Mark Zuckerberg & Priscilla Chan: How AI Will Cure All Disease: Mark Zuckerberg, Priscilla Chan, and a16z

⬅️ Back to Podcasts

PODCAST INFORMATION

  • Title: 🎙️ Podcast: Mark Zuckerberg & Priscilla Chan
  • Podcast/Series: a16z Podcast
  • Episode: Mark Zuckerberg & Priscilla Chan: How AI Will Cure All Disease
  • Hosts: Ben Horowitz, Erik Torenberg, Vineeta Agarwala
  • Guests: Mark Zuckerberg (CEO, Meta; Co-founder, CZI) and Priscilla Chan (Co-founder, CZI; Pediatrician)
  • Duration: Approximately 44 minutes

📓 Podcast Episode on Apple Podcasts

🎧 Listen here

📺 Watch here

🎣 HOOK

Biology still doesn’t have its equivalent of the periodic table. While AI companies chase artificial general intelligence, one of the world’s most ambitious philanthropic organizations is building something arguably more consequential: a complete computational simulation of human cells that could let scientists cure disease in silicon before ever touching a pipette. Mark Zuckerberg and Priscilla Chan aren’t just funding research; they’re rewiring how science itself gets done, and the implications could reshape medicine within our lifetime.

💡ONE-SENTENCE TAKEAWAY

The path to curing all disease isn’t through incremental grants but through building AI-powered virtual cell models and shared scientific infrastructure that democratizes hypothesis-testing, enabling scientists to take risks previously too expensive for wet labs.

📝 SUMMARY

Mark Zuckerberg and Priscilla Chan, co-founders of the Chan Zuckerberg Initiative (CZI), sit down with a16z partners to explain their audacious mission: accelerating the cure, prevention, and management of all disease by century’s end. Priscilla, a trained pediatrician, describes her frustration treating children with mysterious conditions; variants of unknown significance printed on PDFs that offered no real guidance. Mark frames the challenge through a tools-first lens: scientific breakthroughs historically follow new observational tools (microscopes, telescopes), and biology in 2025 is “coding without a debugger.”

The core problem they identified is a funding gap. NIH grants, while essential, are parcelled into relatively small awards for near-term, individual investigator projects. The tools biology needs (imaging systems, AI models, massive datasets) require $100 million to $1 billion investments over 10-15 years. This creates a niche where philanthropy can be most additive: funding shared infrastructure that no single lab could justify but that elevates the entire field.

Their solution centers on the CZI Biohub network: three sites (San Francisco, Chicago, New York) each tackling frontier biology paired with frontier AI. The San Francisco hub focuses on deep imaging and transcriptomics; Chicago on tissue cell communication; New York on cellular engineering. The organizational innovation is physical: forcing biologists and engineers to sit side-by-side, creating cross-disciplinary collisions impossible in traditional academic silos.

The centerpiece is the Cell Atlas and ultimately biology’s missing periodic table. What began as an annotation tool (CellxGene) to solve a workflow bottleneck has become a community-driven resource with millions of cells catalogued in standardized, open-source format. Only 25% was CZI-funded; 75% came from the scientific community recognizing its value. This data now feeds virtual cell models; hierarchical AI simulations from proteins to whole cells to immune systems. These models let scientists test high-risk hypotheses computationally before expensive wet lab validation.

The announcement in this episode is twofold: unifying CZI’s decentralized efforts under Alex Rives (head of EvolutionaryScale, formerly Meta’s protein folding team) as AI-driven leadership, and making the Biohub the primary focus of their philanthropy going forward. Zuckerberg believes AI advances make the “cure all disease” goal achievable significantly sooner than 2100.

The 10-15 Year Horizon Strategy

CZI deliberately targets challenges where “you see a path” but everything isn’t solved. This timeframe (similar to venture-backed companies) matches how long teams can cohere while maintaining risk appetite. Their model identifies gaps in the scientific portfolio: projects too large for NIH grants but too infrastructure-focused for biotech. By funding data generation, tool-building, and model training as integrated loops, they create flywheels where AI reveals blind spots, which shape new data collection, which improves models.

