By Marie Daghlian
The buzz around artificial intelligence, or AI, is palpable—similar to the buzz around CRISPR and immuno-oncology just a few years ago. VCs are throwing money at startups that are using AI. Proponents say it can solve the world’s problems, do what humans cannot. Others worry that it will run away from humans, and foresee a future where intelligent robots run the world. Perhaps, but the reality today is neither.
“These are software toolkits that we are all trying to learn how to use and apply to biomedicine,” Atul Butte, director of the Institute for Computational Health Sciences at UCSF, says. He spoke at the 10th annual Precision Medicine World Conference held in January at the Computer History Museum in Silicon Valley where artificial intelligence as a tool for shaping and enabling patient-centric health care was a prominent theme throughout the three-day event. More than 1500 people packed the event space to hear about the latest developments in bringing precision medicine to patients.
Why is there so much interest in AI in biomedicine now?
Butte claims that it is due to the convergence of three events: amazing hardware coming from the gaming industry; amazing software support, often available as open source; and a plethora of data sources, molecular—DNA, RNA, proteins, electronic health record data, and digital health tracking data. The hope is that we will use AI to provide answers to unsolved questions, such as how to get more efficient drug discovery, how to predict patient outcomes, and how to do population modeling.
Butte offered a quote: “Asking doctors to treat cancer patients without the benefit of modern software is just the same as asking someone to drive at night without headlights,” made by Eric Lefkofsky, a co-founder of Groupon. He and Lefkofsky, now the founder and CEO of Tempus, discussed the use of AI to deliver real-time data driven decision-making to doctors. Tempus is a technology company that empowers physicians to deliver personalized medicine through data analytics.
Entrepreneur and venture investor Vinod Khosla says AI is intelligence “that does things that traditional humans can’t comprehend, and exceeding human capability, building models and complexities we couldn’t develop with traditional techniques.” AI depends on data, a lot of it. With enough data, he says it will be better than humans as “the medical specialist.” But we have a long way to go to get there.
“It takes twenty years for any new discovery to show up in practice,” Khosla says. “We’ve started labeling everything AI now and that’s a mistake. I see very simple statistical correlations called AI—this biomarker does that, or this metabolic pathway results in that. To me, those are linear systems and AI is not linear. What is exciting about new AI—what’s your microbiome doing, what’s your transcriptome doing, what’s your proteome doing, what’s your metabolome doing—add all this together and you start to get the basis, across a million people to diagnose most disease ahead of symptoms. This complexity is what will result in dramatic changes in medicine. My big beef is we are not collecting the right data; we are only collecting the data humans can read. I’m wearing an Apple watch. It’s running a predictive model of what my heart rate should be. When I get on my treadmill, the prediction goes up. Then it takes an ECG and the neural net running on the watch matches the results to a model of my heart—not your heart—on the watch itself. It tells me when I am out of bounds. It will take ten to twenty years for this to get into general practice.”
Khosla tells prospective doctors to be interdisciplinary in their studies: “understand math as much as you understand biology.” He predicts that in twenty years there will be no specialists: “I jokingly say the most important person in the future will be the primary care physician or the nurse practitioner because all of the AI, the decision-making part will be done by systems. You’ll have the best AI oncologist, the best AI cardiologist, the best AI gastroenterologist, and the integrating piece will be the primary care physician. And he’ll be focused on the human element of healing, not being Dr. House.”
He also bets that AI implementation will not start through physicians but through patients. These systems will enable patients to be CEOs of their own health, Khosla says, empowering themselves by generating their own data through mobile apps. It’s hard, it’s chaotic, it’s not systematic, it’s open to criticism, but it is the future.