Greater Consideration of Real-World Application in Development of AI Technologies for Health Urged: New Paper
In the latest edition of the Journal for Medical Internet Research, PCHSS investigator Professor Enrico Coiera argues that the creation and implementation of artificial intelligence in the health field should be driven by human needs rather than simply by what is technologically achievable. In “The last mile: where artificial intelligence meets reality”, Professor Coiera writes that “only the real world can tell us which problems are most worth solving”.
The development of AI technologies tends to follow three stages. The “first mile” entails gathering and organising data on a given area of concern, the “middle mile” concerns creating and refining algorithms and predictive models on this basis of the collected data, and the last mile involves the implementation of the technologies. But in a world of “technology neophilia” says Professor Coiera in his paper, more thought is often given to what is possible than what is useful. In the context of health, he gives the example of thyroid cancer:
[I]t is one thing to demonstrate machine learning can interpret thyroid scans for cancer as well or better than humans—a technical feat; it is another for that feat to be meaningful. In the current setting, where thyroid cancer is both overdiagnosed and overtreated, we do not necessarily need better diagnoses.
To address this challenge, Professor Coiera advocates restructuring AI development away from the linear three-mile framework and toward a more agile and iterative process. In this connection, he recommends giving increased attention to training and retraining AI in consideration of specific, local end populations and shifting “from measuring technical accuracy to evaluating impact on processes and people”.