Jacob Kimmel and Dwarkesh Patel discuss evolutionary explanations of ageing, epigenetic reprogramming, genetic delivery and models for drug discovery. Seven research notes distinguish laboratory findings, early human evidence and commercial forecasts. The discussion also examines research costs and healthcare spending. Patel discloses that he is a small angel investor in Kimmel’s company, NewLimit.
HostDwarkesh Patel
GuestJacob Kimmel
About these timings. Publisher-supplied paragraph starts from the official transcript; paragraph end times are unknown. Discussion boundaries are navigation ranges, not measured speech endings. Audio duration is rounded up from publisher metadata.
Kimmel offers several evolutionary explanations for limited longevity and suggests that medicine may improve traits evolution did not maximise.
Background and context
Parker and colleagues’ 2020 study compared fruit-fly lines selected for later reproductive ageing with control lines. The researchers then reduced expression of candidate genes and measured lifespan and reproduction. They identified genes whose effects on those outcomes pulled in opposing directions. This provides an experimental way to examine the trade-offs discussed in the opening, rather than relying only on analogies with an optimiser.
The study concerns selected laboratory fly populations and particular genetic interventions. It does not demonstrate that humans have a single programmed deadline, establish which explanation dominates human ageing, or show that a treatment can safely remove every trade-off. Kimmel’s broader engineering optimism is an inference beyond these experiments. The useful result is narrower: lifespan and reproductive performance can respond to genetic change in complicated, sometimes opposing ways, so maximising one outcome need not maximise the other.
Laboratory fly results do not determine human lifespan or clinical feasibility.
Kimmel uses TRIM5alpha and an extinct retrovirus to illustrate how past infections may shape present-day host defences.
Background and context
Kaiser, Malik and Emerman’s June 2007 experiment reconstructed the core protein of an ancient retrovirus using sequences retained in chimpanzee genomes. They found that human TRIM5alpha restricted the reconstructed viral system. The authors suggested that evolutionary changes associated with resistance to that virus could have contributed to susceptibility to HIV-1.
The distinction between result and interpretation matters. Restriction in an experimental system was measured; the historical trade-off is the authors’ proposed explanation. Neither establishes universal resistance or susceptibility across all primates and all simian immunodeficiency viruses. Nor does a molecular restriction result constitute a demonstrated HIV treatment. The paper supports Kimmel’s central example of an evolutionary legacy while leaving his broader comparative shorthand and therapeutic extrapolation as separate questions. A defence can be effective against one viral target without being a general shield against every related infection.
The assay is virus-specific and does not demonstrate a clinical intervention.
Kimmel contrasts Yamanaka’s identifiable pluripotency result with the harder task of making aged cells function better while retaining their identity.
Background and context
Takahashi and Yamanaka’s August 2006 paper showed that introducing four factors could produce induced pluripotent cells from mouse fibroblasts under specified culture conditions. The resulting cells expressed characteristic markers, contributed to embryonic development and formed tumours containing multiple tissue types when transplanted into mice. Those assays addressed whether the cells had acquired a different developmental potential.
They did not establish an anti-ageing medicine or preservation of a mature cell’s normal role in an adult organ. Indeed, the tumour assay that helped demonstrate pluripotency illustrates why unrestricted conversion is not automatically a useful therapeutic endpoint. Kimmel’s distinction is therefore substantive: a successful identity-change experiment can have an obvious readout, while useful rejuvenation also requires retained cell identity, function and acceptable safety. The 2006 result is an experimental starting point, not clinical validation of the newer programme he describes.
The cited study used mouse cells and did not test human rejuvenation.
Kimmel describes lipid nanoparticles and viral vectors as useful but limited ways to deliver genetic medicines.
Background and context
Gillmore and colleagues’ June 2021 report provides a specific human example of lipid-nanoparticle delivery. In the initial six participants with hereditary transthyretin amyloidosis, an investigational CRISPR-based treatment delivered components intended to reduce production of transthyretin. The early report measured safety and changes in circulating protein over the first 28 days; reductions differed between the two dose groups.
This was a targeted gene-editing programme for a defined disease. It was not epigenetic age reprogramming, delivery to every cell type, or proof of long-term clinical benefit from rejuvenation. Its value here is to make the word delivery concrete: an intervention must reach an appropriate tissue, perform its intended molecular action and then be evaluated for outcomes that matter to patients. Progress on one of those steps does not establish all the others or resolve the whole-body problem Patel raises.
Six early trial participants and a 28-day molecular endpoint cannot establish whole-body efficacy or long-term safety.
Kimmel invokes Eroom’s Law to explain why better laboratory tools have not automatically made drug development more productive.
Background and context
Scannell and colleagues’ March 2012 analysis described a historical decline in new drug approvals per inflation-adjusted research dollar. They estimated that this measure had approximately halved every nine years from 1950 through the period they studied. The article also offered possible explanations, including a rising bar for improving on existing medicines and problems in how discovery efforts were organised.
The measure counts approvals against spending; it does not directly measure the health benefit of each medicine, the difficulty of every disease or the productivity of a particular company. The proposed causes are the authors’ interpretation, not a controlled experiment that isolates one mechanism. Nor is a historical pattern a physical law guaranteeing future decline. Kimmel’s proposed platform could be assessed against such a metric, but the 2012 article cannot establish that his programme or AI generally has already reversed the trend.
This is historical analysis with proposed explanations, not a universal law or validation of an AI platform.
Kimmel explains how perturbation screens connect a genetic intervention with a cell’s expression profile and why technical improvements matter for scale.
Background and context
Dixit and colleagues’ December 2016 Perturb-seq paper combined pooled CRISPR perturbations with single-cell RNA sequencing. The team analysed about 200,000 cells across immune cells and cell lines, using the resulting profiles to study how interventions affected genes, cell states and interactions. The paper also discussed the difficulty of exploring combinations as the number of possible interventions grows.
This directly supports the assay concept described in the interview. It does not show that every measured expression change is beneficial, that a predictive model generalises to every human cell, or that a laboratory result becomes a medicine without further testing. Kimmel’s distinction between predicting a state and deciding which state is desirable remains essential. More labelled perturbation data can improve the questions researchers can ask, while choosing meaningful outcomes and establishing safety are additional experimental tasks.
The 2016 assay does not validate every later model, dataset-size claim or treatment programme.
Kimmel gives an approximate drug-spending share before arguing that effective preventive medicines could reduce other healthcare costs.
Background and context
The CMS National Health Expenditure Accounts team’s December 2024 report estimated US health spending at $4.9 trillion in 2023, or 17.6% of GDP. In that published vintage, retail prescription drugs accounted for about 9% of national health expenditure, rather than the approximate 7% mentioned in the conversation. These are dated estimates; subsequent releases can revise historical totals.
The category is specifically retail prescription drugs. It should not be treated as a complete measure of every medicine supplied in hospitals or clinics, or confused with the much broader denominator of national income. The accounts describe spending; they do not establish how much an unproven reprogramming therapy would save. That forecast would require evidence about its clinical effects, price, administration and duration of benefit. Keeping the accounting categories distinct makes Kimmel’s economic hypothesis testable without treating it as an observed saving.
A national spending account is not an estimate of savings from a hypothetical therapy.
Remember a word from the conversation? Find an approximate listening time without reading through a transcript.
Transcript paragraph starts; matching words may occur later. All times are approximate. Use up to six words; matching uses 30-second windows, not exact phrases. Transcription can be wrong, so listen to check. Searches stay in your browser.
PodCompanions