Mark Zuckerberg & Priscilla Chan: How AI Will Help Cure Disease

Mark Zuckerberg and Priscilla Chan explain their mission to help cure and prevent disease by the end of the century through the Chan Zuckerberg Initiative (CZI). Rather than funding individual treatments, they focus on building foundational tools and infrastructure that the entire scientific community can use. Priscilla, a trained pediatrician, emphasizes how her clinical experience revealed the importance of basic science—encountering patients with undiagnosed conditions made clear that fundamental understanding precedes effective treatment.

Their core theory is that major scientific breakthroughs are typically preceded by new tools to observe phenomena differently, citing examples like the microscope and telescope. They identified a funding gap: while most science funding comes through NIH grants for near-term individual lab work, longer-term tool development (costing $100 million to $1 billion over 10-15 years) lacks adequate support. This is where CZI intervenes.

They structure grand challenges around 10-15 year time horizons—long enough to tackle meaningful problems but short enough to maintain team cohesion and show progress. Their three Biohubs focus on different complementary challenges: New York on cell engineering, Chicago on tissue and cell-cell communication, and San Francisco on deep imaging and transcriptomics. They deliberately locate these near major research universities to leverage existing academic strength.

A major project is the Cell Atlas, which began as a practical solution to an annotation bottleneck in single-cell research. They created CellxGene as an annotation tool, and as researchers adopted it for standardized data formats, a network effect emerged where 75% of the atlas data came from the community rather than CZI-funded work alone. This created a shared resource of millions of standardized cells accessible to the entire scientific community.

Virtual cell models represent their most ambitious current direction—building computational simulations that combine cellular data with AI to create models of how cells behave. They're developing specialized models: VariantFormer for predicting CRISPR edits, diffusion models for generating synthetic cell types, cryo-EM spatial models, and reasoning models for causal biological inference. These build hierarchically from protein interactions upward to cellular and eventually systemic models.

They argue virtual cell models will enable riskier scientific exploration because researchers can test hypotheses computationally before expensive wet lab work, democratizing access to expensive experimentation. The models don't need to be 100% accurate—directional signal is valuable for de-risking ideas.

A key organizational decision is consolidating their previously decentralized AI and biology work under unified leadership (Alex Reeves from Evolutionary Scale), creating a "flywheel" where AI models can identify gaps that drive new data collection, and new data improves models. They're restructuring CZI so the Biohub becomes their primary philanthropic focus.

They emphasize the importance of bringing biologists and engineers together physically in shared spaces, a simple principle that has proven powerful because it wasn't the default in academia. They're also expanding computational resources (planning 10,000-GPU clusters) and inviting external scientists to collaborate on large-scale questions requiring significant compute.

Reflecting on 10 years of work, Mark notes that their biggest validation is seeing what they initially set out to do actually deliver more results than expected, giving them confidence to double down. They describe themselves as both patient with long time horizons but impatient about execution, maintaining ambiguity while pursuing big goals. They argue precision medicine requires understanding individual biology rather than treating diseases as monolithic categories, which their tools enable.