An NBER working paper published earlier this year uses the framework of “AI exposure” versus “AI adaptivity” to show the different ways AI could impact healthcare productivity and employment. A job that is “exposed” to AI is one where many of its tasks can be performed by AI. A job that is “adaptive” to AI is one where the skills required will be in more demand as AI proliferates. This produces a two-by-two grid.
 
The jobs most likely to get automated and eliminated are the “high exposure, low adaptivity” roles, like medical scribes and administrative assistants. “High exposure, high adaptivity” jobs, which includes most clinicians, could be greatly transformed by AI, but for every job or task that is automated away, another one should be created that is a better use of their time. “Low exposure, high adaptivity” jobs, such as lower-license nursing roles, are less likely to be transformed by AI, but could see a productivity boost by automating away certain routine tasks. Finally, “low exposure, low adaptivity” jobs should see little impact from AI, for better or for worse. These jobs tend to involve manual labor and do not need much training. The throughline is that AI’s productivity impact comes from increasing revenue and decreasing labor costs, but not all jobs will do both. Different jobs will be cut to save money, transformed to make more money, or almost entirely unaffected. Provider organizations looking to make the most of AI should be sensitive to which is which.