From news to practice.
A convincing response does not prove the clinical effectiveness of a tool. In healthcare, evidence, evaluations, and defined responsibilities are necessary.
Possible uses
In the guidelines published in January 2024, the World Health Organization examined large multimodal models: systems capable of processing different types of data, such as texts and images. Considered applications include clinical assistance, administrative tasks, training, and drug research.
Errors that cannot be ignored
WHO calls attention to the risk of false, incomplete or distorted answers, along with data quality and cybersecurity issues. It also highlights the risk of relying excessively on automation: a convincing response can be accepted without the checks that a doubtful piece of information from other sources would have received.
Questions for hospitals
What task is entrusted to the system? On which populations has it been evaluated? Who detects errors and can stop its use? These are useful questions to report hospital projects rigorously. To describe an experiment, documents, evaluation criteria, and confrontation with those conducting it are needed: the distinction between research and clinical use must remain explicit.
Sources and documents
Document date or reference period: 18 January 2024 · WHO guidelines.
Text prepared with AI assistance based on the indicated sources. Interpretive evaluations are distinct from data and statements from sources.
