Nov 6, 2025, 9:45 AM ETInterview, Fireside Chat

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

Centralization of Operations: The Chan Zuckerberg Initiative (CZI) is transitioning from a decentralized portfolio of biohubs and software teams into a single, unified operating philanthropy known as "The Biohub."
Leadership Change: Alex Reeves, formerly of Meta's AI research teams and now leading Evolutionary Scale, has been appointed head of the science program to lead the unification.
Mission Focus Shift: While CZI continues to support education and community initiatives, the Biohub is now the primary thrust of the organization's philanthropy, representing a "doubling down" on biology following a decade of high returns in that sector.
Rationale for Unification: The move aims to close the feedback loop between AI model training and data generation, allowing scientists and AI engineers to work "shoulder-to-shoulder" rather than in silos to build more accurate virtual cell models.

Core Objective: The goal is to create a "periodic table of elements equivalent for biology" by building standardized, open-source datasets and virtual cell models to accelerate the pace of basic science.
Time Horizon: Projects are structured around 10-to-15-year horizons, balancing the need for long-term, high-risk development with the ability to make incremental progress.
Funding Model: CZI funds the development of large-scale tools (estimated at $100M to $1B over 10–15 years) that are typically too expensive or long-term for standard NIH grants, which focus on near-term, smaller-scale investigations.
Community Adoption: The "Cell by Cell" annotation tool and resulting data atlases achieved a network effect where 75% of the data contributions now come from the broader scientific community rather than CZI funding.

Hierarchical Modeling Approach: The strategy involves building models hierarchically: starting with state-of-the-art protein models, expanding to cellular models (transcriptomics), and eventually integrating into complex systems like a virtual immune system.
New Model Capabilities:
- Reasoning Models: Introduction of early-stage AI models designed to reason causally ("why" things happen) rather than just identifying correlations.
- Simulation Models: Deployment of diffusion models to generate synthetic cell types, allowing for the simulation of rare biological configurations.
- Predictive Editing: Models like "VariantFormer" can predict the outcome of CRISPR edits on cellular states.
Compute Expansion: The Biohub is scaling its infrastructure from a 1,000 GPU cluster to plans for a 10,000 GPU system to enable larger-scale biological simulations that individual academic labs cannot support.

Precision Medicine Definition: The vision shifts away from demographic-based treatment (age, ancestry) toward treating "rare diseases" individually based on unique biological signatures and variant significance.
De-risking Research: Virtual cell models aim to allow researchers to run high-risk, high-reward simulations in silico before investing in expensive and slow wet-lab experiments.
Success Metric: Success is defined by an "explosion" of new waves of precision medicine startups and diagnostics, particularly for diseases currently classified as "idiopathic" (of unknown cause), such as idiopathic pulmonary fibrosis.
Open Source Commitment: All data, visualization tools, and query systems are open-sourced to maximize community utility, with feedback loops driving improvements in user interfaces to lower barriers for non-computational biologists.

Initial Skepticism: The original goal to "cure and prevent all disease by 2100" was initially dismissed by the scientific community as "crazy" and by the AI community as "boring" or inevitable without specific intervention.
Pathway Realization: The team identified that the lack of shared tools and standardized datasets was the primary barrier, leading to the creation of the Biohub strategy.
Validation: Ten years into the mission, the initiative reports higher-than-expected returns, validating the hypothesis that funding the development of foundational tools yields greater leverage than funding individual grants.