Artificial intelligence is now accelerating biological research so dramatically that experiments Harvard researcher David Sinclair says would take decades—or even centuries—using traditional methods can be digitally narrowed down to a manageable shortlist in a matter of months. Sinclair outlined that shift Friday on a new episode of The Joe Rogan Experience, where AI, notably, became one of the central themes.
The longevity researcher says his team uses AI to screen vast numbers of potential molecules, analyze biological datasets, and evaluate human cells. According to Sinclair, this changes not just the speed of research but also who can do complex science.
“AI has already revolutionized biology,” Sinclair says. Where researchers once had to manually process large volumes of scientific literature, AI systems can now search through research data on a completely different scale.
The development touches directly on one of biotech’s most ambitious frontiers: restoring damaged and aged human cells.
How David Sinclair’s lab actually uses AI
Sinclair uses AI to pick candidates for new treatments from massive molecular libraries. The catch is simple: the theoretical chemical search space is so enormous that it’s impossible to make and test every candidate in a lab.
On the podcast, Sinclair describes a project in which his group is hunting for a single molecule that can influence multiple desired activities in a cell at once. He says the search began with a digital collection that ultimately swelled to about one trillion possible or known molecules.
The lab doesn’t have to test them all physically. Computers first predict how molecules might interact with relevant proteins and enzymes, creating a digital funnel.
Sinclair says his team reduced the vast pool to roughly 200 candidates that can actually be tested in human cells.
His comparison is striking: one digital search that took about two months would have required around 160 years using traditional laboratory methods, he says. With an even larger search space, practical testing could take thousands of years.
Those figures are Sinclair’s own estimates and don’t mean AI literally performed 160 years of full biological experimentation. They reflect the time it would take to probe a very large set of candidates via conventional experimental methods.
The broader trend is clear: AI is shifting part of drug discovery from the physical lab to computational triage—eliminate millions or billions of options digitally, then test only the most promising candidates.
AlphaFold is a key driver of that speed-up
A crucial piece, Sinclair says, is Google DeepMind’s AlphaFold, which predicts the three-dimensional structure of proteins—data researchers need to understand how biological processes work and where drugs might act.
Google DeepMind has now released predictions for more than 200 million protein structures. The AlphaFold database covers virtually all proteins known in scientific repositories.
That’s a big deal for researchers designing new molecules. A protein’s shape largely determines how it functions and how other molecules can interact with it.
On the podcast, Sinclair explains how his team uses compute to virtually test candidate molecules against relevant proteins and enzymes. Thanks to AlphaFold, many protein structures can be predicted in advance, shrinking a major obstacle.
AlphaFold’s impact has been formally recognized. In 2024, Demis Hassabis and John Jumper of Google DeepMind shared the Nobel Prize in Chemistry with David Baker for work on protein structures and computational protein design.
AI also gauges whether human cells look ‘younger’
AI’s role doesn’t stop at molecule hunting. Sinclair’s group also uses it to analyze changes in human cells.
He says AI can help assess whether old human cells, after an experimental treatment, exhibit features that more closely resemble young cells.
That creates a second acceleration.
An algorithm can help decide which compounds to test—and then help interpret the results. In theory, that enables an increasingly automated research loop: design candidates, predict which will work, run experiments, analyze results, and select new candidates based on the data.
That combination is exactly why AI is so compelling for pharma.
Drug development is traditionally defined by massive attrition. Only a small fraction of candidate drugs reach patients. If AI can weed out long-shot candidates early, researchers can focus time and money on a much smaller, stronger set.
That doesn’t prove a drug is safe or effective. Animal studies, toxicology, and clinical trials remain essential. The technology mainly changes how fast teams can arrive at candidates worthy of those experiments.
A 19-year-old used AI to comb gigantic biology datasets
Sinclair shared another example that may better illustrate how AI is reshaping scientific work.
A 19-year-old researcher joined his lab this summer, he says. Sinclair gave him a fundamental question: where does a cell store the information it needs to behave like a younger version of itself?
Sinclair’s research partly centers on the hypothesis that biological aging is linked to loss of epigenetic information. The epigenome encompasses mechanisms that switch DNA regions on or off without altering the DNA code itself.
The young researcher, according to Sinclair, gained access to massive existing biological datasets and used AI to search for patterns.
Two weeks later, he returned with what Sinclair describes as a potential “smoking gun” for his hypothesis.
That’s not an independently confirmed breakthrough. Sinclair did not say the finding has been peer-reviewed. The example mainly shows how AI lets researchers probe datasets too large for individuals to process manually.
