Read the full transcript of Dr Edmund Jackson’s talk titled “How AI Can Heal Healthcare” at TEDxNashville 2024 conference.
Listen to the audio version here:
TRANSCRIPT:
A Pivotal Encounter
DR EDMUND JACKSON: 10 years ago, I found myself in the office of a singular individual, Dr. John Perlin. At the time, he was the chief medical officer for a hospital company called HCA Healthcare, where I was applying for a job as a data scientist.
And I said to him, “Dr. Perlin, what is it that you’d have me, a data guy, do for you?” And he proceeded to tell me about a disease called sepsis. It’s a systemic bloodstream infection that if identified early, can be very readily treated with about 25 cents worth of antibiotics and fluid.
And if it goes unrecognized, spreads through the body, leading to systemic inflammation, organ failure, and all too frequently, death. And Dr. Perlin proceeded to tell me at length about the etiology, biochemistry, pharmacology of this disease, and I understood none of what he was telling me.
And I said, “John, I really don’t understand. I’m not that kind of doctor. I’m a doctor of numbers.” And he said, “I understand. I tell you all this to convince you that as clinicians, we’ve tried everything we can do in the fight against sepsis, and we are still losing. Our care simply does not arrive in time to be relevant.
We need an early warning of sepsis. We believe it’s in the data, and that you and your fellow data scientists can find it.” What Dr. Perlin didn’t know was that 20 years prior to that conversation, sepsis had killed my father. He was 58, and I was a teenager.
So I found myself in a room opposite a man telling me that for want of 25 cents worth of drugs, my dad might still be alive, and moreover, that I could play a part in preventing other families going through what I did and my family did.
Well, I entered that office seeking a job, and I left carrying a mission.
The Paradox of Modern Healthcare
But what little did I know? You see, healthcare itself is sick. Something is very, very wrong in healthcare. You see, we’ve never had more well-educated, more dedicated, more specialized physicians. We’ve never had instruments and labs and diagnostic tools that are more precise. We’ve never had therapies and drugs that are more potent and can render what was once a death sentence as a mere inconvenience or a meme.
And yet despite all of this, healthcare is failing. On any measure that we care about, be it clinical quality, patient experience, and most of all, the four and a half trillion dollar bill that healthcare levies on America every year, things are getting worse.
But why? People tell me, “Edmund, it’s complicated.” No. It’s complexity.
In healthcare, we have never slain the demon of complexity. You see, in other parts of the economy, like technology or consumer goods, there are patterns and modes of thought, like standardization, automation, playbooks, that enable us to simplify what we’re doing by doing the same thing over and over again.
While Apple Watch is a great example of this, a tremendously complicated device, and yet one of millions of identical devices, that standardization enables Apple to make a device at low cost and high quality.
But in healthcare, we are the product, and there’s no such thing as a standard human. Heck, I don’t even know what size my shoe is half the time. And so everything we’ve learned and applied in the rest of the economy about simplification doesn’t work in healthcare.
And the cruel irony is that those new specializations of physicians, those new instruments, those new therapies, while individually good, combine to increase complexity and so make things worse. It’s a pure torture. What we need is a new mode of thought to simplify. We need new technology, and that technology is AI.
The Evolution of AI
What is it? An AI is simply a machine that thinks like a human, but at scale. The first generation of AIs were AIs that could do logic, as you and I do. There were enormous branches, trees of if-then-else statements, inductive-abductive logic, and they worked, but they were incredibly intricate, expensive to produce, and even more expensive to maintain.
And so they fell into disuse. They were the first and the second AI boom-bust cycle. The second generation of AIs were AIs that can do mathematics and see patterns, as we do. To share with data, the AI, through machine learning, can infer a model which makes predictions about the future. You know, we used to joke about the weatherman and how useless and unreliable they were.
And we don’t anymore, because aided by AI, they’re really damn good.
So, yeah, the third and current generation of AIs are machines that understand language, as we do. Kat just gave great examples of that. These are machines that can read and write, listen and speak in all of our human languages, but at scale.
And as a society, we haven’t yet figured out exactly what we’re going to do with all of this, but we know it’s big. One of the leading proponents is a company called OpenAI. It’s valued at $80 billion on revenues of one and losses of five billion.
