By Sinead Prince and Julian Savulescu
Centre for Biomedical Ethics, Yong Loo Lin School of Medicine, National University of Singapore
Prince, S., & Savulescu, J. (2025). When is black-box AI justifiable to use in healthcare?. Big Data & Society, 12(4), https://doi.org/10.1177/20539517251386037. (Original work published 2025)
This research explores a pressing dilemma in healthcare today: when is it justifiable to use black-box AI, which lacks transparent explanations for its decisions, in medical practice? The topic drew our interest due to the growing tension between the potential benefits these AI systems offer—such as more accurate diagnostics—and the strict regulations in regions like Australia and Europe that prohibit their use without adequate transparency.
In AI ethics literature, many argue that AI in healthcare must be explainable. However, there are crucial situations in which accuracy and reliability become more important than transparency. For example, in urgent or serious cases, faster and more precise AI-driven decisions can save lives even if the reasoning is inscrutable. Explainability cannot be the only ethical requirement—context, accuracy, bias, seriousness, urgency, and human involvement also play essential roles in determining when to use black-box AI in medical decision making.
Our main argument is that automatic bans on black-box AI overlook its practical value. Rather than demanding explainability as a universal rule, we must consider the context and balance multiple factors. We advocate for a pluralistic framework: black-box AI is ethically permissible if it is highly accurate and reliable, and if other conditions—such as patient consent and regulatory oversight—are met. Accuracy alone isn’t enough, but neither is explainability; justification depends on the specifics of each decision.
We should therefore adopt a flexible approach to AI ethics in healthcare. Instead of one-size-fits-all standards, ethical use should depend on balancing factors like patient needs, urgency, benefits, and risks. Black-box AI, deployed responsibly, can be justified in healthcare settings—even without perfect transparency.
