AI needs human tissue data to advance medicine, not just hype about cancer cures

The promise that artificial intelligence will soon cure cancer has become more marketing catchphrase than achievable goal, according to both skeptical tech leaders and a biotech startup trying to address the root problem: today's AI systems are trained on data that doesn't represent actual human biology.

The data problem holding back AI medicine

Vivodyne, a University of Pennsylvania spinout, contends that most AI drug-discovery models rely on incomplete information sources — primarily animal studies, isolated cell cultures, or individual protein structures. Without access to how living human tissue actually responds to treatments, these models struggle to identify truly effective drugs. CEO Andrei Georgescu frames the limitation bluntly: absent direct human testing, AI drug developers "are going to cure cancer in mice."

Even industry leaders are pushing back against exaggerated claims. Anthropic's Dario Amodei recently noted that the pitch has become tiresome, with Silicon Valley executives—including OpenAI's Sam Altman and Google DeepMind's Demis Hassabis—having repeatedly positioned AI-driven cures as justification for massive research investments and computing expenditure.

The gap between promise and delivery

The disconnect between rhetoric and reality is stark. While a handful of AI-designed drug candidates have advanced to human trials, and one has reached Phase III testing, meaningful breakthroughs remain elusive. DeepMind's celebrated AlphaFold—which won Nobel recognition for mapping protein structures—has not yet led to a single approved drug. Isomorphic Labs, the company built to capitalize on AlphaFold's insights, now expects its first human trials by year-end 2026, having initially targeted 2025.

The pharmaceutical industry already faces this challenge directly: roughly 90% of drugs that show promise in animal testing fail to gain regulatory approval when tested in humans, exposing a fundamental gap between preclinical and clinical outcomes.

A new approach to drug testing

Vivodyne's solution centers on HIVE, modular robotic laboratory systems that can cultivate 20 different types of human tissue, then autonomously administer treatments and track responses. The company claims its synthetic tissues achieve remarkable fidelity to human biology—liver cells show 94% accuracy in predicting human toxicity outcomes, airway tissue matches real human airway behavior 96% of the time, and bone marrow has demonstrated 100% concordance with chemotherapy drug screening data.

Having raised approximately $80 million, the startup recently opened what it describes as the world's largest "human data center" near San Francisco. Georgescu asserts the facility already produces twice the experimental throughput of all U.S. animal testing combined.