This also shifts the profile of a successful scientist. Biology knowledge remains essential, but the ability to steer AI systems, fuse datasets, and test computational hypotheses is becoming critical.
Sinclair sums it up bluntly: the future increasingly belongs to people who can direct AI well.
Why aging research is a natural fit for AI
Longevity research is almost an ideal use case for AI because aging isn’t a single biological process.
DNA, gene expression, proteins, metabolism, mitochondria, inflammation, damaged cells—countless processes interact. Humans can study individual mechanisms, but fusing huge volumes of molecular data is exactly what modern AI excels at.
Sinclair’s lab studies so-called partial epigenetic reprogramming.
It uses three transcription factors: OCT4, SOX2, and KLF4, together abbreviated as OSK. The goal is to roll back certain features of older cells to a younger state without reverting them fully into stem cells.
That last part is crucial. Full reprogramming can strip a cell of its identity and raises safety risks.
The research has now reached human trials
What a few years ago was mostly tested in mice entered a new phase in 2026.
On June 9, Life Biosciences announced the first participant was dosed in a phase 1 trial of ER-100. The experimental therapy uses controlled expression of OCT4, SOX2, and KLF4 and is being tested in conditions that damage the optic nerve.
The U.S. FDA cleared the study earlier this year. The trial initially focuses on safety and tolerability in people with open-angle glaucoma and non-arteritic anterior ischemic optic neuropathy, a condition that disrupts blood flow to the optic nerve.
The study also tracks multiple vision measures.
That distinction is essential. ER-100 is not yet a proven anti-aging treatment, and the ongoing trial does not prove human aging can be reversed. It’s an early clinical safety study of a technology that showed promising preclinical results.
Sinclair says on the podcast that there’s “nothing bad to report” so far from the first patient’s treatment, but no efficacy conclusions can be drawn from that.
AI could deliver a new class of drugs much faster
As biological mechanisms become clearer, AI can massively accelerate what happens next, Sinclair argues.
His lab says it continually receives new candidate molecules from computational analyses. The bottleneck is shifting.
The old problem was finding interesting candidates. The new problem is increasingly that scientists generate more promising leads than they can physically test.
Sinclair is explicit: AI feeds his group new molecules, but it still takes years to evaluate them through animal studies and, ultimately, clinical trials.
That’s a fundamental change for science.
When digital discovery scales exponentially but physical experiments don’t, labs, animal studies, clinical infrastructure, funding, and regulation become the new chokepoints.
More compute alone won’t solve that.
The AI scientist’s toolkit is rapidly expanding
These shifts fit a broader move toward semi-automated science.
AI can parse scientific papers, predict protein structures, design molecules, mine biological datasets, and spot patterns in microscopic images. Robots can then run a portion of the experiments.
Sinclair points to companies combining AI and robotics to further automate drug discovery.
The endgame could be a near-closed research loop: an AI proposes candidates, software selects the highest-yield experiments, robots execute them, and new models analyze the results.
Humans don’t necessarily disappear. Their role shifts toward defining questions, validating results, designing experiments, and assessing safety and ethics.
Will anyone be able to design drugs on a phone?
Sinclair goes a step further. He expects advanced AI tools will become so accessible that people can design drugs—or at least generate candidate molecules—from a phone.
That sounds futuristic, but we’ve seen similar democratization in other AI domains.
Software development once required deep coding skills. Generative AI can now produce entire code blocks from plain language. Image generation, translation, and data analysis have followed the same arc.
Biology, however, has a crucial difference: bad software can crash; bad biology can cause physical harm.
That’s the dark edge of Sinclair’s vision.
The same AI could lower the bar for biological threats
Rogan responds to Sinclair’s prediction about AI-designed medicines with one word: bombs.
Sinclair adds bioweapons.
That turn hits a growing issue in AI safety. A model that’s great at designing therapeutics may also hold knowledge that could be misused to design or tweak harmful biological systems.
The very traits that make AI valuable in drug discovery create risk.
AI can analyze large datasets faster than humans. It can link genetics, protein structures, and molecular interactions. It can generate proposals a researcher might not consider.
In therapy, that’s an advantage. In malicious hands, it’s a threat.
Sinclair says he maintains a database of known pathogens and potential new threats—part of preparing for future pandemics or unnatural biological incidents.
The AI revolution in biology thus has two faces: faster medicine—and a greater need to design biological AI systems safely.
AI is also changing how consumers use health data
AI’s influence isn’t confined to labs, Sinclair argues.
