By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.
Peter Attia MD309 ‒ AI in medicine: its potential to revolutionize disease prediction, diagnosis, and outcomes
Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.
At a glance
WHAT IT’S REALLY ABOUT
Zak Kohane on AI's Path to Transforming Medicine and Beyond
- Kohane traces AI from early perceptrons and 1980s rule-based expert systems to today's data-driven wave, which he attributes to large datasets, deep neural networks and GPUs, with transformers arriving as a further surprise.
- He argues that current technology already suffices to transform medicine by lifting the bottom half of doctors to the level of the top half and by equipping nurse practitioners and physician assistants to cover a growing primary care shortage.
- The conversation covers AI's promise for early disease prediction, such as from retina images, gait, speech and health records, and the obstacles around data access, HIPAA and unfriendly health data systems.
- Kohane predicts robot-controlled routine prostatectomy within ten years, sees AI as potentially acceptable for some mental health care, and worries that the medical establishment may use AI to preserve the status quo.
- Beyond medicine, the two discuss misuse by people, military autonomy, job displacement, social media chaos, AI creativity and idols, and digital legacy through tools like Rewind.ai.
IDEAS WORTH REMEMBERING
5 ideasToday's AI rests on three ingredients
Kohane says large online datasets, multi-layer deep neural networks, and GPU hardware drove the breakthroughs, with image recognition leaping forward around 2012 and transformers arriving later.
Goalposts for intelligence keep moving
Kohane notes that each AI milestone, from chess to face recognition to writing a rap, is dismissed once achieved, and he finds the debate about whether it is truly intelligent no longer interesting.
No new technology is needed to transform medicine
Kohane argues that even if progress stopped now, the barriers are building models, fitting them into workflow, and creating payment and business models, and that raising the bottom half of doctors to the top half would be transformative.
AI augments before it replaces, especially amid shortages
Kohane says AI has not replaced radiologists partly because there aren't enough of them and context-aware transformers were needed, and he sees AI-equipped nurse practitioners and physician assistants filling primary care gaps.
Early prediction is feasible, but data is the bottleneck
Kohane believes detecting conditions like Alzheimer's early from retina, gait, speech and eye movements is fairly straightforward, and that long-term records such as those from Clalit and Kaiser, plus patient access to their own data, can substitute for waiting thirty years.
WORDS WORTH SAVING
5 quotesIf everything stopped now, we could already transform medicine. It's just a matter of the sweat equity to create the models, figure out how to include them in the workflow, how to pay for them, um, how to create a reimbursement system that, uh, and a business model that, uh, works for our society. But there's no technological, um, barrier.
— Isaac "Zak" Kohane
I know we think that we're all very smart, but the fact that you and I, without thinking too hard about it, can come up with fluid speech, okay, so the model is now a trillion parameters. It's not a simple Markov model, but it's still a model.
— Isaac "Zak" Kohane
The rest of us, I think because we value our own intellects so much, we couldn't imagine how we would get that kind of conversational expertise just by scaling up the model and the dataset.
— Isaac "Zak" Kohane
What I'm most concerned about is how it could be used by the medical establishment to keep things the way they are, to pour con- pour concrete over practices.
— Isaac "Zak" Kohane
Waiting thirty years to have the perfect dataset is not the right answer to help patients now.
— Isaac "Zak" Kohane
High quality AI-generated summary created from speaker-labeled transcript.
