Conference Presentation, Panel
Will AI Deliver on the Promise of Better, Faster, Cheaper Health Care? | Asia Summit 2025
Milken InstituteSung Hee Choe, Lohendran (Logen) Baskaran, David Berry, Adrian Lam, Wan Yue, Dr. Logan Baskerin, Dr. Yuan Yueyuan, Dr. Wan, Dr.-Vosker, Wanyu, Logan, UA, David Rockefeller Jr.
- Transformative AI Applications in Genomics and Drug Development
- DNA and RNA foundation models are now instrumental in personalized medicine, enabling clinicians to predict individual drug responses and adverse reactions based on specific genetic variants.
- AI has accelerated the development of next-generation mRNA therapeutics, resulting in molecules that produce higher protein levels, exhibit greater in-body stability, and allow for improved manufacturing compared to early pandemic-era vaccines.
- Clinical Efficiency and Diagnostic Accuracy
- AI algorithms have reduced cardiac CT scan analysis time from 3–4 hours (manual research) or 10–30 minutes (expert clinical reporting) to just six minutes for end-to-end processing, addressing significant capacity bottlenecks where daily scan volume has risen from 5 to 20.
- Pilot programs utilizing computational fluid dynamics and AI allow clinicians to infer dynamic hemodynamic data from static CT images, enabling non-invasive decisions on stenting procedures that previously required invasive cath lab visits.
- Entrepreneurial Challenges and Patient-Centric Design
- Successful AI health tools must navigate complex stakeholder ecosystems involving back-office administration, IT departments, nursing staff, and physicians, rather than focusing solely on clinical problems.
- Companies are increasingly leveraging low-cost, ubiquitous sensors (e.g., voltage gradients, PPG) combined with advanced machine learning to upgrade diagnostic capabilities without requiring expensive new hardware.
- Patient receptivity is high for non-invasive, rapid (3.5-minute) diagnostic tools that eliminate the need for hour-long travel to specialists, particularly in underserved regions like the US Deep South.
- Investment Thesis and Future Trajectories
- Investors are prioritizing companies that leverage AI to solve the human limitation of "dynamic incorporation," specifically targeting misdiagnosis rates in complex areas like mental health where the average patient receives three incorrect diagnoses before the correct one.
- The consensus among panelists is that AI will serve as an augmentative assistant rather than a replacement for doctors, as human liability, the "human touch," and complex decision-making regarding interim care remain essential.
- Rate-Limiting Factors: Data, Regulation, and Equity
- Data Sanctity: Strict, divergent global data rules are constructing "walls" that prevent the sharing of anonymized data, slowing AI development and risking the creation of models that only serve specific populations.
- Geographic Bias: Current clinical guidelines derived from US/European cohorts show 30–80% inaccuracy in Singapore due to differing risk factor weightings, necessitating the development of population-specific risk scores (e.g., the 60,000-patient GPCATS study) that account for race, environment, and lifestyle.
- Explainability: There is a critical distinction between "black box" deep learning models suitable for drug discovery (where preclinical checks exist) and "white box" statistical models required for frontline diagnostics to avoid unacceptable Type 1 or Type 2 errors.
- Forward-Looking Statements (Next 5 Years)
- Panelists envision a seamless preventive care funnel utilizing routine blood tests and demographics to identify future disease risks and intervention strategies before invasive testing is required.
- There is a projected shift toward remote patient monitoring and home-based diagnostic tools capable of processing poor input data to make rapid, actionable health decisions (e.g., overnight pediatric triage).
- Strategic Recommendations and Calls to Action
- Policy and Reimbursement: Policymakers and payers must adopt flexible risk-taking models for reimbursement to enable the commercial adoption of new AI technologies.
- Data Sharing: Regulators must take calculated risks to dismantle data silos, enabling the creation of diverse datasets that address racial and geographic biases in AI training.
- AI Literacy: Increasing public understanding of AI for both youth and the elderly is essential to bridge the growing gap between excitement and skepticism, fostering broader acceptance and usage.
- Philanthropy: Private philanthropy needs to increase investment in non-commercializable AI research to advance the field in areas that do not yield immediate commercial products.