Recently, a friend from the world of business told me a story.
They had a new project and reached out to one of the partners from a different company to get things started, to get things rolling.
They had a generative artificial intelligence (AI) program write out the preliminary proposal, including the goals of the project, the proposed timeline, the expected budget, and the rest of those nuanced business-y details.
When they got a response, they asked their AI widget to read the email, summarize it, and prepare a response. There was quite a bit of back-and-forth, and eventually a final fully fleshed-out project was produced and ready to be undertaken.
That’s when someone took a look, and realized the entire project had been developed without any human intervention — just two different AI chatbots talking to each other, back and forth, refining, changing, each responding to the other one.
Now, this story may be apocryphal; just be one of those things people are starting to fear, or maybe hope for, in the new world where AI is so deeply enmeshed in everything we do.
AI is starting to burrow its way into healthcare. In many ways it’s been there in ways we have not even known about, and it’s likely coming in ways we cannot even imagine.
Those of us who have seen this sort of tidal wave of change before know that not everything is all it’s cracked up to be, the final be-all and end-all that’s going to make our lives and our patients’ lives better.
At this point, none of us feel that we’re ready to turn over real medical decision-making to AI, but we do foresee a time when it will be able to take some of the routine follow-up and rote escalation pathways and standardized clinical care and help us implement them more efficiently and effectively.
Right now, all of us in primary care are frankly overwhelmed and a bit burned out by the volume of stuff coming at us.
We are being asked to see more patients with less support, crush more and more patients into an already overbooked schedule, extend our hours, handle things through calls and video visits and portal messages when we otherwise wouldn’t be expected to be working.
And each of these clinical interactions leads to more work, without concomitant increased support, and more results to deal with, more consults to make and interpret, not to mention the onslaught of portal messages.
The portal has led to patients feeling they can send us a long list of things they want us to take care of, and we’ll just do it, because we have nothing better to do.
We hear from patients saying we’re not spending enough time with them, we’re not really explaining things, we’re not really listening, we’re not helping them understand their health the way they wanted it explained.
We’re told that simple advice like eat healthier, get more sleep, exercise more, is not nuanced and tailored enough, and that if we were really going to help them make lasting changes, we should be putting in a lot more effort.
I wish we had unlimited time to spend explaining everything: why we order every test, why we sent someone to a consultant, what are the potential next steps, and on and on. But we don’t have the time.
Our patients are starting to get health information from these AI programs, putting their constellation of symptoms in and saying, what could this be? Then they turn around and send us the output, and ask us what they should do next, or request we order all the things the AI recommended.
Right now, when someone sends us a portal message that we need to expend clinical energy on, patients are told to expect to be billed for these services.
We’ve occasionally gotten pushback from people who sent portal messages with a whole bunch of issues they wanted addressed, including labs, imaging, referrals, and other questions, and then questioned why they got a bill for service they thought they were entitled to.
Maybe someone out there is working on a program that will respond to these AI-generated lists of questions from patients and come up with a series of AI-generated answers.
True, our electronic medical record in its current iteration uses AI to draft a first response to the questions in patients’ portal messages, which we are expected to review and edit and send back to the patient. This program has gotten quite a bit better from the first version, which pretty much regurgitated the patient’s question and said sorry that they had this symptom and recommended getting evaluated with an office visit.
Now the answers are much more nuanced, and are starting to look like the kind of things we would write, with little editing required.
It’s not quite all the way there yet, not yet ready for prime time. But, much like the business email conversation I described, perhaps if the patient has AI-generated questions, they should be prepared to receive back AI-generated answers, even if lightly edited by us.
Hopefully soon these programs will begin to act clinically, going down the clinical pathways our brains turn to as we take a history and do a physical and assess lab and imaging results.
But for me, an even better solution would be to train more primary care doctors and give us the professional support we need, so we can give patients the time and attention they want. Until we see this, patients aren’t going to be happy without the nuanced answers they expect.
But if that’s not going to happen, if we’re not going to rebuild this country’s healthcare system with a massive influx of primary care providers paid well and supported professionally, then maybe they should let us participate in the process of building these AI solutions.
Or else we’re just going to have your AI chatbot talking to mine.
