Silicon Valley GirlHow AI Is Breaking the Rules of Biology | Dr. Priscilla Chan, Chan Zuckerberg Initiative
CHAPTERS
- 0:00 – 1:09
Science will look fundamentally different soon: the promise of AI-driven biology
A fast-paced opening sets the stakes: Priscilla Chan believes biology and biomedical science are on the verge of a step-change powered by AI. Marina frames the central patient question—what these breakthroughs will actually mean for real people in the near future.
- •Chan’s claim that science will be “fundamentally different” within ~5 years
- •The patient-centered framing: impact matters more than tools alone
- •Positioning AI + data as the catalyst for the next era of biology
- 1:09 – 2:34
Why Priscilla’s miscarriage openness mattered: isolation, stigma, and shared experience
Marina shares her own miscarriage journey and explains how Chan’s public honesty helped her feel less alone. Chan echoes how common miscarriages are but how rarely they’re discussed, and they reflect on the emotional weight and social stigma.
- •Marina recounts years-long fertility struggles and the impact of public sharing
- •Miscarriage is common but under-discussed; patients often feel isolated
- •Harmful clinician messaging (“it was your fault”) and emotional consequences
- •Public narratives can create community and resilience
- 2:34 – 3:24
The turning point: becoming parents and launching CZI to build a healthier future
Chan explains how having their first child made “the future” feel immediate and personal, motivating a large-scale commitment to long-term societal impact. She describes CZI as their way to contribute to a world where kids can be healthy and thrive.
- •Parenthood as the catalyst for committing resources and focus
- •CZI framed as building a better world for the next generation
- •Moving beyond “nesting” toward systems-level investments in health and opportunity
- 3:24 – 4:40
Mission and strategy: curing or preventing all disease by accelerating every scientist
Chan lays out CZI/Biohub’s ambitious mission—cure or prevent all disease—and argues it’s now plausible sooner due to AI advances. The core strategy is enabling the broader scientific community with tools, datasets, and the freedom to take bold risks, rather than trying to solve everything in-house.
- •Mission: cure or prevent all disease; timeline pulled forward by AI progress
- •Key lever: make scientists faster, more efficient, and more risk-tolerant
- •Tool-building and ecosystem enablement over single-organization heroics
- •Biohubs combine frontier biology with frontier AI for direct human impact
- 4:40 – 6:56
From skepticism to momentum: how reactions changed from 10 years ago to the LLM era
Chan recounts early skepticism—people called the mission unrealistic—and how that pushback clarified what was missing: better tools, datasets, lab techniques, and multidisciplinary teams. With large datasets in place, LLMs made a pathway to extracting meaningful biological knowledge feel tangible.
- •Early reactions: “You’re nuts”—used as a prompt to identify concrete blockers
- •Identified needs: datasets, lab methods, new techniques, cross-discipline collaboration
- •Growth of Biohubs and major single-cell datasets before the LLM inflection
- •LLMs as the interpretive engine for biology-scale data
- 6:56 – 9:22
Clinic-to-lab motivation: the UCSF pediatric cases that exposed medicine’s limits
Chan explains the clinical experiences that shifted her from frontline pediatrics toward investing in basic science. Seeing children with unnamed, untreatable conditions—and research that couldn’t translate into care—made clear that scientific progress is the source of future hope.
- •UCSF as a referral center for the hardest pediatric cases
- •Confronting diseases with no diagnosis, name, or treatment
- •Research PDFs that couldn’t be operationalized into clinical decisions
- •Conclusion: advancing basic biology is essential to creating medical options
- 9:22 – 11:17
Why CZI builds tools (not single cures): mapping cell types to understand what breaks
Chan describes CZI’s non-disease-specific approach: build platforms that make all scientists more effective. She explains cell-by-gene mapping—how the same DNA yields different cell types—and why understanding healthy vs. broken cellular states enables precise interventions.
- •Tool-and-platform orientation rather than disease-by-disease funding
- •Cell-by-gene maps to distinguish cell types and states across the body
- •Same genome, different outcomes: regulation drives cell identity and function
- •Understanding cellular errors enables targeted, corrective therapies
- 11:17 – 12:22
Virtual cells explained: human-relevant experimentation that’s faster and cheaper
The conversation turns to “virtual cell” models—computer simulations of how cells behave—so experiments can be run on human-relevant systems without relying as heavily on animal models. Marina underscores the potential: faster drug discovery, digital twins, and programmable biology.
- •Virtual cells as in-silico models of real cellular behavior
- •Benefits: cheaper, faster experimentation with more direct clinical relevance
- •Reducing translation gaps from flies/mice to humans
- •Potential outcomes: rapid drug discovery, personalized medicine, digital twins
- 12:22 – 14:37
What patients feel first: personalized medicine and rethinking “common diseases”
Chan connects virtual-cell capabilities to near-term patient impact: moving beyond averages toward individualized predictions of drug response and disease pathways. She argues that categories like hypertension or depression likely contain many sub-diseases that require different treatments.
