“At this point, saying that AI will cure cancer is more a cliche than it is inspiring, and most people think it is deceptive,” Anthropic CEO Dario Amodei tweeted last weekend. “The thing that will work is actually curing cancer. I think by far the most accurate criticism of AI companies including Anthropic is that we haven’t yet delivered on our big promises to benefit the world. That is totally on us.”
Mentioning “curing cancer,” if not “curing all disease,” has been a nearly universal feature of the AI CEO messaging for some time.
“If AI stays on the trajectory that we think it will, then amazing things will be possible,” OpenAI CEO Sam Altman blogged last September. “Maybe with 10GW of compute, AI can figure out how to cure cancer.”
Over $70b was thrown toward integrating AI into life sciences between 2020 and 2024. Globally, AI data centers have already harnessed an estimated 30GW of computing power, with new contracts and construction projects announced every week. And still, every week, more than 188,000 people die from cancer. Not one AI-discovered drug — at least not one that’s been labelled as such — has hit the market.
The message of Amodei’s tweets is that the AI industry has failed to earn the public’s trust, and its redemption arc hinges in no small part on making biomedical miracles happen. Frontier AI developers are racing to build self-improving AI models under the assumption that, to solve problems as mind-bendingly difficult as cancer (much less all disease), we need superintelligence. In fact, skepticism that superintelligence will be enough is often dismissed as a symptom of not grasping superintelligent AI’s full, glorious potential.
As a mortal human, I’d love to open a chat window and confidently type, “Cure cancer. Make no mistakes. –dangerously-skip-permissions.” But there are plenty of reasons why no amount of geniuses in data centers can break through some of the frustratingly human problems standing between us and our cancer-free future — at least not as quickly as Amodei expects.
To convert raw intelligence into actionable discoveries, you need instruments, institutions and, most importantly, real-world data. Cells don’t divide instantaneously, and mice need time to reproduce. Humans can take entire lifetimes to get sick and heal again. No brain, however godlike and silicon-based, can get around that.
Five years ago, just a few months after COVID vaccines let university labs mostly reopen, I grabbed burritos with a grad school friend. He was a structural biologist, deep in the weeds of studying a specific protein dotting the membranes of neurons, flash-freezing the cells and taking pictures of them under an electron microscope.
“Have you seen this new AlphaFold database?” he asked. “I think everything I’ve been working on is just … done.”
Proteins are long strings of amino acids, tangled up in precise 3D formations that determine what they can stick to. And figuring out a protein’s exact shape has historically been really hard. My friend used an extremely powerful microscope to take pictures of proteins that wound up looking like gray blobs, hinting at the knotted cord underneath. Getting a handful of proteins to exist in isolation long enough to be photographed at all can take a PhD student researcher the better part of their 20s.
On that day in July 2021, DeepMind released the predicted 3D structure of every protein in the human body. It took human researchers about 50 years to do half of that work manually, and was recognized by a Nobel Prize in 2024.
It worked that well because it solved one of biology’s most tractable problems. Conveniently, DeepMind had access to the Protein Data Bank, a preexisting gigantic dataset of clean, standardized inputs. It also knew exactly what to output: a 3D shape, just like it had been trained on. And scientists already had tools to check whether the predictions were correct — they’d been doing this on their own for generations. Structural biologists were effectively mined for training data.
“Curing cancer” is a much gnarlier problem. There is no Cancer Data Bank, and the thing we’d want AI to produce could be anything from a small-molecule drug to a living cell therapy. The data that scientists need isn’t sitting in a cryogenic freezer somewhere, waiting to be organized. It has to be grown, imperfectly, at the pace of natural aging. Superintelligence cannot speedrun time.
Biology spans several levels of abstraction, from individual molecules to entire organisms. Problems sitting at the molecular level are both the most solvable for AI models such as AlphaFold, and the furthest removed from the diseases AI companies need to cure.
While some diseases, such as cystic fibrosis and sickle cell anemia, can be pinned to a single molecular culprit, the conditions that the vast majority of people have to worry about as they get older — cancer, heart disease, and neurodegenerative diseases such as Alzheimer’s — cannot. These conditions are not even “diseases,” per se. The term “cancer,” for instance, refers to over 200 different diseases, each affecting different combinations of cell types and body parts. Even something that sounds localized, like breast cancer, affects far more than breast tissue. Before a single tumor cell metastasizes, it’s already sending chemical signals — proteins — via the bloodstream to other organs, preparing them for the cancer’s arrival. AlphaFold could perfectly predict the structure of every protein in existence and then some, and we still wouldn’t be much closer to understanding why they trigger metastasis, or how to stop it.
