematically checking their sources, the High Authority for Health (HAS) said on October 30.
“ Generative AI systems can be a lever for improvement to promote quality in the healthcare system.", the HAS judges in a series of recommendations. But " Therefore, they must be used in a reasoned manner, for the benefit of individuals and to support professionals.", warns this independent authority, whose opinions serve as a framework for health policies in France.
These tools, which notably include OpenAI's ChatGPT assistant with its 800 million weekly users, represent a major technological innovation and are now widely used daily for a multitude of purposes. The HAS (French National Authority for Health) has therefore examined their use in the healthcare sector.
She believes they can be used for a wide variety of purposes: to establish a synthesis of the scientific literature on a given subject, to help institutions manage their resources, to translate certain information into clear terms for patients… But she also calls for great vigilance, recalling for example the ability of AI tools to “hallucinate”, that is to say to present as fact statements that do not correspond to any reality.
Read alsoWhen AI lowers doctors' expertise
A first working basis
The HAS therefore recommends, in particular, that healthcare professionals always check the sources used by AI and to " consult as needed to familiarize yourself with their content and verify them by referring to other reliable sources"She also highlights the issue of confidentiality with regard to patients: it is necessary to..." verify in each request that no information allowing direct or indirect identification or relating to medical confidentiality is shared“.
Above all, she recommends not relying entirely on AI, so as not to stifle one's own skills: Thus, a triage nurse in an emergency department using a generative AI system to assist in patient triage, while still retaining some triage capabilities without this assistance, will be able to perform this triage even if the system malfunctions.", she cites as an example.
This document is an initial working document, but the HAS plans to develop further guidelines. Future recommendations will be addressed directly to patients, as AI tools are used by many users to answer questions about their health, potentially leading to unreliable opinions.
English 
French