- •Medicine today is built on population averages; individuals aren’t average
- •Goal: predict responses based on genetics and personal biology
- •Reduce trial-and-error prescribing and the suffering it causes
- •Reframe common diagnoses as heterogeneous clusters of distinct mechanisms
- 14:37 – 17:11
First breakthroughs: immune system as a programmable lever (autoimmune, neuro, cardio)
Asked which diseases might be cured first, Chan focuses on the immune system: a built-in system that protects health but can also cause disease when imbalanced. She describes both understanding immune “levers” and potentially engineering immune cells for tasks like preventing autoimmune attacks or even clearing arterial plaques.
- •Immune balance as central: underactive vs. overactive states drive disease
- •Autoimmune disease as a high-promise area if imbalance can be mapped and corrected
- •Engineering immune cells to patrol organs and perform targeted actions
- •Examples discussed: multiple sclerosis/neurodegeneration; atherosclerotic plaque cleanup
- 17:11 – 19:49
Scaling the cell atlas: from 100M to 1B cells, and the multi-omics road ahead
Chan explains that mapping is accelerating dramatically—years to reach 100 million cells, months to reach a billion—driven by faster lab hardware and clearer goals. But transcriptomics is only one dimension; true understanding requires proteins, spatial organization, and live-cell behavior, enabled by tight AI–wet-lab feedback loops.
- •Acceleration in cell mapping speed (100M in years → 1B in months)
- •Drivers: improved hardware tools plus focus and coordination
- •Beyond RNA: proteins, spatial maps, live-cell dynamics, context dependence
- •AI + wet lab “flywheel”: models identify blind spots; experiments fill them; tools remove bottlenecks
- 19:49 – 21:46
Virtual immune system and wearable sensors: measuring how immune cells ‘talk’
Chan introduces the “virtual immune system” as modeling not just single cells but immune networks communicating across the body. She highlights a tiny sensor (analogous to a glucose monitor) that can read immune-cell communication signals in living organisms, providing the data foundation for virtual modeling and intervention testing.
- •Virtual immune system models multi-cell communication and coordination
- •Immune system has no single organ—distributed signaling is key
- •Wearable/implantable-style sensor captures dynamic immune signals in vivo
- •Data enables virtual manipulation of parameters to study many diseases
- 21:46 – 23:34
A 2040 care model: early-warning patches, flare prevention, and custom drugs
In a speculative but grounded scenario, Chan imagines continuous monitoring for autoimmune risk (e.g., lupus) to detect molecular shifts before symptoms appear. Virtual-cell models could then help identify misbehaving proteins and support designing personalized therapies to prevent flares rather than reacting to damage.
- •Proactive monitoring to detect the earliest disease signals
- •Autoimmune example: lupus risk + molecular trigger tracking
- •Prevent flares before organ damage and severe symptoms
- •Use virtual models to pinpoint mechanisms and design individualized drugs
- 23:34 – 25:58
What worries her: speed, removing barriers, and the future physician’s role with AI
Chan says what keeps her up is moving quickly—her job is removing obstacles so scientists can work efficiently. She advises future scientists/doctors to enter the field now, and describes physicians’ evolving role: pairing deep patient insight with science, and learning to ask the right questions of AI while continuing to serve as healers.
- •Primary concern: urgency—progress depends on working fast
- •Her role: identify and eliminate barriers to scientific execution
- •Advice: this is the most exciting time to go into science/medicine
- •Physicians as translators: patient experience + deep science is “magic”
- •Clinical AI examples (skin/retina): doctors must ask the right questions and guide care
- 25:58 – 27:54
When the mission becomes personal: reproductive biology, rare disease communities, and hope
Chan shares deeply personal intersections with the work, including studying single-cell expression in female reproductive organs and the fact that labor’s trigger cascade is still not fully understood. She also describes “Rare As One,” empowering patient groups to participate in research—sometimes delivering something as meaningful as a diagnosis and a name.
- •Reproductive biology research motivated by pregnancy experience
- •Major unknowns remain (e.g., what triggers labor) despite modern medicine
- •Rare As One: training/resources for rare disease groups to engage in research
- •Impact isn’t only cures—diagnosis/naming a condition reduces isolation and powerlessness
- 27:54 – 30:33
Cures within our lifetime, mission fulfillment, and balancing work with motherhood
Chan argues many cures are plausible within our lifetime, citing rapid progress over the last decade and examples where genetic understanding enables correction. She doesn’t define a single “mission fulfilled” endpoint, and closes with practical advice on parenting while building: strict time boundaries and disciplined scheduling.
- •Optimism grounded in recent history: huge progress in just 10 years
- •Most ‘ripe’ targets: diseases with clear genetic/molecular mechanisms
- •Need more diseases mapped to mechanism level + support for bold, risky science
- •No fixed finish line; there’s always more important work
- •Work–family balance via disciplined scheduling and protected kid time
