Panel, Conference Presentation
Artificial Intelligence and Machine Learning in Medicine: Hope or Hype?
Milken InstituteRick Berke, Rowan Chapman, Iya Khalil, Lloyd Minor, George Yancopoulos, Megan Zweig, Dan Furstenberg
- Artificial intelligence and machine learning are predicted to reduce drug discovery timelines, lower R&D costs, enable more precise treatments, and improve access to healthcare while allowing physicians to focus on high-touch care through automation of data mining and voice-recognition tools.
- Significant expansion of genomic data resources is planned, including a private biopharmaceutical collaboration aiming to sequence hundreds of thousands and soon millions of individuals, with a specific near-term target of over 250,000 people via partnerships with health systems like Geisinger, contrasting with the current zero linkage in the NIH Precision Medicine Initiative.
- Material limitations in current drug approval include a global average of 10 to 20 new first-in-class drugs annually, a clinical trial failure rate of 99%, and a lack of new treatments for major diseases like obesity and type 2 diabetes, though specific therapies for liver disease in morbidly obese patients are expected within a few years and the Virta program shows 60% diabetes reversal at one year pending approval.
- Structural risks and disincentives stem from the US insurance model which typically covers patients for only one to two years, creating a misalignment for preventative AI investments that yield benefits four to five years later, alongside rising clinical trial costs of approximately 10% if molecular and phenotypic data collection is mandated without tax offsets.
- Talent acquisition and integration challenges include a shortage of advanced engineers with 20 years of experience, a reliance on physicists and mathematicians transitioning into biology, and a predicted slow evolution of medical school curricula, though the field expects a merger of biological sciences and data science rather than the replacement of physicians.
- Ethical and regulatory risks involve potential "Minority Report"-style scenarios from unregulated AI behavior predictions, concerns over passive data collection in mental health apps, and "AI washing" in the venture landscape, while the FDA is encouraging AI use through incubator programs and has already approved AI diagnostics for diabetic retinopathy.
- Future technology goals include the development of a Bloomberg Terminal-style platform for healthcare professionals, the transition from empirical discovery to a quantitative data cycle, and a shift from pattern recognition to learning causal mechanisms, with expectations that sequencing costs will continue to drop to make data science skill sets more viable.
- The current healthcare system is characterized as overwhelmingly reactive with diagnostic technologies at a fraction of their potential and Electronic Health Records (EHR) limited by billing needs rather than clinical decision support, necessitating a massive shift in investment toward prediction and prevention over short-term venture exits.