Democratizing Discovery Through Interface Design

A critical insight is that lowering barriers to entry accelerates breakthroughs. CellxGene’s interface was intentionally designed for non-computational biologists, enabling immunologists to explore neurodegeneration data and spot connections. The goal is making complex biological data queryable by domain experts without requiring them to be AI specialists; creating network effects where specialists from different fields can “come for the annotation, stay for the virtual cell model.”

🧠 INSIGHTS

Core Insights

  • The Tools-First Theory of Scientific Breakthroughs: Most major scientific advances are preceded by new observational tools (microscopes, telescopes). Biology currently lacks the equivalent computational tools, creating a ceiling on discovery speed. CZI’s strategy is to build these $100M+ infrastructure projects that governments won’t fund and companies can’t monetize.

  • The Grant Funding Risk Paradox: Traditional funding forces scientists toward low-risk projects to maintain careers (tenure, publishing). Virtual cell models invert this by making high-risk hypothesis testing computationally cheap, potentially unlocking breakthrough ideas that would never survive peer review for wet lab funding.

  • Network Effects in Scientific Data: The Cell Atlas succeeded because standardization (CellxGene annotation tool) created a data format network effect. Scientists adopted it for their own benefit, then contributed data back, building a commons where 75% of content is community-generated; proving open-source infrastructure can work in biology.

  • AI-Biology Integration Requires AI-Native Leadership: Appointing Alex Rives (AI researcher who understands biology) rather than a biologist who understands AI signals a fundamental shift. The bottleneck isn’t domain expertise but fluency in both; requiring organizational structures where AI shapes data collection, not just analyzes it.

  • The Lab Space Paradox: Modern biology labs are expanding compute, not square footage. GPU clusters (1,000→10,000) are the new essential lab infrastructure, more expensive than wet lab space and enabling questions impossible with traditional setups; reshaping what a “biology lab” looks like.

How This Connects to Broader Trends/Topics

  • The End of Big Science Funding Gaps: Governments excel at basic research (NIH) and private markets at applied development (biotech), but $100M+ “pre-competitive” infrastructure falls through cracks. This reveals a structural market failure where philanthropy can be more than additive; it can be catalytic, creating new categories of scientific institutions.

  • AI’s Over/Underestimation Loop: Zuckerberg notes AI is “simultaneously the most overestimated and underestimated technology.” While AGI hype dominates headlines, domain-specific AI applied to biology may have more immediate, tangible impact; echoing how early internet infrastructure seemed boring until applications emerged.

  • The Interface Layer as moat: In an era where models and data become commoditized, user interfaces designed for specific scientific workflows create defensible value. The insight that “a scientist isn’t going to talk to it like ChatGPT” suggests the winning AI tools won’t be general-purpose but will speak the language and respect the mental models of each domain.

🏗️ FRAMEWORKS & MODELS

The Biohub Master Plan

CZI’s unifying framework for structuring scientific philanthropy around AI-enabled biology.

  1. Frontier Biology + Frontier AI: Pair cutting-edge biological data generation with state-of-the-art AI capabilities in the same organization, preventing the “two cultures” problem where biology and AI researchers work in isolation.

  2. Grand Challenge Selection: Target 10-15 year projects that are ambitious enough to matter but credible enough to see a path forward, filling the gap between NIH grants and century-long moonshots.

  3. Physical Collocation: Biologists and engineers must work shoulder-to-shoulder, not just collaborate across institutions. Communication bandwidth matters more than organizational charts.

  4. Open-Source Infrastructure: Build tools (CellxGene, Cell Atlas) that become standards for the community, creating network effects where usage drives contribution and accelerates the entire field.

The Virtual Cell Hierarchy

A multi-scale modeling approach that builds biology from the bottom up.

  1. Protein Foundation Layer: Start with state-of-the-art protein models (EvolutionaryScale) to understand sub-cellular components accurately.

