Vivodyne, a biotech startup spun out of the University of Pennsylvania in 2021, is betting that its robotic lab system can fix what it sees as a core flaw in the AI drug-discovery sector: a lack of human-relevant data. The company has built modular machines it calls HIVE that can grow 20 types of human tissue, then autonomously dose and monitor those samples. Vivodyne says this process generates causal biological data that current AI models lack, which today mostly comes from animal testing or studies of single cells and proteins. CEO Andrei Georgescu argues that without human-tissue data, AI models will produce drugs that only work in mice.

The startup’s pitch arrives as prominent tech leaders make bold claims about AI’s medical potential. Anthropic CEO Dario Amodei recently wrote that the idea AI will cure cancer has become more cliche than credible, while Sam Altman and Google DeepMind’s Demis Hassabis have previously suggested AI could lead to cures for all disease. Actual results remain modest, with only a handful of AI-designed drugs in human trials, one as far as Phase III. Alphabet’s Nobel-prize winning Alphafold has yet to produce a new drug, and Isomorphic Labs, the company founded to build on it, has pushed its first expected trials to the end of this year.

Georgescu says the industry needs a sanity check because existing models lack the data to capture human biology’s complexity. The pharmaceutical industry already faces a steep failure rate, with 90% of drugs that pass animal testing failing to win regulatory approval in human trials. Vivodyne claims its lab-grown tissues offer a better predictor: its liver cells have shown 94% accuracy in predicting human toxicity outcomes, its airway tissue matches real human tissue behavior 96% of the time, and its bone marrow has shown full concordance in tests of 20 chemotherapy drugs.

Last week, the company opened what it calls the world’s largest human data center just outside San Francisco, having raised just under $80 million across two rounds led by Khosla Ventures. Georgescu says the facility is already achieving twice the throughput of all animal trials conducted in the US. The goal is to give drugmakers more confidence before they spend tens of millions on clinical trials, a process he likens to automotive crash tests, where automakers expect to pass requirements before testing.

Vivodyne says it is working with multiple major pharma companies, though it will not name them publicly. Beyond these partnerships, Georgescu sees the HIVE machines as a way to generate the causal data needed to train new models on human biology. He points to a study in Nature Methods last month that found no clear data scaling laws when training generative AI models on existing cellular data, which he says is based on static snapshots of cells that do not account for how a cell reached a particular state.

The HIVE machines track hundreds of thousands of ongoing experiments where diseased tissue is exposed to stimuli, which Georgescu expects to enable reinforcement learning for new AI models. He believes this approach will also be essential for combination therapies, which target multiple pathways and dramatically expand the space of possible drug combinations. “You have to say, I want this effect to happen, so what cause should I invoke?” Georgescu said, adding that establishing causality in human biology is the basis of the entire effort.