Ethical Imperatives in the Integration of Artificial Intelligence in Assisted Reproductive Technology

摘要

Dear Editor, Assisted reproductive technology (ART) has revolutionised the world of reproductive medicine and offered hope to millions of people and couples through solutions for infertility. In the recent years, the most promising area that could advance ART is robotics and artificial intelligence (AI) integration. From enhancing embryo selection through machine learning algorithms to optimising ovarian stimulation protocols, AI will revolutionise the way clinical decision-making and patient care are rendered.[1] This sensitive, highly ethically complex field requires caution and full consideration of the implications required, such as data privacy, informed consent, bias in the decision-making process and the need for human oversight. Transparency and explainability in AI systems used in ART are critical for ethical practice. Many AI algorithms function as “black boxes,” generating recommendations without clearly explaining the reasoning behind their conclusions. This lack of clarity can undermine informed consent, making it difficult for both clinicians and patients to critically assess AI-driven decisions.[2] The integration of AI into ART calls for informed consent from patients about the use of their anonymised data. Patients should be made aware that their data could be shared with companies developing AI solutions to refine algorithms and improve predictive accuracy.[3] Providing interpretable outputs is essential to maintain trust, accountability and confidence in the technology. Equally important is the issue of algorithmic bias, which is a significant ethical concern in ART. AI models are trained on biased datasets that inadvertently may amplify inequalities in access and effectiveness of fertility treatment. These include genetic and demographic imbalances that may lead to incorrect embryo viability predictions across different populations. Such biases can only be reduced using diverse, representative datasets combined with rigorous validation protocols to ensure that all patients will have equal results.[1] Studies have shown that the development and testing of efficient and accurate AI-driven technology depends on access to large data sets, reliability and quality of the data shared and bias in data collection. With increasing regulation, access to large datasets can be challenging. Data obtained from multiple centers can be heterogeneous and diverse. These issues can impact the design of precise and accurate AI algorithms.[4] On the other hand, data leakage is a scenario in which information is unintentionally included in the training data that would not be present in real-world clinical scenarios, thereby causing the performance metrics to be misleadingly high. This has brought multifaceted privacy and security challenges into ART that require ethical oversight beyond traditional medical data protection. The use of biometric information, such as reproductive tissue imaging and hormone level tracking, requires secure protocols for patient identification and data management to prevent unauthorised access and misuse. This would mean that the privacy standards of AI algorithms in use in ART will be very high, include anonymising training datasets and safeguard model inference privacy while ensuring algorithmic transparency in protecting the autonomy and trust of patients.[5] The major moral dilemma that arises with AI-driven ART solutions involves figuring out who is liable in the case of errors or poor medical decision-making.[6] The question rises whether to blame the clinician, the hospital, or the AI provider. This is another fact that needs to be taken into regards. This calls for clear guidelines on legal and ethical responsibilities to improve patient safety. Although AI will optimize efficiency in ART, losing clinical touch is still a subject of concern. AI can be a good assistant for decision-making but cannot replace human clinicians in their empathetic and personalised care.[7] In conclusion, AI will hav