It’s incredible that ChatGPT for Health will revolutionize patient access to health insights… BUT I worry about the shift from insights to AI self-diagnosis - and the downstream healthcare costs that will be borne by society.
Look, I'm a pro-AI tech startup founder. I use AI daily to solve critical problems in my business. I too consult AI for areas outside my personal expertise.
But here’s the difference: if AI gives me bad business advice (which it confidently is prone to do on occasion), I will bear the brunt of that cost. If I get bad corporate legal advice from AI that would otherwise have been caught by a reasonable lawyer, and I choose to follow it, it’s my business that goes under. Society doesn’t pay for that with their tax dollars or their lives.
In Healthcare, however, those “costs” are borne by all of society. If a patient uses ChatGPT for Health to self-diagnose (I know they say it’s not meant for that, but c’mon, it’ll happen), it’s going to report a wide range of possibilities - some likely, some not. While someone clinically trained like myself has the judgment to filter through the noise, the average patient won’t have that skillset.
With the convenience of AI self-diagnosis, we will see a surge in unnecessary tests, appointments, ER visits and procedures. This doesn’t just potentially harm the patient, but it creates downstream harm to acutely sick patients who will face delays in care. Because with AI self-diagnosis, we will see dramatically increased demand for care while supply remains finite.
We’ve discussed this very same concern with the growth of full body MRI startups - except that with those companies, demand was bottlenecked by wealth (i.e. only affluent folks would pay $2,500 out of pocket for a full-body MRI). But as AI like ChatGPT is close to free and universally accessible, these downstream consequences go up 1000x.
AI for consumers isn’t designed to think about society. It’ll do whatever the consumer wants.
Whereas clinicians, administrators and policymakers work to be stewards of limited healthcare resources. They think about cost-benefit at a population level. They realize we can’t afford to chase every unlikely diagnosis.
AI can aggregate your personal health data, analyze your test results, interpret medical images and suggest treatments for infinity… BUT what good is that, if for actual care delivery, the resources - the beds, ERs, nurses, doctors - are all constrained?
If we can’t stop the AI self-diagnosis train (and I don’t think we can), then the impetus is on clinicians to ensure we have a hand in designing purpose-built clinical AI that patients will use instead - so that we can help ensure AI can be an ally for sustainable care delivery, rather than a machine for churning out "what-ifs" that break the system.
