AI Creators
Peter Diamandis (Moonshots - AI playlist)
Optimus Surgeons in 3 Years | MOONSHOTS
Mar 20, 20261h 30m watch
Key Points
- 1.Three medical-AI founders discuss the realistic 5-year path for AI in healthcare with refreshingly little hype.
- 2.Consensus: AI augments rather than replaces, but the day-to-day work itself changes substantially.
- 3.Concrete bottlenecks: multi-site data access, regulatory alignment across jurisdictions, and clinician trust calibration.
- 4.A useful disagreement on whether primary care or specialty care will see the biggest near-term impact.
Summary
What's already real
Radiology second opinions running on millions of scans annually, ambient scribing in primary care visits (cutting documentation time by 30-60%), and triage chat handling routine intake before a clinician sees the patient - all in production today at meaningful scale, with measurable patient and provider outcomes.
What's blocked
Three different unlocks needed: multi-site data access (HIPAA-compliant federation across health systems is harder than the engineering problem suggests), payer/regulator alignment (FDA approval pathways for adaptive AI are still being defined), and clinician trust calibration (overconfidence kills patients; underconfidence wastes the technology).
Where they disagree
One founder argues primary care sees the biggest impact first because the cognitive load is high and the diagnostic surface is broad. Another argues specialty care wins because the data is cleaner and the workflows are more amenable to automation. The third sits in the middle and points out that the bottleneck is reimbursement, not technology - whichever pathway gets paid for first will define 'where AI helped'.
The trust calibration problem
An extended discussion on what it takes to make a clinician genuinely trust an AI second opinion - neither dismissing it as noise nor deferring to it as oracle. The founders agree this is more important than raw model accuracy and that current product UX largely fails at it. Examples of what works: confidence intervals shown by default, citing the training cohort, surfacing dissimilar past cases.
Five-year picture
Doctors keep their jobs; their day-to-day looks unrecognizable. Documentation time collapses, diagnostic breadth increases, and a new role of 'AI oversight specialist' emerges in larger systems. Patient-facing AI tools become normal but always sit behind a clinician for anything beyond information triage. The thing nobody on the panel is willing to predict: what happens to medical training when the diagnostic task itself is partially automated.
Worth watching for
Founders, clinicians, and policy folks tracking AI in health.
- healthcare
- panel
- long-form