Price:

AI & TECH

Wachter warns AI deskilling could trigger a medical death spiral

Friday, June 12, 2026 · from 1 podcast
  • AI scribes are ending the ‘pajama time’ physicians spend documenting after hours.
  • AI models now detect heart disease signals in EKGs invisible to elite cardiologists.
  • Clinicians who lean on AI assistants can rapidly lose their baseline diagnostic skills.

The productivity paradox is finally breaking. For decades, digitizing healthcare forced doctors to become high-paid data entry clerks, spending eight hours a week on “pajama time” filling out charts at home. The system was so sclerotic it remained the country’s largest user of fax machines long after drug dealers abandoned pagers.

Dr. Bob Wachter argues the first real win is the ambient scribe. On Freakonomics Radio, he described tools that record patient visits and synthesize notes, letting doctors at places like UCSF look patients in the eye again. Generative AI can now summarize a 600-page medical history in 30 seconds, catching a 20-year-old diagnostic error buried on page 397. It’s a wingman that satisfies both clinician sanity and hospital billing.

The real shift is diagnostic. Dr. Pierre Elias developed EchoNet, an AI that scans cheap, routine EKGs for hidden signs of structural heart disease. In a trial, elite cardiologists identified disease 64% of the time - barely better than a coin flip. The AI hit 78%. It doesn’t think like a doctor; it finds mathematical patterns in electrical activity that are literally invisible to the human eye. Elias cited a patient misdiagnosed with asthma whom the AI flagged for severe heart failure, leading to a life-saving transplant.

“The promise of AI ‘co-pilots’ carries a significant psychological risk: the atrophy of human expertise.”

- Dr. Bob Wachter, Freakonomics Radio

The breakthrough carries a trap. Wachter pointed to a Lancet study where gastroenterologists using AI to spot lesions during colonoscopies saw their accuracy improve - but when the AI was switched off, their performance plummeted below their original baseline. He calls it a potential “death spiral” where leaning on the digital crutch erodes the very clinical reasoning skills needed if the system fails.

Regulation is another hurdle. Wachter contends agencies like the FDA, built for static drugs and devices, aren’t fit for software that updates daily. He argues for a light touch to avoid stifling innovation, noting that professional conservatism and malpractice fear are stronger guardrails than any government rule.

The system is too broken to just hire more humans. Wachter’s optimism is rooted in that dysfunction. But success hinges on culture and workflow, not just code. The race is on to integrate AI without triggering the deskilling it’s meant to cure.

Source Intelligence

- Deep dive into what was said in the episodes

677. Can Backgammon Save Us from Ourselves?Jun 12

  • Watcher argues healthcare is the largest user of fax machines in the country and only recently ditched pagers after drug dealers did.
  • The transition from paper to digital records didn't transform medicine and in some ways made things worse, creating a 'productivity paradox' where technology failed to improve efficiency.
  • AI scribes using ambient intelligence now summarize patient conversations, grouping related symptoms and filtering out personal anecdotes, which restores eye contact and reduces physician 'pajama time'.
  • Watcher cites a case where a human error - misinterpreting 'PE' for physical exam as pulmonary embolism - stuck in a patient's record for 20 years, arguing AI summarization can reduce such mistakes despite hallucination risks.
  • Pierre Elias says traditional cardiovascular disease detection via cardiac catheterization is too invasive for healthy patients, and echocardiograms are too expensive for population screening, creating a diagnostic gap.
  • Elias's AI model, EchoNet, analyzed ECGs from nearly 20,000 patients and predicted structural heart disease with 78% accuracy, outperforming cardiologists at 64%.
  • In the CACTUS trial, AI screening in eight New York emergency departments found half of high-risk patients for undiagnosed structural heart disease weren't getting follow-up echocardiograms.
  • Watcher says Epic dominates the EHR market with records for 325 million people, winning through integrated solutions, but its closed system may slow AI innovation by locking out third-party tools.
  • Watcher argues FDA-style regulation isn't fit for most AI tools because they shape-shift and update constantly, unlike static drugs or devices, and that overregulation now poses a bigger risk than underregulation.
  • A Lancet study found gastroenterologists' colonoscopy performance declined after a three-month AI assist was removed, showing rapid deskilling even in experienced doctors.
  • Watcher says patients in blinded trials often rated AI as more empathetic than doctors, though the AI is merely mimicking empathy effectively.
  • Watcher remains optimistic about AI in healthcare because the system is 'so screwed up' and can't afford to simply hire more humans, but success depends on culture, regulation, and workflow integration more than pure tech capability.
Also from this episode: (3)

Business (2)

  • Bob Watcher says healthcare delivery remains sloppy due to high fixed costs, powerful incumbents, and a 'funky' economic model where customers don't directly choose vendors.
  • Watcher says billing creates perverse incentives, as reimbursement hinges on how a doctor phrases notes, not just the care provided.

Culture (1)

  • A 2012 JAMA crayon drawing by a seven-year-old captured the doctor-patient relationship breakdown, showing the doctor typing with his back turned.