I’ve been learning everything I can about what AI will do to help healthcare achieve its potential, and especially how it will help e-patients be stronger contributors. The game’s not over (this game will never be over) but so far, this is the book! The AI Revolution in Medicine: GPT-4 and Beyond.
The authors – a computer whiz (Lee), a patient / journalist (Goldberg), and a physician / informaticist (Kohane) – know the technology intimately: they started using it before it was even called ChatGPT and finished writing the book this March, when GPT-4 was a few months old.
The authors had all caught my eye earlier
In September I heard Lee and Kohane speak at different events: Lee shared the opening keynote at the HL7 FHIR annual meeting, and Kohane gave one of the most important talks at the DCI Network AI workshop. I know Goldberg’s work as the “Commonhealth” health and science reporter for WBUR (Boston NPR – I worked with her on Warner Slack’s obituary) and as Boston bureau chief for the New York Times. What struck me about both Lee’s and Kohane’s talks is that while many observers pontificate from outside the “swimming pool,” both of these guys speak like experienced divers who can describe what’s really important and why: what mistakes can get you into trouble, what innovations are truly astounding.
I want to emphasize that again: in the hundreds of conferences I’ve spoken at, I’ve met many many people who are just learning and earnestly doing their best to understand new topics – it’s a good reason to attend such events – but I’ve only seen a handful of presenters who really delivered a mind-pop that explains so much and opens new vistas, because they’ve really been there on the front lines and they know what matters. These three authors are all in that class.
Similarly, in print there are many “oh wow” writings about ChatGPT and its peers. But if you want to get up to speed and down to earth, to understand what the important issues are (and will be) with generative AI in medicine, this is the book. My favorite Amazon review (a physician scientist) says he read it in one sitting, and cites:
…never-before-possible opportunities to level the playing field for access to medical care, and equally important, to improve the business end of healthcare — physician overload, financial devastation of missed pre-auths, and the like. This book points the way to a once-in-a-generation revolution in healthcare that’s unfolding before us in real-time.
Yes, a once-in-a-generation revolution that’s unfolding right now.
A few highlights
I’ll add a few of the high-impact highlights that struck me. This is far from a complete list.
- Clinical work: it’s already helping clinicians (docs & nurses) with decision making, diagnosis, and lots more. This is hard work, and it really helps to have a smart assistant.
- It helps with the killing amounts of paperwork and busywork they do: insurance paperwork like prior authorizations and much more. All that time is unpaid, aka volunteer time!
- It can write visit notes into the computer in standard formats such as SOAP (subjective, objective, assessment and plan) and include appropriate ICD-10 (diagnosis codes) and CPT (Current Procedural Terminology) billing codes.
- For patients, it can summarize long tedious visit notes into human-readable instructions (as I did with mine)
- Summarizing highly technical research articles. (I did that today: “Please summarize this article in 500 words or less, at 12th grade reading level.” Boom: instead of 20 minutes of hard thinking, it took 45 seconds.)
But the BIG benefit will be…
… that it will help the billions of people who can’t access care at all.
That includes impoverished nations and it includes developed nations where many can’t get an appointment soon.
At the DCI Network event, Zak Kohane said a mind-bending thing. We’d been talking about various ethical challenges (a favorite AI discussion), and he said “The real ethical problem is that half of humanity has no access to doctors at all.”
He challenged us to imagine a remote village where the best available medical person is a nurse with modest training. Imagine a patient with complex symptoms, alone in that village. Now imagine that the nurse feeds ChatGPT whatever information is available, and asks, “What could this be?” (Read the book to see the kind of answers it produces to such questions.)
That is an astounding thing to say.
A key factor is that it can bring any level of knowledge down to your level, which will make it possible for you to do things you never could. And that will vastly multiply how much expertise is available anywhere in the world. Which is especially important where there is no expertise.
Developed nations fret about how doctors and governments should regulate AI, but there are two countries that are eager to implement: China and India (1.4 billion each), because half their countries have no primary care at all.
Note – I say this sharply: access to care isn’t just a third world problem. Many people in the US, UK, etc have to wait weeks or months to get an appointment. In such situations you and I now have an expert resource available that we never had before.
It ain’t perfect but it will change nearly everything
To be clear, this type of AI is both astoundingly sharp and astoundingly dumb, and nobody knows why. Nobody told it to go learn math, but it did. Nobody told it to learn programming, yet it did. Nobody told it how to express a prescription using HL7 FHIR code so that other computers can ingest it – but GPT-4 somehow learned it.
Yet it can’t solve Sudoku puzzles! And unlike most of its errors, it can’t find its own Sudoku mistakes and fix them when pointed out!
Without question, for the foreseeable future it will have to be used with a “grownup” (a human) checking its work. As the book says:
Coming to grips with this dichotomy–that it is at once both smarter and dumber than any person you’ve ever met–is going to be one of the biggest questions and challenges in the integration of GPT-4 into our lives, and especially in medicine when life-and-death decisions might hang in the balance.
Start using it. Yes, you.
I dare you to give it a try chat.openai.org on something, anything, and see what you get. This book finally got me to do that, today, because (like swimming) you won’t know until you do. I asked it to outline a new speech I’ll be giving next week. I expected a half-clueless good guess, but look what it did.
Give it a try, and if you want to understand the future of healthcare, read this book.