Illustration from Google Labs (Prompt: sketch of a computer giving medical advice)
Projections about the impact of artificial intelligence (AI) on health care delivery range widely. We fail to appreciate how much AI is already used in medical care (such as radiology readings and AI-based transcription). I think some overestimate how much change we’ll see in the next year or two, and many underestimate how much AI will change the world in a decade.*
Harvard economist David Cutler has a short, thoughtful article in JAMA Health Forum (no paywall) with these observations:
Most labor displacement in medical care will be replacing rote tasks - and this is likely to be gradual
AI is likely to complement rather than displace clinicians, who can recognize subtle patterns that AI could miss.
AI will allow clinicians with less expertise to function at a higher level -allowing substitution of less expensive clinicians for higher priced clinicians.
AI will enable out-of-hospital monitoring, which could substantially lower total health care costs
AI will need to ingest vast quantities of data to continue to improve
I’d add to his observations that we demand a higher degree of accuracy from computers than from flesh-and-blood clinicians. AI can be used to improve insights into social determinants of health, but it is trained on large amounts of historical data that incorporates structural racism and has the potential to exacerbate existing bias.
Cutler concludes that we will still need clinical trials and quasi experimental research to inform policy decisions which will need to be made by humans.
Implications for employers:
AI may improve both the business of medicine and patient care, although that won’t necessarily lower costs. For instance, AI’s ability to model drugs could lead to more pharmaceutical breakthroughs for which we will pay top dollar.
AI can be used to assess patient risk, which can help focus programs on those who will benefit most.
AI can be misused, too. Reporting this week about a Medicare Advantage plan which automatically denied continuing facility care shows the peril of algorithmic decision making without adequate human supervision.
Employers should monitor their vendors to be sure that AI-based decision-making does not exacerbate existing bias
* This is an adaptation of a Bill Gates quote “We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten years.”
Tomorrow: We have great drugs to lower cholesterol, but most with heart disease are not adequately treated
Please “like” this post and spread the word. Thanks!