And what is it? It’s a text box.
But it’s a text box, critically, that contains the whole internet. You see, OpenAI has a machine that has read the internet and will answer all of your questions about it. And that a text box containing the internet should be worth an unfathomable sum should come as no surprise, because that is precisely exactly what Google was for the last generation of AIs.
Moreover, it’s a pattern of thought. OpenAI did not seek to simplify the internet. They wrapped it in a very simple surface, that text box. They took the internet, can of worms, and they put the can around it.
AI in Healthcare: Wrapping Complexity
But that mode of wrapping complexity is what we need in healthcare. I’ll give some examples. My friend David, he’s a primary care physician, one of the most mission-centered and hardworking people I’ve ever met. He used to see between 20 and 30 patients a day in his practice, and then he would go to his computer and write notes about all of those interactions into the electronic medical record. Those notes feed the leviathan of healthcare administration.
All of the revenue cycle, insurance claims, billing, government compliance, regulation, legal disputes, quality, everything comes from those notes.
Now David has an AI. It’s on his phone and sits on the table while he’s consulting with his patients. It listens to their interaction and then writes the notes for him. A machine that listens and writes.
What it does is it wraps the whole administrative complexity for him in an interface so simple it seems invisible. Aided by this AI, David now sees between 60 and 70 patients a day, an unimaginable increase in his productivity and decrease in the cost that his patients experience.
Now David may be Superman, but he’s only one man.
Can we apply the same logic to clinical teams or whole hospitals? A hospital is about the most profound place in modern American life. Most of us came into this life in a hospital, and most of us will depart it from one.
And with all of the ambulances and helicopters arriving with gunshot victims and trauma victims, babies being born, hearts being transplanted, the complexity is overwhelming. It’s more than anybody can hold in their mind, but it’s not more than an AI can hold.
Imagine an AI wrapping the hospital, aware of the status of all of the patients, their journey and inferring the next best actions for all the clinicians and workers in the hospital and sharing it with them in their language, coordinating that care. Wrapping that hospital administration for the providers will dramatically reduce their complexity and increase their quality and productivity. A small example of this is the sepsis work I described earlier.
HCA brought together teams of clinicians and engineers, and I was privileged to be one of them. Together we wrapped the subtle complexity of sepsis development in an algorithm and a workflow, giving clinicians hours of advanced warning of sepsis.
And with that tool, they’re able to save thousands of lives a year. It was a lifetime privilege to be on those teams.
The Future of AI in Healthcare
But it’s no longer all that can be done. Machines that can read and write give us new opportunities. Consider the case of Martha Mills. She was a British teenager who, on a vacation with her family, slipped on her bicycle and jabbed herself in the abdomen with the handlebars. The pain was intense.
She went to the ER. It didn’t get better, and she was admitted. She deteriorated over the weekend and ultimately succumbed to death. It was a tragedy, an avoidable tragedy, because the signs and symptoms of sepsis were all over that chart, but they were missed in the clinical and all-too-human complexity of modern medicine.
Now imagine an AI that has read the medical textbooks, has read the clinical literature, and is connected to those charts in real time and can read them. Such an AI would know about sepsis and would not miss it. Most excitingly, imagine that AI exists outside of the traditional healthcare system, allied and aligned to Martha’s family, who were at her bedside, disempowered through this whole experience.
Such an AI could have told them, “Martha is critically, deathly ill and needs to be in a paediatric ICU right now. Here’s why.” Armed with that information, her parents would have advocated for her transfer and probably life-saving care. All of that technology is available today.
Conclusion
So, what ails us in healthcare is complexity, and the bitter irony, as I said, is that every attempt we make to improve matters with better tools, with better therapy, with better drugs, makes things worse by adding to complexity. We need to simplify. We can’t simplify by standardizing, so we must simplify by wrapping, and I’ve shown how we can wrap administration for a single physician, hospital operations for clinical teams in hospitals.
But most of all, I want you to leave knowing that AI can wrap the experience of healthcare for you and for me as consumers, empowering us as advocates, knowers, and deciders in our care and in the care of those we love. Thank you.
