Kira Radinsky, CEO of Diagnostic Robotics and co-founder of Mana.bio, is an AI entrepreneur who turns prediction into action. She discusses using ML to automate healthcare admin, build causal systems that predict outbreaks and patient twins, and accelerate drug delivery with closed-loop labs. Conversation covers incentivizing bold algorithms, ROI-aware models, and avoiding bias in healthcare AI.
38:47
forum Ask episode
web_stories AI Snips
view_agenda Chapters
auto_awesome Transcript
info_circle Episode notes
Use Patient Twins To Simulate Controlled Trials
Do use causal inference to move from prediction to recommended actions by finding patient 'twins' and mimicking controlled trials in observational data.
Kira's team finds matched patients who had different physician actions historically and infers which clinical next step prevented deterioration.
Mental Health Often Trumps Direct Clinical Priorities
Algorithms surfaced nonobvious clinical priorities, e.g., treating depression first reduced overall medication nonadherence more than switching antihypertensives.
That ranking came from historical outcomes showing behavioral effects cascade across other conditions.
Explain From Literature Not Just From Data
Explainability must go beyond dataset correlations to literature and biological mechanisms to convince clinicians.
Diagnostic Robotics built a system that reads medical literature and links correlations to external biological explanations before clinicians act.
Get the Snipd Podcast app to discover more snips from this episode
Of all the industries AI will transform, Kira Radinsky believes chemistry and biology will change the most.
Kira is the co-founder and CTO of Diagnostic Robotics, which uses AI to automate the administrative work that's crushing healthcare teams — so clinicians can actually focus on patients. She's also the co-founder of Mana.bio, where they're accelerating drug discovery by orders of magnitude.
She'll tell you she's terrible in the lab. Not because she isn't brilliant, but because she can't pipette without killing the cells. So she’s thrilled that thanks to her skills in data and AI she was able to realize her childhood dream of being a scientist:
“I'm not trying to automate everything… Like when, when you say automate drug discovery, I'm not gonna discover everything. I just want to accelerate it, which comes back to my childhood dream: I just didn't want to do it myself. I just want AI to replace me as a scientist. That's it.”
But this episode is about more than healthcare. It's about how to build systems that get smarter over time — feedback loops, causal inference, incentivizing algorithms to take risks, and knowing when to optimize for ROI instead of accuracy. Lessons that apply whether you're building in biotech or not.
We cover:
How growing up Jewish in Soviet Ukraine — and fleeing to Israel just before the Gulf War — shaped Kira's obsession with predicting the future
How she built a system that successfully predicted real-world events, including Cuba's first cholera outbreak in Cuba in 130 years
How Mana.bio is using AI to build "rocketships" that deliver drugs to the right cells — and how they've done in three months what used to take 20 years
Why predictions are only valuable if there's something you can do about them — and why that makes healthcare an ideal field for AI
How to incentivize algorithms to make bolder predictions (it's easy to predict there won't be an earthquake today; it's much harder to say there will be)
Why causal inference is the most underrated tool in machine learning right now
How healthcare AI can perpetuate racial bias — and what builders need to do differently
Note: this interview originally aired in October 2024.
Chapters:
(01:44) - Why predictions are so important to Kira: lessons from fleeing Soviet-era Kyiv
(05:10) - Building a prediction engine from 150 years of news
(08:35) - How Kira predicted the Cuba cholera outbreak
(09:50) - Returning to biology by way of data
(12:50) - Predicting healthcare outcomes by finding your patient's twin
(17:53) - The racial bias hiding in healthcare AI
(19:15) - Building Mana.bio and accelerating drug discovery
(24:33) - "In three months, what did what used to take 20 years"
(31:44) - Builder tips: ROI, causal inference, and teaching algorithms to explore
(35:07) - Planning: Where generative AI needs improve
Future Around & Find Out