The tech industry has spent years predicting that artificial intelligence will eventually cure cancer. But according to TechCrunch AI, a biotech startup called Vivodyne argues the AI drug discovery sector is currently hitting a massive wall: it has a fundamental data problem. Current models are largely trained on animal testing or static single-cell data, which limits their real-world human application.
The Biological Bottleneck
OpenAI’s Sam Altman and Google DeepMind’s Demis Hassabis have both cited curing diseases as a core justification for massive AI compute buildouts. Anthropic CEO Dario Amodei has echoed similar sentiments. Yet, the reality remains complicated.
While breakthroughs like the Nobel-winning AlphaFold mapped the building blocks of life, they have not yet produced an approved new drug. Vivodyne CEO Andrei Georgescu notes that absent actual human testing data, these AI models are only going to figure out how to cure cancer in mice. This disconnect explains why 90% of drugs that succeed in animal testing still fail to receive regulatory approval for humans.
Generating Causal Human Data
To bridge this gap, Vivodyne built modular robotic labs called HIVE. These machines can grow 20 different kinds of human tissue, autonomously dose them with drugs, and continuously monitor the reactions.
The startup, which recently raised nearly $80 million and opened a massive “human data center” near San Francisco, says its tissues closely match real human organ behavior. Their liver cells boast 94% predictive accuracy compared to human toxicity trials, and their bone marrow models achieved 100% concordance in tests of 20 different chemotherapy drugs.
Fixing the Scaling Law Problem
What stands out here is the critical shift from static to causal data. Recent studies show that generative AI models trained on existing cellular data do not exhibit the same clear scaling laws we see in large language models. Because traditional biological training relies on static snapshots of cells, the models learn what a cell state looks like, but not the mechanisms of how it got there.
Vivodyne’s continuous monitoring provides the missing context needed to advance the field:
- Continuous tracking: Observing hundreds of thousands of tissue experiments in real-time.
- Cause and effect: Recording exactly how a specific drug stimulus causes a cell to change states.
- Reinforcement learning: Providing the dynamic feedback loops necessary to train models that truly understand complex human biology.
Strategic Recommendation
For AI practitioners and pharmaceutical companies, the takeaway is clear: compute power alone will not solve biology. The next major moat in AI drug discovery will belong to the organizations that can generate high-fidelity, causal human data at scale.
As modern medicine moves toward complex combination therapies that target multiple biological pathways, traditional trial-and-error experimental approaches will no longer cut it. To actually cure complex diseases, the AI industry must first build systems that understand cause and effect in living human tissue. Readers can explore the full details of Vivodyne’s approach at the original source.