Two sides of the same algorithm: a qualitative study of ART professionals' and patients' views on the use of machine learning for embryo assessment and selection.

PURPOSE: Machine learning (ML) is increasingly being introduced into assisted reproduction clinical practice, particularly for embryo grading and selection. This study aimed to explore how assisted reproductive technology (ART) professionals, regulators and patients understand the use of ML in embryo assessment. METHODS: Semi-structured interviews were conducted with ten ART professionals/regulators and ten patients undertaking assisted reproduction in Australian clinics. Interviews explored themes identified as ethically salient to clinical implementation. RESULTS: Both professionals and patients expressed mistrust of ML and emphasised the importance of human oversight in embryo selection. Participants identified uncertainties regarding how ML produces knowledge and raised questions about accountability and responsibility for ML-assisted decisions. Concerns were expressed that ML may lead to the discarding of viable embryos in the pursuit of selecting embryos. Participants called for full disclosure from clinics regarding the use of ML. Most professionals expressed a preference for transparent ML models and raised concerns about incorrect scoring due to poorly trained algorithms, handling errors, and potential algorithmic bias. Some professionals also expressed concern about over-reliance on ML and the potential for workforce deskilling. CONCLUSION: ART professionals and patients are cautious about the introduction of ML into embryo selection and emphasize the continued importance of human oversight, transparency, and accountability. Knowledge production with ML remains contested and there is uncertainty on how best to incorporate ML and reconfigure the work of embryologists and their professional expertise.