  2. Cellular Simulation Layer: Integrate single-cell transcriptomics, spatial imaging (cryoEM), and cellular engineering data to model whole cell behaviors.

  3. Systems Biology Layer: Combine into virtual immune systems, tissue models, and eventually organism-level simulations for specific questions.

  4. Feedback Loop: Use model predictions to identify data gaps, then design targeted experiments to fill them; treating the model as a hypothesis engine, not just an analysis tool.

The “Periodic Table for Biology” Mental Model

CZI’s inspiration from physics: biology needs foundational, standardized classification systems before it can become a predictive science.

  • Standardization as Enabler: Just as the periodic table enabled chemistry, a complete Cell Atlas (every cell type, every state) becomes the reference framework for all biological questions.
  • Static Reference + Dynamic Models: The atlas provides the “what” (taxonomy), while virtual cells provide the “how” (mechanism), together creating a complete computational biology stack.
  • Community Ownership: Only 25% CZI-funded; the rest emerges from scientific community adoption, ensuring long-term sustainability beyond philanthropic timelines.

💬 QUOTES

  1. “When we first set out that the goal to cure and prevent disease by the end of the century, people like honestly most scientists couldn’t look at us with a straight face. And it was true because if you just decided to spend the money funding the next best grant for every single lab in the country, like you there was no pathway to that being true.” - Priscilla Chan, on why traditional funding models couldn’t achieve their mission

  2. “If you look at the history of science, most major breakthroughs are basically preceded by the invention of a new tool to observe phenomena in a new way. It’s without those kind of tools, it’s kind of like you’re coding without being able to step through the code and debug things.” - Mark Zuckerberg, articulating the tools-first theory of scientific progress

  3. “The biology folks looked at it as if it were crazy ambitious. And then the AI folks are like, well, that’s kind of boring. That’s just automatically going to happen. It’s like, okay, there’s something in between there that needs to be bridged.” - Mark Zuckerberg, on the AI-biology perception gap

  4. “Most diseases should be thought of as rare diseases because each one of our biology is different and right now we just get lumped… truly each one of our biology is different.” - Priscilla Chan, redefining disease classification for precision medicine

  5. “All models are wrong. Some are useful. This is hopefully has utility on certain axes.” - Mark Zuckerberg, on the pragmatism of virtual cell accuracy

  6. “We’re not expanding like a lot of square footage per se, but we’re expanding our compute. It’s just like in a sense that’s new lab space. It’s much more expensive than wet lab space.” - Priscilla Chan, on how AI is reshaping research infrastructure

  7. “The user interface was intentionally designed to not need to have a computational or really a very deep biological background to be able to use because you want people coming from different fields to look at the problem.” - Priscilla Chan, on democratizing scientific tools

  8. “The conversation is what we keep finding out ends up being very, very important… a scientist isn’t going to talk to it like you know I talked to chat at GPT or whatever.” - Mark Zuckerberg, on domain-specific AI interfaces

  9. “Being willing to have a long time horizon, but be impatient at the same time… it’s all those iterations along the way that have sort of allowed us to get to this place where you know to get lucky ready having built data sets to take advantage of AI.” - Priscilla Chan, on navigating ambiguity in philanthropic science

  10. “There is a space that and that there’s just going to be a huge amount of leverage with AI and it is um yeah it’s it still seems like there could be a lot more effort in the space around building tools and just accelerate the whole thing a lot better.” - Mark Zuckerberg, on the underinvestment in scientific infrastructure

🎯 HABITS

Scientific Leadership Habits

  • Impatient Patience: Tolerate decade-long ambiguity while pushing rapid iterations. Priscilla notes they “craved feedback” early on but learned to value long-term signals over short-term validation.

  • Tool-First Framing: Always ask “what’s the missing tool?” before funding research. This redirects energy from individual experiments to infrastructure that multiplies impact across the entire field.

