Viewpoint: AI is incompatible with the real-world practice of medicine

Though some artificial intelligence algorithms developed for use in healthcare have advanced to the point of outperforming clinicians, they still lack the fundamental ability to interpret clinical data as only humans can.

Because of this, according to a recent commentary in The Lancet, while AI can certainly play an invaluable role in patient care, it is ill-equipped to analyze patient data to identify anything other than a specific, immediate issue, the parameters of which must be explicitly defined when building each algorithm.

The op-ed's authors — comprising researchers from the Harvard School of Public Health in Boston and Microsoft Research in Redmond, Wash., among others — described three technical challenges associated with the use of AI in healthcare:

1. Clinical data is messy: Data found in EHRs, medical histories and other patient records are often disorganized, incomplete and prone to bias. Though human clinicians can typically correctly interpret this data based on their past interactions with a patient, algorithms are unable to make assumptions without sufficient data to back them up.

2. Algorithms cannot account for the unknown: Other analytical issues arise when real-world data differs from the data upon which an AI model was trained, such as when an algorithm trained on data from an urban hospital is applied to a rural setting, or when a new disease outbreak occurs.

3. Medical practices are far from objective: Not only do clinicians use a multitude of labels and wording to describe the same conditions, but they also rely heavily on human judgment to diagnose those conditions — a level of subjectivity that would be nearly impossible to program into every single AI algorithm.

More articles on AI:
Mayo Clinic invests in breast cancer imaging AI
3 lessons NewYork-Presbyterian learned from using AI to reduce length of stay
CHI Franciscan launches AI-powered Mission Control hub for care coordination

© Copyright ASC COMMUNICATIONS 2019. Interested in LINKING to or REPRINTING this content? View our policies by clicking here.

 

Top 40 Articles from the Past 6 Months