Simulating interactions between molecules in silico or even poking at cells in a petri dish simply can’t capture these complexities. To understand how multiple organ systems work together, in sickness and in health, researchers need to observe them in a real-life laboratory — including, at least for now, in living, breathing, tumor-growing animals. “AI will be bottlenecked for questions that require us to generate data that we can’t scale,” Martin Borch Jensen, founder and CSO of Gordian Biotechnology, tells Transformer.
AI has played a role in science for years. Humble machine learning techniques are nearly ubiquitous in large-scale biomedical research, and were used in the development of the Moderna cancer vaccine that’s making headlines this week. But claiming this as evidence that “cancer vaccines [are] now being discovered with AI,” as many Silicon Valley types have done on X, is a huge stretch. Phase 1 clinical trials started in September 2022, months before ChatGPT’s launch. Preclinical drug development had been ongoing for years prior.
Somewhere in San Francisco, someone is furiously typing:
You’re talking about today’s AI models! Once there are billions of superintelligences and armies of robots running labs, they’ll design better experiments than we can. They’ll need fewer measurements and simulate those we couldn’t make. They’ll discover patterns in long-forgotten datasets and PDF documents. You’re just not feeling the ASI, bro. Trust me, bro. It’ll simulate the human condition from first principles.
It’s true that many issues in biology could be solved with a little extra intelligence. Lord knows the scripts I coded as a neuroscience PhD student were deeply embarrassing and could have been much improved by AI. Superhuman AI could dramatically improve every researcher’s workflow, from homing in on which experiments are worth running to streamlining data processing pipelines.
But it can’t simulate the progression of a disease when its progression hasn’t been fully recorded inside a living body yet. Unlike a field such as meteorology, where researchers can describe the atmosphere with equations, we don’t understand the rules of biology. We can’t list everything our bodies contain, much less what everything does. AlphaFold didn’t figure out protein structures from first principles — it learned from generations of human biologists taking pictures of real proteins.
That which has not been measured cannot become training data. And that which must be measured is often really, really annoying to measure. Say you’re trying to figure out why a tumor stopped responding to a drug, so you take a biopsy. That tissue sample gives you a subset of cells, frozen in time — better than nothing, but a woefully incomplete representation of a population of fast-growing cells that carry different genetic mutations. You can’t watch them evolve, because measuring what’s in a cell kills it. And any model of cancer you can watch evolve, like a lab-grown tissue culture in a petri dish, hardly captures the complexity of the living human body.
One might dream of a world where incredibly dexterous robots can do all of this for us, tirelessly testing drug candidates in warehouses full of cancerous lab mice. But Jacob Kimmel, co-founder and CEO of NewLimit, a longevity-focused biotech company, says that industrial robotics still struggle at biology research, however eager people are to assign robots bench work.
Unlike manufacturing, where a robot can repeat the same coded sequence of motions over and over, biomedical research requires making tons of tiny real-time decisions based on faint signals. Superintelligence, combined with physical hardware development, could make autonomous wet lab research more realistic, Kimmel argues. And in some cases, robots are already doing certain easily automatable tasks in wet labs. But “the human hand is still an object so complex, we cannot reproduce anything like [it],” Kimmel says. To do so, we’ll first need to solve the very hard problem of physical intelligence. Even if it improves at a superhuman pace, the field of robotics needs years, if not decades, to build machines capable of handling wriggling mouse pups, or feeling the vibes-based resistance of an incorrectly-placed electrode.
But people have been doing biomedical research for decades, our ASI-pilled friend interjects. There’s gotta be a bunch of data lying around! Why don’t AI companies just grab it, like Anthropic grabbed all those old books?
While there is indeed a bunch of biomedical data lying around, most software engineers would be horrified by its state. When I started grad school less than a decade ago, I was taught to use niche licensed desktop programs running on ancient PCs and record surgical measurements in a generic wide-ruled notebook. “Sure, we have piles and piles of data,” medicinal chemist Derek Lowe wrote. “We don’t know how to clean this stuff up and categorize it … and honestly, we don’t even know if it can be.” For most of the history of biomedicine, data wasn’t collected with the intention of feeding it to large language models. Important variables almost certainly went unrecorded, or got trapped in dusty floppy disks and lab notebooks.
A growing number of biotechnology companies are taking on the mission of running wet lab experiments en masse to create high-quality physiological training sets for future AIs. Jory Bell, who leads Playground Global’s AI-driven biology investments, tells Transformer that these days, his AI-centric biotech investments tend to favor “companies who are working in an area of biology or modality which lacks a current existing data set, and have developed a unique way to develop that data set.”
