Exécution de notre outil de synthèse sur un nouvel article
Vérifiez l’article ci-dessous, puis lancez sa synthèse.
Hallucinations and Bias in Clinical AI: Retrieval-Augmented Generation as a Mitigation Strategy (Preprint)
Prajita Niraula, Mallika Upreti, Suman Kadariya, Bishal Poudel, Sujan Kadariya, Shreedhar Shreedhar Kunwar, Satenla Ruit
Résumé : UNSTRUCTURED KEY MESSAGES Standalone AI systems risk clinical harm due to inaccuracies (“hallucinations”) and biases, limiting their reliability for diagnosis or documentation. Retrieval-augmented generation (RAG) improves safety by grounding AI outputs in real-time medical evidence, but its success hinges on high-quality data and equitable design. Policymakers must prioritize adaptive regulation, including standardized bias audits, interoperability standards, and global access to prevent AI from exacerbating healthcare disparities. Clinicians, not AI, must retain final authority; RAG tools should augment judgment with explainable, verifiable recommendations while minimizing workflow disruption.