How will 'Chat-IRB' impact research ethics review in LMICs?

While the use of generative artificial intelligence (AI) in research has sparked controversy internationally, the use of large language models (LLMs) in the ethics review of research protocols is particularly contentious. Against this backdrop, Porsdam Mann et al recently published a comprehensive, well-balanced and carefully considered paper on the use of application-specific LLMs by Research Ethics Committees (RECs). Although we support the potential advantages that such curated LLMs can bring in improving the speed and efficiency of REC processes globally, there are some challenges that are unique to resource-constrained settings. While many of these challenges relate to infrastructural constraints, linguistic diversity, paper-based submission systems and the digital divide, substantive concerns are linked to the availability, relevance and quality of training data required for LLMs. Protection of confidential data submitted to RECs over decades is another concern. This is especially important where clinical trials are concerned. Furthermore, RECs in low- and middle-income countries (LMICs) have nuanced and historical considerations relating to research ethics that have arisen in the context of asymmetrical power differentials in international collaborative research. Currently, such considerations are not easily delegated to AI systems. There are also risks, especially automation bias, in contexts where skills development in research ethics review is much needed and progressing well. Mitigating such risks may be possible in the future by implementing various guardrails. However, exploring potentially different ways in which LLMs could be used, especially in improving capacity development of REC members, is critical.