Some of these experiments can be automated, with robots running controlled trials on hundreds of tissue samples in parallel. But in many cases, researchers can’t escape the fact life marches on at a certain pace, regardless of AI’s capabilities. The measurements that matter the most, including whether a drug meaningfully extends a cancer patient’s life, take the longest to make. Kimmel, who studies longevity, feels the pinch of this bottleneck acutely. When you’re optimizing for the lifespan of a primate, for example, “you’re just not going to be able to scale ‘lifespan of primate’ measurements in any pragmatic way.”
In his most recent essay, Policy on the AI Exponential, Dario Amodei wrote: “...while AI itself is likely to present novel challenges that emerge very quickly and that we have no prior experience in handling, other fields accelerated by AI are likely to encounter a very different problem: regulatory systems that were designed for a slower pace of innovation and are not prepared to handle the deluge of new products and advances that AI will bring.”
Specifically, he predicts a future where AI dramatically increases both the quantity and quality of new drugs, and called for regulatory reform that would fast-track high-quality drug candidates through the approval process. “If my predictions about AI are correct,” Amodei wrote, “there will soon be many instances of interventions that work really well out of the blue, and the regulatory system should be prepared to take them seriously and not adopt a posture of excessive skepticism.”
In his (admittedly strange and widely panned) manifesto published earlier this month, Mark Zuckerberg also called upon the US government to “accelerate society’s ability to develop new cures” by “streamlining how the FDA and other regulators test and approve new treatments.” The general sentiment, it seems, is that science is about to go wild, and regulators need to get out of the way.
Saying that the clinical trial system needs fixing is far from a hot take — many scientists would agree. But Ruxandra Teslo, a scientist who writes about drug development, argues against the sort of fix Amodei proposes. Sure, bureaucracy is frustratingly slow. But part of what slows clinical trials down is their endpoints. To find out whether a novel cancer treatment improves survival odds, or increases the likelihood that someone stays in complete remission for at least five years, you simply need to wait. Improving the odds that a drug works, Teslo notes, doesn’t speed up the process of proving it.
She advocates for more and faster early-stage human trials, while Amodei wants these trials to demand less. Both want better endpoints, which could shorten the wait time from “until a person dies” to “until X biomarker hits Y level.” It’s all a bit circular: Powerful AI could help validate biomarkers that would help human clinical trials go faster. But the data to train those models in the first place would have to come from human clinical trials — which, for the time being, would have to happen at their current pace.
“Will AI cure cancer? No,” says Eric Kelsic, co-founder and CEO of Dyno Therapeutics. “Will AI be a critical part of how we cure all disease, and that’s how we cure cancer?” he continued. “Absolutely yes.”
If you look past the splashy taglines claiming to “solve all disease” or “decod[e] biology to radically improve lives,” it’s clear even the most bullish CEOs scope their projects to account for biological and temporal constraints AI can’t surmount. “You have to find a problem where the AI approach just gives you the solution,” Bell says.
As an investor, Bell worries about the backlash companies may face for taking millions or billions of dollars with a promise to cure all disease, if those cures don’t materialize. There could be “negative cycles of disillusionment with what AI can do to untangle the complexity of biology,” he says. “Those companies are, you know, potentially poisoning the well,” deflating people’s confidence in AI’s usefulness.
AI companies can only promise to cure cancer for so long before people will expect cancer to be cured. At some point in the near future — or perhaps, the recent past — all of the visceral negatives of AI won’t be worth it. Berkeley AI professor Emma Pierson, once Dario Amodei’s mentee, carries a genetic mutation that gives her a roughly one in four chance of getting breast cancer. Still, she wrote: “I will wait a little longer for a cure — even if it means losing my fertility and living under the shadow of risk — if it lets us approach this new world more carefully, and ensure that, in curing cancer, we do not lose the things that make cancer worth curing.”
In any case, the pretty-powerful-but-not-superintelligence available today still can and should be put to work in the life sciences. Its helpfulness is jagged: game-changing where data already exists, and premature where it doesn’t. Structural prediction models are already making it easier to design new and better drugs, accelerating at least one major part of the research pipeline. This could, as Amodei predicted, make some clinical trials faster and cheaper in the long run. Human researchers, armed with more awareness than they had 10 years ago, can start to fill in the physiological gaps, teaching their silicon labmates enough about the biological world that someday, they’ll be able to simulate it in all its messy complexity. Humans — particularly those in the US — can also begin repairing their deeply broken biomedical research institutions.
“Biomedicine and science writ large will end up being limited not just by the amount of cleverness we have available — whether we create that cleverness from electrons or by way of electrons that were once food — but rather by our ability to go out and gather the most useful observations of the universe,” Kimmel says. Scientists still need to test medications, analyze biopsies, and see what happens. AI does not simply “cure cancer.” But it can point scientists toward the slow, unsexy, physical work that might.
Thank you for this post!