AI has become a familiar part of the conversation about future medicine, but Vivodyne says the harder problem is not hype or computing power. The biotech startup argues that AI drug discovery needs better human biology data before it can make meaningful progress against complex disease.
The company has built HIVE, a system of modular robotic labs that can grow 20 kinds of human tissue, dose them autonomously, and monitor what happens. Vivodyne says that kind of living-tissue evidence is what today’s AI models are missing.
The data gap behind AI drug discovery
Vivodyne’s central claim is direct: many AI systems are trying to learn biology from inputs that do not fully represent how human tissue behaves. The source of data matters because drug candidates often look promising before they are tested in people.
Today, much of the relevant evidence comes from animal testing, or from studies of single cells or proteins. Those sources can be useful, but Vivodyne argues they are not enough to capture the complexity of living human tissue.
Andrei Georgescu, Vivodyne’s CEO and co-founder, frames the issue as a practical limitation for AI models. Without human testing, he asks what those models can actually learn to do. His answer is that they risk solving disease in mice rather than in people.
That criticism lands in a field where major AI leaders have made large claims about medicine. Anthropic CEO Dario Amodei recently wrote that claims about AI curing cancer have become more cliche than credible. Sam Altman has repeatedly used curing cancer as part of the case for OpenAI’s push toward AGI and larger compute buildouts, while Google DeepMind’s Demis Hassabis said last year that AI could potentially cure all disease within a decade.
The record so far is more restrained. A handful of AI-designed drugs have reached human trials, with one reaching Phase III, widespread human testing. But Vivodyne’s point is that the biggest obstacles may not be problems AI can solve today without a stronger biological foundation.
What HIVE is designed to produce
HIVE is Vivodyne’s attempt to build that foundation. The company says its robotic labs can grow human tissue, expose it to drug candidates or other stimuli, and observe the resulting biological changes over time.
That time element is important. Georgescu argues that many current models are trained on static snapshots of cells. In his view, those models may learn that one cell state exists and another cell state exists, but not the causal path that connects them.
Vivodyne wants its machines to generate data about cause and effect in living human tissue. The company says HIVE machines are tracking hundreds of thousands of ongoing experiments, including diseased tissue exposed to some stimulus. Georgescu expects that work to support reinforcement learning for models of human biology.
The company’s claims about its tissue systems are specific. Vivodyne says its liver cells have 94% predictive accuracy compared to human trials that test for toxicity. It says its airway tissue matches the behavior of real human tissue 96% of the time, and that its bone marrow has achieved 100% concordance in tests of 20 different chemotherapy drugs.
Those numbers are central to Vivodyne’s argument. If human tissue grown in the lab can better anticipate how people respond, then AI drug discovery may have a stronger way to evaluate which candidates deserve to move forward.
Why animal results are not enough
The problem Vivodyne is targeting is already familiar in pharmaceutical development. The source article says 90% of drugs that are effective in animal testing to enter clinical trials do not receive regulatory approval for humans.
That failure rate matters because clinical trials are expensive and slow compared with earlier testing. The source article notes that a clinical trial typically costs tens of millions of dollars. Drugmakers therefore have a strong reason to improve confidence before taking a candidate into human testing.
Georgescu compares the issue to automotive crash tests. An automaker is typically confident that a car will pass NHTSA requirements before the formal test. Drugmakers, by contrast, often enter clinical trials without the same level of confidence, and most candidates fail to win FDA approval.
Vivodyne says it is working with multiple major pharma companies, though it has not named them publicly. The goal is not simply to make more experiments possible, but to make earlier experiments more relevant to human outcomes.
- Better screening: HIVE could help identify drug candidates that appear more likely to work before a clinical trial begins.
- More relevant evidence: Human tissue data may show effects that single-cell, protein, or animal data do not capture.
- Training data for models: Repeated experiments on living tissue could give AI systems more causal information about biology.
The company behind the approach
Vivodyne was spun out of the University of Pennsylvania in 2021, after Georgescu received a PhD in bioengineering there. The company has raised just under $80 million across two rounds led by Khosla Ventures.
Last week, Vivodyne opened what it calls the world’s largest human data center just outside of San Francisco. Georgescu says the team is already achieving twice the throughput of all the animal trials being held in the US.
The phrase human data center points to the company’s larger ambition. Vivodyne is not only trying to test individual drug candidates. It wants to create a stream of human biology data that can train future AI models.
That ambition also reflects a critique of current AI model training. Georgescu points to studies such as one published in Nature Methods last month, which found no clear data scaling laws when training generative AI models on existing cellular data.
In plain terms, simply adding more of the same cellular data may not be enough. Vivodyne’s bet is that AI systems need a different kind of data: experiments that show what caused a biological state to change, not just what that state looked like at one point in time.
What this means for future medicine
Vivodyne’s argument is especially relevant for complex diseases. Georgescu believes future treatments may need combination therapies and drugs that target multiple pathways, unlike the majority of drugs available today.
That creates a search problem. If researchers must explore many possible combinations and biological pathways, the number of options can grow beyond what a purely experimental approach can handle. Georgescu’s view is that AI will need causal models of human biology to navigate that space.
For now, Vivodyne’s message is more grounded than the broad claim that AI will cure cancer. The company is saying that progress depends on the data AI learns from, and that living human tissue may be a missing layer between lab results and human trials.
If HIVE can help drugmakers understand which candidates are more likely to succeed, it could make AI drug discovery more practical. The near-term stake is better evidence before costly trials. The larger stake is whether AI can move from pattern recognition toward a deeper understanding of cause and effect in human biology.