During the conversation, he shows two smart rings and a continuous glucose monitor. These wearables constantly collect data on sleep, heart rate, activity, and metabolism.
According to Sinclair, the companion apps are getting smarter as AI automatically extracts patterns. Instead of just showing a heart-rate or sleep chart, software might flag that a specific meal coincided with a sharp glucose spike.
That sets up a new model of personal health.
The first generation of wearables gathered data. The next is trying to interpret it.
Over time, AI systems could fuse wearable data with blood tests, genetics, medical records, nutrition, exercise, and medications—creating a continuously updated digital view of someone’s health.
That inevitably raises questions about privacy, medical reliability, and accountability. A consumer app misreading a bad night’s sleep is annoying. An AI giving medical advice from incomplete data can have far bigger consequences.
AI may change how we communicate science
Another notable moment comes when Rogan looks up information live on JRT, an experimental LSD-derived compound.
In 2025, UC Davis researchers described a modified psychedelic that showed interesting preclinical effects without LSD’s hallucinogenic properties.
On air, Rogan uses AI search engine Perplexity to pull background almost instantly.
It looks minor, but it signals a second revolution in scientific knowledge.
It’s not just researchers using AI to produce science. Journalists, podcasters, investors, physicians, and consumers are using the same tech to find and interpret it.
The classic route via search engines involves a query, a list of websites, multiple articles, and your own source vetting.
Generative search tries to compress that into a single answer.
It’s faster—but introduces a new trust problem. If the AI misreads a study, lacks context, or cites a weak source, an error can be delivered as a confident answer.
For medical information, source checking becomes more important, not less.
The real AI breakthrough may not be chatbots
The Rogan–Sinclair conversation highlights how limited the public image of AI still is.
For millions, AI means chatbots, images, videos, and smart search. Inside labs, a different AI revolution is unfolding.
There, AI is a tool to shrink nature’s search space.
From billions of molecules, a handful of candidates must emerge. From millions of papers, relevant links must surface. From huge genetic datasets, patterns must be extracted that humans can’t see.
The economic impact could be significant.
Pharma spends years and billions developing drugs, while most candidates fail. Any improvement in early selection can move the needle on cost, speed, and the volume of treatments explored in parallel.
The biggest bottleneck is shifting from ideas to testing
The most striking takeaway from Sinclair’s story isn’t that AI has independently found a cure for aging.
It hasn’t.
The bigger shift is that machines can now generate hypotheses and candidates faster than physical science can vet them.
An algorithm can crunch thousands of options quickly. A clinical trial cannot run thousands of times faster without introducing safety risks.
That creates a new tension.
AI is accelerating the digital side of science exponentially, while biology remains bound to cells that must grow, animals that must be observed, and humans whose safety must be established over time.
In the coming years, it won’t just be the quality of scientific AI that matters. The capacity to validate its output safely will be strategic.
From AI model to medicine is still a long road
Sinclair is bullish on the future of both AI and longevity, but his boldest predictions must be clearly separated from clinically proven results.
A computer flagging a molecule doesn’t make it a drug.
Old human cells showing youthful features in a dish doesn’t mean an older person becomes decades younger.
And an experimental gene therapy entering the clinic doesn’t guarantee it will be safe and effective.
Precisely because AI can generate such convincing possible solutions, that nuance matters more.
The next step in this revolution won’t happen solely in data centers.
It will happen in labs, clinics, and ultimately with patients. If AI plus biotech delivers what researchers like Sinclair expect, artificial intelligence could be far more than a tool for finding information or drafting text.
It could become the technology that directly dictates how fast we discover new medicines, understand disease, and learn to influence biological processes.
That also reframes the core question. No longer just: what can AI think up?
But above all: how fast can we safely find out if it really works?
AI’s spectacular compute speed doesn’t change a basic fact of biology: predictions must survive the physical world. A molecule must actually do what a model predicts. A treatment must prove safe in cells, animals, and people. Effects must be reproducible. And promising results must hold up outside a single lab.
That’s where a major gap may open between the AI revolution on our screens and the one unfolding in science. A chatbot can produce an answer in seconds.
An AI model can traverse a search space in weeks that used to take researchers decades. But treating a patient is something else entirely.
Which makes Sinclair’s “160 years” example most interesting for what it doesn’t mean. AI didn’t do 160 years of science in two months. It collapsed 160 years of potential searching into a much smaller set of experiments humans can actually run. And that may be the real revolution.
Scientists can spend less time blindly hunting for a needle in a haystack. AI can shrink the haystack first. Humans still have to prove the remaining needle is the one we were looking for.
Loading