AI isn’t ready to take over medical literature reviews
A Nature World View piece argues that AI tools, while being explored to speed up health-related literature reviews, are not yet reliable enough to replace human experts. AI can replicate steps like study identification, data extraction, and writing, but meaningful review questions, relevance assessment, and interpretation require human judgment. AI models can hallucinate and lack the necessary context, and many tools are private/black-box, raising independence and bias concerns. Training times are long and current workflows often take longer than manual reviews. The author advocates building cooperative human–AI systems rather than generating entire reviews, to safeguard accuracy and policy relevance.