Overview of #485 — The New Science of Cancer
Sam Harris speaks with oncologist and cancer biologist Siddhartha Mukherjee about the updated edition of The Emperor of All Maladies and how cancer science has evolved over the last 15 years. The conversation is organized around three major themes — prevention, detection, and treatment/cure — with a strong emphasis on the growing role of AI in cancer research and medicine. Mukherjee argues that cancer is best understood as many genetically distinct diseases with shared biological features, and that major progress is now coming from more precise prevention strategies, better risk-stratified screening, and increasingly effective targeted and immune-based therapies.
Cancer as a Disease: One Disease or Many?
Mukherjee’s core framing is that cancer is:
- Not one disease, but not entirely separate diseases either
- Genetically unique in each individual tumor specimen
- Unified by shared behaviors, especially:
- cells that keep dividing
- cells that often resist death
- cells that hijack normal survival and movement pathways
Main conceptual takeaway
A future “cure for cancer” will almost certainly involve many different cures, not a single universal one, though there will be common principles that run across them.
Prevention: The Most Difficult and Most Underappreciated Front
Mukherjee argues that prevention should not be an afterthought, but it has historically been harder than treatment because:
- Prevention studies must show that something doesn’t happen
- They often require long timelines
- Cancer lacks a perfect equivalent to cholesterol or blood pressure as a surrogate marker
How cancer prevention is studied
He outlines three main approaches:
- Ames test: detects mutagens that can cause DNA mutations
- Animal studies: expose animals to suspected carcinogens
- Epidemiology: compare cancer rates in exposed vs. unexposed populations
Cell phones and brain cancer
Mukherjee dismisses the old cell phone-brain cancer scare:
- Brain tumor mortality has remained flat over decades
- Cell phone radiation is physically different from x-ray radiation
- The rise in smartphone use has not produced a matching rise in brain cancer deaths
A major new idea: “Inflamogens”
Mukherjee says the field is now recognizing a new class of carcinogens that:
- do not primarily cause mutations
- instead alter the tissue environment
- create chronic inflammation
- help dormant cancers begin growing
Examples discussed:
- Particulate air pollution
- possibly asbestos
He calls these substances “inflamogens.”
Why this matters
This opens up new possibilities for prevention:
- developing tests for harmful chronic inflammation
- identifying people at risk before cancer appears
- removing environmental triggers that promote tumor growth
Chemoprevention
He says we are closer than many people think to effective preventive drugs, but with an important exception already in use:
- Tamoxifen and other anti-estrogen drugs can prevent some ER-positive breast cancers in high-risk people
HPV vaccine
Mukherjee is especially emphatic about HPV prevention:
- HPV causes certain cancers, especially cervical cancer
- The HPV vaccine is highly effective
- He strongly recommends vaccination for young boys and girls
- He cites a Swedish randomized study showing cervical cancer risk can be driven to zero with appropriate vaccination in the right age group
Risk, Genetics, and the Psychology of “Previvors”
Mukherjee discusses how people should interpret elevated cancer risk from:
- family history
- BRCA1/BRCA2 and other inherited mutations
- polygenic risk scores
- viral infections such as HPV
His practical framework
He recommends thinking in a “grayscale” of risk:
-
Very high inherited risk
- e.g. BRCA1/BRCA2, p53 mutations
- see a genetic counselor
- consider enhanced screening and prevention trials
-
Moderately elevated polygenic risk
- risk estimates are still evolving
- counsel without panic
-
Low apparent risk but strong exposure history
- e.g. asbestos exposure
- may warrant deeper evaluation
-
Virus-related cancer risk
- especially HPV
- should be managed with medical follow-up and vaccination
Psychological point
He warns against becoming a “previvor” — someone consumed by the fear of getting cancer before they have it.
Detection: Promise, But Bayes Matters
A major segment focuses on liquid biopsies and cell-free DNA blood tests.
His caution
Mukherjee says the common public experience with these tests is often not atypical, but typical:
- false positives create anxiety
- positive results can lead to unnecessary scans and biopsies
- the math is dominated by Bayesian prior probability
Key principle: Bayes’ theorem
His central argument:
- A test’s usefulness depends not just on sensitivity and specificity
- It depends on the base rate of cancer in the population being tested
- If cancer is rare in the tested population, even a good test can yield many false positives
Who liquid biopsy is most useful for
He says these tests are likely most valuable in higher-risk groups, such as:
- people with strong inherited risk
- people with elevated polygenic risk
- people with prior cancer, where recurrence is more plausible
Whole-body MRI
Mukherjee is skeptical of routine whole-body MRI screening:
- the false-positive burden is high
- it can trigger invasive and risky follow-up
- the key question is not whether a scan finds “something,” but whether it reduces mortality
Lead-time bias
He emphasizes that early detection can falsely appear beneficial if one measures survival from the date of diagnosis rather than actual mortality reduction.
Minimal Residual Disease: A More Promising Use of Detection
Mukherjee is much more optimistic about detecting minimal residual disease (MRD) after treatment.
Why MRD matters
For patients in remission, MRD can help identify:
- tiny amounts of remaining cancer
- recurrence earlier than conventional scans
- when to restart treatment in a targeted way
Best use case
He highlights multiple myeloma as a major example where MRD-guided monitoring has already been valuable.
