How should health systems put ethical AI into practice?

Healthcare systems are increasingly aware of the risks associated with artificial intelligence (AI), such as algorithmic bias and flawed modeling techniques, yet understanding these hazards doesn't necessarily translate to knowing how to mitigate them. In an upcoming panel discussion at the HIMSS24 AI in Healthcare Forum, experts will address questions surrounding the identification of biased AI systems, risk mitigation strategies related to training on incorrect datasets, and the establishment of an ethical AI culture emphasizing transparency and trustworthiness.

The discussion will be moderated by Anne Snowdon, HIMSS director of clinical research, and feature Michael J. Pencina, chief data scientist at Duke, and Brian Anderson, former chief digital health physician at MITRE and newly appointed CEO of the Coalition for Health AI. Anderson highlights challenges in ensuring AI fairness, particularly regarding biased training data sourced from academic medical centers, and the need for intentional partnerships to address dataset biases.

Anderson emphasizes the importance of transparency and accountability in the development, deployment, and monitoring of AI models in healthcare systems. He advocates for upskilling healthcare providers to critically evaluate AI models' suitability for patient care and stresses the necessity of ongoing governance and monitoring to address model drift and performance degradation. Anderson acknowledges the resource constraints faced by many health systems and suggests scalable solutions and partnerships to support responsible AI adoption.

Looking ahead, Anderson discusses the urgent need for consensus on evaluation frameworks for generative AI and the potential of public-private partnerships to drive progress in AI regulation and governance. He expresses excitement about the collaboration between leading technical experts from the private sector and government agencies in shaping the future of responsible AI in healthcare.