John Halamka: AI Helps Specialists Detect Cancer, And More, Before Clinical Diagnosis
4 minute read
Between Code and Consciousness is a series by The Beautiful Truth asking the question: What does it mean to think, create or decide in the age of AI? Nine leading voices reflect on artificial intelligence – not as an abstract force, but as a tool whose worth depends on how it honours our humanity.
Dr. John Halamka is a former emergency physician turned global health technology leader. When the Obama administration needed a blueprint for making medical records shareable, they turned to him. Today, as President of Mayo Clinic Platform, he oversees one of the world’s largest health AI ecosystems, blending millions of de-identified records, scans and genomes.
“Anyone can make an algorithm; that’s not the hard part. The real challenge is making it trustworthy, explainable and equitable.”
For patients, the promise of AI carries a bright future. But for clinicians, how do they adapt?
People want self-driving cars to be 10,000 times safer than human drivers. If a person bends a fender, it’s just a human mistake. If AI does it, it’s front-page news. That’s the cultural challenge around trust.
But peer experience changes things. A doctor says: “I used this tool and it helped me with a tough case.” Or: “It suggested an option I hadn’t even considered.” And suddenly, colleagues pay attention.
When did early detection move from theory to reality?
In the 1980s, we were just capturing digital data. In the 1990s, the problem was standardisation: ‘hypertension’, ‘high blood pressure’, ‘elevated blood pressure’ all meant the same thing. By the 2000s, I was working on national data standards so information could be computable and exchangeable.
By the 2010s, we could finally turn data into wisdom. My wife was diagnosed with stage 3A breast cancer in 2011. She was 50, an artist, and asked: “What’s the best chemotherapy for someone like me, who needs to keep using her hands?” At the time, no system could personalise care at that level. So, I started building decision-support tools to learn from millions of past patients, to guide the one sitting in front of you.
“[My wife] was 50, an artist, and asked: “What’s the best chemotherapy for someone like me, who needs to keep using her hands?” At the time, no system could personalise care at that level. ”
John Halamka
What can Mayo Clinic do today that simply wasn’t possible before?
In 2020, I was asked to curate Mayo Clinic’s 150 years of data – text, images, heart telemetry, genomes, pathology slides – to build a multimodal dataset for AI, while respecting privacy and consent. Since then, we’ve created hundreds of predictive algorithms and several foundation models that do what humans can’t.
My father-in-law died of pancreatic cancer in 2014. It’s usually found too late, invisible on scans. Now Mayo Clinic has an AI model that can help specialists detect it at stage 0 – up to three years before clinical diagnosis. We can also forecast conditions you don’t yet have, such as atrial fibrillation. And in endoscopy, a specialist might miss 15% of small colon polyps, but a specialist using AI only misses 3%.
After decades of working at the intersection of medicine and technology, what gives you the most optimism about where AI is heading?
Timing matters. If you regulate too early, before you even know what you’re regulating, you quash innovation. If you ignore the risks, you create harm. The way forward is collaboration – government and industry working side by side, defining ethical practices and turning those into regulation when the problems are understood.
Anyone can make an algorithm; that’s not the hard part. The real challenge is making it trustworthy, explainable and equitable. At Mayo Clinic, we show that AI isn’t about replacing doctors, it’s about giving them additional insights. If we do that right, we won’t just improve medicine in Boston or London – we’ll bring safe, personalised care to billions of people worldwide.