Important distinction
MRD should be seen more as a tool for intervention than as an alarm bell.
Treatment: Major Progress Has Already Happened
Mukherjee argues that cancer mortality is falling overall, and some cancers have been transformed from lethal diseases into chronic or sometimes curable ones.
Areas of dramatic progress
He highlights several examples:
- Non-small cell lung cancer
- some patients now survive many years with immunotherapy
- Bladder cancer
- strong responses to immune-based treatments
- Breast cancer
- some advanced cases now yield many years of good-quality life
- Multiple myeloma
- survival has improved steadily over decades
- Childhood ALL
- CAR-T and other cellular therapies have rescued many relapsed/refractory cases
Immunotherapy
These drugs work by:
- removing the “cloak” cancer uses to hide from the immune system
- helping immune cells recognize and attack tumors
CAR-T therapy
He notes CAR-T has been highly successful in blood cancers but remains difficult in solid tumors.
Why solid tumors are harder
- They have a hostile microenvironment
- The tumor “soil” blocks CAR-T cells from penetrating effectively
- Researchers are now trying combination approaches to overcome this barrier
A new RAS inhibitor
Mukherjee cites a recent pancreatic cancer advance:
- a RAS-targeting drug showed improved survival in a trial
- median survival improved from about 6 months to 13 months
- he describes this as a first foothold, not the final answer
Drug Prices, Generics, and Manufacturing Quality
Mukherjee addresses the high cost of cancer drugs and the promise of generics.
Why generics matter
- several major cancer drugs are going off patent
- generics should reduce costs substantially
- he supports protecting patents for a limited period, but not extending them indefinitely
Quality-control concerns
He acknowledges real problems in generic drug manufacturing, especially historically, but says they can be addressed through:
- continuous audits
- independent testing
- verification that active ingredients match
AI in Cancer: Real Utility, Not Just Hype
Mukherjee is broadly optimistic about AI, but says it is not one AI — it is many specialized systems.
Where AI can help now
1. Prevention
AI can combine:
- genetics
- exposures (“exposome”)
- microbiome
- behaviors
- diet
- other multidimensional features
to find hidden risk patterns that humans struggle to see.
2. Detection and diagnosis
AI is already useful in:
- mammography
- lung cancer screening
- skin lesion analysis
- pathology image review
He frames AI as a second reader or companion diagnostician.
3. Drug discovery
This is the focus of his company, Manus AI, which he co-founded with Ujwal Singh and Reid Hoffman.
He says AI drug discovery requires:
- teaching AI the rules of medicinal chemistry
- giving it enough structure to design viable molecules
- using AI for target discovery as well as molecule generation
4. Clinical trials
AI can help:
- identify eligible patients in electronic medical records
- optimize trial design
- support adaptive trials that learn and re-balance over time
The Larger Outlook: Will Cancer Be “Behind Us”?
Mukherjee is cautiously optimistic.
Why he’s optimistic
In his lifetime, many cancers have shifted from:
- incurable acute disease
- to manageable chronic disease
- and in some cases to curable disease
Why cancer will not disappear completely
He says cancer will likely always exist because:
- it is tied to aging
- some cases are due to plain bad luck
- environmental risks will never be eliminated perfectly
Still, the direction is clear
He expects:
- more cancers will become treatable
- fewer will remain fully out of reach
- prevention, early detection, and immunotherapy will continue to improve outcomes
Science and Politics: Damage to U.S. Medical Research
The conversation ends with a critical discussion of U.S. science policy under the Trump administration.
Mukherjee’s concerns
He says the administration has been broadly anti-science, citing:
- funding cuts to the CDC and other agencies
- chaos in academic and drug-development ecosystems
- unnecessary scrutiny in some places and insufficient scrutiny in others
- damage that will take years to reverse
Broader institutional warning
He emphasizes that:
- destroying institutions is easy
- rebuilding them takes years
- scientific trust must be restored
Supply chains and offshoring
Mukherjee warns that the U.S. has become dangerously dependent on foreign manufacturing for essential medicines and supplies.
He gives the example of intravenous saline, noting that a supply disruption could severely impact hospitals and surgery.
Key Takeaways
- Cancer is many diseases, but with shared biological logic
- Prevention is harder than treatment, but new work on inflammation-based carcinogenesis is promising
- HPV vaccination is one of the clearest success stories in cancer prevention
- Liquid biopsies and whole-body scans are overhyped for average-risk screening
- Bayesian reasoning is essential for interpreting screening tests
- MRD is one of the most promising detection tools in relapse-prone cancers
- Immunotherapy, CAR-T, and targeted drugs have already transformed outcomes for some cancers
- AI is likely to be genuinely useful in prevention, pathology, drug discovery, and trial design
- U.S. science and manufacturing infrastructure need repair to sustain progress
Notable Insights
“Cancer is not one disease, but each specimen is its own disease in the genetic sense.”
“Prevention is the most difficult science, because you’re trying to make something not happen.”
“Bayes is the answer.”
“Minimal residual disease is a tool, not an alarm.”
“AI in medicine will be companion mode.”