  • Cross-Disciplinary Physical Integration: Literally seat biologists next to engineers, refusing to let organizational silos mirror disciplinary ones. The “attack the communication bandwidth problem” approach.

Product Development Habits

  • Solve Your Own Bottleneck: CellxGene was built because annotation was slowing down the Cell Atlas project. The best tools emerge from friction in your own workflow, ensuring product-market fit.

  • Standardize for Network Effects: Intentionally design data formats and interfaces to become community standards, not just internal tools. The 25%/75% CZI/community funding split becomes a metric of success.

  • Interface as Moat: Build UIs that lower barrier to entry for non-experts, creating viral adoption loops where immunologists can query neurodegeneration data without computational expertise.

Personal Habits

  • Founder-Market Fit Verification: Continuously ask “if we didn’t exist, would this be a problem?” to ensure unique value proposition. This prevents philanthropic drift into crowded spaces.

  • Seek Underestimated Domains: Deliberately target areas where AI is simultaneously overhyped (AGI) and underhyped (domain-specific applications), finding leverage in the gap.

  • Measure Through Usage, Not Credit: Focus on tool adoption and community growth rather than publication authorship, “that’s why it’s philanthropy” but with startup-style customer feedback loops.

📚 REFERENCES

  • Chan Zuckerberg Initiative (CZI): Philanthropic organization founded in 2015 with mission to cure, prevent, and manage all disease by 2100.
  • CZI Biohub Network: Three research centers (San Francisco, Chicago, New York) focused on frontier biology + frontier AI integration.
  • Cell Atlas / CellxGene: Open-source, community-driven database with millions of single-cell profiles; biology’s “periodic table” with standardized annotation tools.
  • NIH Grant System: Traditional U.S. government funding mechanism that parcels science into small, near-term investigator grants, leaving large infrastructure projects unfunded.
  • AlphaFold: DeepMind protein structure prediction model used as example of AI built on public datasets; CZI aims to create datasets specifically for training next-generation models.
  • EvolutionaryScale: AI company (formerly Meta protein folding researchers) joining CZI Biohub; Alex Rives becomes head of science program.
  • Virtual Cell Models: Hierarchical AI simulations from protein to cellular to systems level, enabling in silico hypothesis testing.
  • Cryo-Electron Microscopy (cryoEM): Imaging technology at near-atomic resolution used to generate spatial cell data for model training.
  • CRISPR Gene Editing: Used in CZI’s “Variantformer” model to predict cellular outcomes of genetic edits.
  • Idiopathic Pulmonary Fibrosis (IPF): Example disease mentioned where CZI’s tools helped a startup identify drug targets in previously “idiopathic” condition.

✅ QUALITY & TRUSTWORTHINESS NOTES

  • Empirical Scale: Cell Atlas contains millions of cells from thousands of experiments; only 25% CZI-funded, demonstrating genuine community adoption and data quality.
  • Peer Review & Publication: Work published in top-tier journals; models like Variantformer and diffusion-based cell simulators are being released to scientific community for validation.
  • Cross-Institutional Validation: Biohubs partner with premier universities (UCSF, Stanford, Berkeley, etc.) ensuring research undergoes academic scrutiny and complementary expertise.
  • Specific Metrics: CZI tracks tool usage rates, community contribution percentages (75% of Cell Atlas), and plans GPU clusters scaling from 1,000→10,000 units; concrete infrastructure commitments.
  • Transparency on Limitations: Zuckerberg explicitly states “All models are wrong. Some are useful,” acknowledging that virtual cells don’t need 100% accuracy to provide directional signals and derisk experiments.
  • Founder Credibility: Chan is UCSF-trained pediatrician who experienced clinical limitations firsthand; Zuckerberg brings Meta’s AI/scale expertise. Both have demonstrated decade-long commitment and doubling down based on results.
  • Unique Positioning: No other organization simultaneously does frontier AI and frontier biology while maintaining open-source infrastructure…a genuine gap in scientific ecosystem.

Crepi il lupo! 🐺