While most of us spent our early teens dealing with the drama of middle school, Daphne Koller was in Israel simultaneously completing high school and college. She was a computer science prodigy on the fast path to a career as a leading AI researcher, an entrepreneur, and now the Chief Computing Officer at Calico, a stealthy, brainy, well-funded startup focused on human longevity.
Along the way, Daphne picked up a MacArthur “Genius” Award and co-founded the online teaching company Coursera – two remarkable accomplishments that we don’t have time to discuss on today’s show. Instead, we learn about her fascinating personal journey from Israel to the Bay Area, then spend most of our time getting up to speed on the current state of AI, and learn where, why, and when it’s likely to palpably impact healthcare.
Of particular interest, Daphne discusses the need for folk who are “bilingual” – who deeply understand both AI and healthcare; such domain knowledge, Daphne says, is critically important, and associated with the development of algorithms that perform the best. We discuss the challenge of balancing the benefits of incorporating domain expertise with the concern that in doing so you might introduce your own preconceived biases.
Thank you to Daphne for bringing her great energy and knowledge to Tech Tonics.
Today’s episode is brought to you by DNAnexus, the secure and compliant cloud platform that enables enterprise users to analyze, collaborate around, and integrate massive amounts of genetic and other health data.
Additional Notes:
· Thoughtful commentary by Andrew Orlowski on an early health/AI effort involving a rule-based approach – and what lessons we (should have) learned.
· Another thoughtful commentary, this by Brandon Ballinger of Cardiogram, entitled “Three Challenges for Artificial Intelligence in Medicine.”
· David’s Forbes piece from earlier this year on AI and healthcare.
This is a riveting talk. Daphne Koller is a thought leader for twenty years on AI and related things (caveat for things people don’t always agree on). But Lisa and David really bring their A game for this one. Clearly they’ve been thinking about this for some time. You can be an optimist but the more you know the more you need to know. It’s so fascinating. Or you can worry about the dystopian risks with the concentration of knowledge. No one can argue with the merits of avoiding disease through early detection, if it avoids overdiagnosis or diagnostic odysseys. But knowing how people are not great at prevention, you can envision an automated way that a therapeutic intervention — chemopreventive or vaccine or bot — could forestall disease progression at an early stage.
Training data, as the discussion covers, will need to be gathered more broadly and deeply. It’s early. I was thinking Martin Arrowsmith (great Sinclair Lewis book) in the 1920s knew his patients from womb to tomb. Does any AI system today have that kind of longitudinal data? I think not.
I think this has to come from people, not doctors. They have too much on their plate right now. And cultural change individually and institutionally comes too slowly. And people are focused on health, not just health care.
Marc, thanks for listening to Tech Tonics! We agree – Daphne is riveting. Best, Lisa