Artificial intelligence has dominated the healthcare news cycle and hospitals have invested millions in its advertised ability to predict, analyze and make decisions faster and more accurately than humans can. Visar Berisha, PhD, and Julie Liss, PhD, believe the technology is overhyped and is due for solutions in their development, according to an op-ed for Scientific American.
Dr. Liss and MD. Berisha argue that academic publications often overhype the accuracy of AI algorithms used by healthcare organizations. They found that published AI models' accuracy declined when working with a larger data set.
When designing AI models, researchers train the model on a portion of their data set. The data left out from training is used to test the accuracy of the AI model.
However, Dr. Liss and Dr. Berisha pointed out that data leakage, sometimes unintentional, can occur and that some scientists build and test their AI models using held-out data. This practice would guarantee a high rate of accuracy as the model is modified.
A solution Dr. Liss and Dr. Berisha suggests using new data sets to test for accuracy. They also spelled out preventive measures.
"We can prevent these issues by being more rigorous about how we validate models and how results are reported in the literature," they write. "After determining that development of an AI model is ethical for a particular application, the first question an algorithm designer should ask is 'Do we have enough data to model a complex construct like human health?' If the answer is yes, then scientists should spend more time on reliable evaluation of models and less time trying to squeeze every ounce of “accuracy” out of a model. Reliable validation of models begins with ensuring we have representative data."
Taking a holistic approach to AI can push past hype and improve clinical outcomes and grow its reach, the writers say.
Dr. Liss, is the chief clinical officer and co-founder of Aural Analytics. Dr. Berisha, is an associate professor at the Temple-based Arizona State University and the chief analytics officer at Aural. Read their full commentary here.