The promises of AI curing cancer and all diseases remain far from decisive practical results, but biotechnology startup Vivodyne believes the main obstacle is not a lack of computing power, but a lack of suitable data about how human tissues work. For this reason, the company developed modular robotic laboratories it calls HIVE, with the aim of producing causal biological data that can be used to train models with a better understanding of the complexities of the human body.
According to the company, HIVE laboratories can grow 20 types of human tissue, then automatically expose them to different doses and monitor their responses. The idea is to move from observing the state of a cell or protein at a specific moment to tracking what happens when a particular stimulus leads to a subsequent biological change.
The Gap Between AI Models and Human Biology
A large portion of the data currently used in drug-discovery research comes from animal experiments or studies of individual cells and proteins. Vivodyne believes these sources do not always provide a sufficient picture of the behavior of living human tissues, particularly when the response results from the interaction of several biological pathways.
Andrei Georgescu, the company’s CEO and co-founder, explained the problem by saying that current models learn to distinguish one “cell state” from another, but do not necessarily learn that the second state resulted from inflammation or a stimulus that caused the transition from the first state. From his perspective, the absence of human tests leaves models vulnerable to producing results that may work in mice but fail to translate to humans.
This observation comes at a time when some claims about AI’s ability to cure cancer have begun to seem more ambitious than the available evidence. Dario Amodei, CEO of Anthropic, has said that talking about curing cancer with AI has become closer to a cliché than a reliable expectation, although he previously discussed this possibility in his writing. Sam Altman has also repeatedly cited curing cancer as a justification for pushing OpenAI toward artificial general intelligence and increasing investment in computing, while Demis Hassabis of Google DeepMind said last year that AI could help cure all diseases within a decade.
What Do the Current Results Say?
Actual results remain limited. A handful of drugs designed with the help of AI have reached human trials, and one has reached Phase 3, but this has not resolved the fundamental obstacles to approving new drugs. AlphaFold, which won a Nobel Prize and contributed to understanding the basic structure of life, has not yet produced an actual new drug, according to the article. Isomorphic Labs, the company founded to build on AlphaFold, expects to begin its first trials by the end of 2026, after they were originally scheduled for 2025.
The article indicates that about 90% of drugs that perform well enough in animal tests to reach clinical trials later fail to receive regulatory approval for human use. Vivodyne says its tissues achieved 94% predictive accuracy in liver-cell toxicity tests compared with human trials, and that airway tissues matched the behavior of real human tissues in 96% of cases, while bone marrow achieved a complete match in tests involving 20 different chemotherapy drugs. These are figures disclosed by the company, not a substitute for broad independent validation.
What Is Vivodyne Actually Building?
Vivodyne spun out of the University of Pennsylvania in 2021 after Georgescu earned a PhD in bioengineering. In the week before the article was published, the company opened what it calls the world’s largest “human data center,” just outside San Francisco. It says the center conducts experiments at a rate equal to twice the total number of animal experiments carried out in the United States, without disclosing the names of its partners, despite confirming collaborations with several major pharmaceutical companies.
The immediate commercial goal is to help drug developers estimate a candidate’s chances of success before spending tens of millions of dollars on a clinical trial. Georgescu compares this with crash testing in the automotive industry: a manufacturer usually knows before testing whether a car will meet NHTSA requirements, whereas drug developers enter clinical trials with less certainty, facing a high likelihood that the drug will fail to receive FDA approval.
Vivodyne has raised just under $80 million through two funding rounds led by Khosla Ventures. The company does not disclose the names of its customers or partners, so the article alone cannot be used to assess the actual scale of commercial use of its platform.
Why Does This Approach Matter?
The significance of Vivodyne’s project lies in its attempt to address a stage often overlooked in discussions about AI: the quality of the data from which a model learns, not merely the model’s size. The company says HIVE laboratories track hundreds of thousands of ongoing experiments in which diseased tissues are exposed to different stimuli, potentially providing training data on causal relationships and perhaps supporting models capable of suggesting the cause that leads to a specific effect.
This type of data could become more important when developing combination therapies that target multiple pathways, since Georgescu believes the number of possible combinations is expanding rapidly and cannot be explored through random experimentation. But the article provides no evidence that Vivodyne is close to curing cancer; instead, it presents a technical bet that improving human data may be a necessary condition for achieving more realistic progress in drug discovery.