Artificial Intelligence
How AI failed in the face of COVID
Five practical lessons and warnings for data scientists
The sudden hit of the COVID-19 pandemic found the doctors and the hospitals completely unprepared. Too little was known about the new virus and too many patients were queueing at the door to diagnose and triage them correctly and quickly enough. AI to the rescue!
Machine learning models for diagnosing COVID or predicting the illness’ severity started springing up like mushrooms. Some of them were adopted by hospitals and used “in production”, impacting which patients were treated, and how. A recent study in the British Medical Journal looked at over 200 such models to find that none were fit for clinical use and some are potentially harmful to the patients.
Many parts of the system are to blame, starting from the lack of cooperation between machine learning and medical researchers, to the way the world of research works, promoting individual inventions rather than collective effort, to the lack of common data standards. All of this is discussed in this excellent article in the MIT Technology Review, alongside some remedy suggestions.
In this article, let me focus on the only part I feel competent enough to discuss: the machine learning algorithms and…