Risks of omission and commission errors when using generative AI in behavioral science research

Generative artificial intelligence (GenAI) is rapidly being adopted across the behavioral sciences, reshaping how researchers generate ideas, design studies, collect data, and analyze results. While GenAI offers substantial benefits such as increasing productivity, enabling more sophisticated methods, and lowering technical barriers, these gains come with potential costs in the form of errors. This article discusses the risks of errors as a foundation to decide when and how to use GenAI in behavioral research. We distinguish between two types of errors of shifting relative threat, depending on how narrowly a research task is specified. Open-ended tasks (e.g., ideation, literature review) leave a large space of plausible outputs and make researchers relatively more liable to errors of omission. In constrained, well-specified tasks (e.g., running a specified analysis, writing code to a stated scheme), the narrow solution space limits the scope for omission, raising the relative threat of errors of commission. We argue that commission errors are potentially easier to detect. Omission errors, by contrast, are harder to detect and pose greater risks to the diversity and scope of scientific inquiry. We propose that GenAI usage should be guided by a comparative assessment of error detectability. We offer practical recommendations for evaluating GenAI applications and for preserving the methodological diversity that sustains cumulative science.