As Sir Winston Churchill once said, “Where there is great power, there is great responsibility.” This is the overarching context of our latest episode about the future of artificial intelligence (AI) in physiology and healthcare. Listen as host Dr. Ryan Melvin (University of Alabama at Birmingham) interviews author Dr. James Zou (Stanford University) about the Review by Zhang et al. on “Leveraging Physiology and Artificial Intelligence to Deliver Advancements in Healt...
As Sir Winston Churchill once said, “Where there is great power, there is great responsibility.” This is the overarching context of our latest episode about the future of artificial intelligence (AI) in physiology and healthcare. Listen as host Dr. Ryan Melvin (University of Alabama at Birmingham) interviews author Dr. James Zou (Stanford University) about the Review by Zhang et al. on “Leveraging Physiology and Artificial Intelligence to Deliver Advancements in Healthcare AI in Physiology and Healthcare.” The authors discuss an important transition in AI –the shift from model development to model deployment, while keeping at the forefront the goal of achieving positive real-world impacts by using AI in healthcare workflows. Zou and co-authors identify 3 key factors in this gap between development and deployment: ensuring AI works reliably and robustly across diverse populations, ensuring the financial sustainability of AI models, addressing new regulatory challenges for AI algorithms. FDA regulators are presented with unique challenges regarding how to continuously evaluate and monitor AI algorithms, while still allowing developers to update algorithms in a relatively frictionless way that ensures algorithms’ behavior is safe and robust. Does AI face a generalizability crisis? According to Dr. Zou, this is an ongoing issue. Instructive fine-tuning, which is intended to make iterative large language models safer, can sometimes lead to surprising AI behavior changes. On the topic of AI and authorship, the experts acknowledge that it is becoming increasingly difficult to distinguish if text is written by AI or by humans, as AI becomes more aligned with human behaviors. This may present unique opportunities, according to Dr. Zou. There are many types of text where human authors would benefit from AI support, for example in generating code and figures, and improving writing quality and precision. In the end, however, humans must take final responsibility for the accuracy of all statements in scholarly articles. How can the medical field play a role in shaping the metrics by which we judge generative AI? Listen to find out.
Angela Zhang, Zhenqin Wu, Eric Wu, Matthew Wu, Michael P. Snyder, James Zou, and Joseph C. Wu Leveraging Physiology and Artificial Intelligence to Deliver Advancements in Healthcare Physiological Reviews, published July 19, 2023. DOI: 10.1152/physrev.00033.2022
View more
