Beyond accuracy: what makes embryo AI trustworthy in practice?

As AI becomes more visible in IVF, much of the conversation has focused on technical performance. Can a model predict pregnancy outcomes? Can it rank embryos more consistently? Can it provide a more objective reference than manual assessment alone?

These are important questions. But they are not the only questions that matter in clinical practice.

A recent qualitative study published in Journal of Assisted Reproduction and Genetics, “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,” shifts the focus from model performance to real-world acceptance.

Unlike many AI papers that center on accuracy or AUC, this study looks at the human side of AI adoption: trust, transparency, responsibility, and the role of professional judgment.

One of the most interesting findings was that ART professionals and patients did not view the issue in completely different ways. Both groups recognized the potential value of machine learning in embryo assessment, but both also expressed caution about how it should be used.

Human oversight was a recurring theme. Participants wanted embryologists and clinicians to remain involved in the final decision, especially because embryo selection is a high-stakes and emotionally sensitive part of treatment. For patients, the human element was closely tied to reassurance, communication, and trust. For professionals, it was also tied to accountability and clinical responsibility.

Transparency was another important issue. Participants wanted to understand how AI-generated results are produced, how much confidence they should place in them, and how the outputs would be used in practice. Concerns about algorithmic bias, incorrect scoring, over-reliance on AI, and potential de-skilling of embryologists were also raised.

The study does not suggest that AI has no place in embryo assessment. In fact, it shows that both patients and professionals can see the value of AI when it is used carefully. The challenge is not simply building a model that performs well, but integrating it into clinical workflows in a way that is understandable, accountable, and aligned with professional expertise.

For IVF teams, this paper is a useful reminder that the future of AI in embryology will not be decided by performance metrics alone. Trust, transparency, and human oversight will be just as important in determining whether AI becomes a meaningful part of routine embryo assessment.

In that sense, embryo AI may be most valuable when it provides an additional, objective reference point — helping teams make more consistent decisions while keeping embryologists and clinicians at the center of care.