Why AI can't be trusted to write scientific reviews (www.nature.com)

The Cochrane Collaboration is exploring the use of AI tools to enhance the efficiency of systematic reviews in health-related research. Despite the potential for AI to streamline literature reviews, the organization warns that current models are not ready for widespread use due to critical limitations. Systematic reviews play a pivotal role in clinical practice and public health, and errors in this process can lead to significant repercussions, including misguided treatment protocols and wasted healthcare resources. AI's ability to replicate the procedural steps of systematic reviews—like study selection and data extraction—falls short because these tasks require nuanced human judgment and contextual understanding. Most existing AI tools, mainly developed by private companies with proprietary algorithms, present transparency issues, specifically concerning their impartiality in evaluating evidence for drugs and medical devices. Cochrane's findings suggest that, rather than replacing human experts, it would be more beneficial to develop AI systems that support collaborative efforts between humans and machines, allowing for a more comprehensive and reliable review process.