What AI Companies Get Wrong About Curing Cancer (with Emilia Javorsky) - Future of Life Institute Podcast Recap

Podcast: Future of Life Institute Podcast

Published: 2026-03-20

Guests: Emilia Javorsky

What Happened

Emilia Javorsky argues that the promise of artificial superintelligence curing cancer is overly simplistic. Curing cancer involves more than solving intelligence problems; it requires addressing data availability, incentives, and coordination. Javorsky highlights the lack of accessible data for AI training, as most biological data does not follow a standardized format like other fields.

Despite the rapid increase in medical knowledge, with the doubling rate now at 73 days, cancer mortality rates have not significantly improved. There is a flat trend in FDA-approved drugs over the decades, and many promising drugs fail to pass through the FDA due to business and intellectual property issues. This reflects a mismatch between the pace of knowledge accumulation and clinical advancements.

Biology's complexity, marked by its lack of first principles, makes it difficult for AI to generate breakthroughs akin to those in physics or mathematics. Most medical literature may be unreliable due to reproducibility issues, and electronic medical records are often designed for billing rather than capturing clinical realities, leading to biases.

Cancer is a dynamic, evolving process with individual variability rather than a single disease. This complexity is demonstrated by the Hallmarks of Cancer papers, which have shifted treatment approaches from organ-based to mutation-targeted methods. Early detection, as seen in South Korea's thyroid cancer screening program, has not always resulted in improved mortality outcomes due to overdiagnosis.

AI's current role in drug development includes identifying drug targets, predicting toxicity, and designing clinical trials. Yet, a significant 97% of treatments that work in cells and mice fail in human trials. Stephen Wolfram's concept of computational irreducibility is relevant here, suggesting that complex biological systems cannot be fully understood from first principles alone.

The pursuit of artificial superintelligence is diverting resources from biotech and cancer research, with biotech investment at a decade low. Javorsky calls for restructuring incentives and institutions to improve oncology and biomedical research, emphasizing the importance of creating large-scale data sets supported by bodies like the NIH and philanthropy.

Regulatory reforms are needed to integrate new measurement modalities into clinical care and make them affordable. Outcomes-linked reimbursement models could incentivize patient health improvements, while mandating electronic data capture for federally funded research could enhance data availability.

Javorsky points to China's progress in biotechnology, attributing it to an AI tools-first approach. Unlike the US, where chronic diseases like cancer receive fewer resources, China's strategy shows the potential benefits of prioritizing AI tools in advancing healthcare.

Key Insights

• AI's potential for curing cancer is limited by data and coordination issues, not intelligence. Superintelligence cannot model cancer without accessible, standardized data.
• Medical knowledge is growing rapidly, with a doubling rate of 73 days, yet cancer mortality rates have not significantly improved. The FDA's stringent approval process and business issues prevent many drugs from reaching the market.
• Biology's complexity and lack of first principles make it challenging for AI to achieve breakthroughs. Most medical literature is not reliable, and electronic medical records introduce biases not representative of clinical reality.
• AI helps in drug development but struggles with the transition from preclinical to human trials, as evidenced by a 97% failure rate. Restructuring research incentives could improve oncology and biomedical research outcomes.