The most interesting AI lab right now might not be in a San Francisco office tower. It is, reportedly, growing in a petri dish. Michael Polansky, Lady Gaga's partner and a veteran of Sean Parker's orbit, is training AI models on living skin tissue, betting that biological feedback loops will teach machines things static datasets never could. The body as training environment. The body as compute substrate.
From Skin to Synapses: Biotech AI's Expanding Ambition
Zoom out and the pattern sharpens. Nature's latest briefing covers a new narcolepsy drug that researchers believe could unlock an entirely novel class of brain therapies, because the mechanism it targets sits at the intersection of sleep, consciousness, and neurochemical regulation. Suddenly a drug about staying awake is a master key to the brain. Both stories share a logic: go deep enough into one biological system and you find a lever that moves everything else. Polansky's living-skin AI is chasing the same insight. Skin is not just an organ. It is a sensing, signaling, adaptive surface that processes environmental data in real time. Training a model on it is less about dermatology than about learning how biological intelligence actually works.
The Antibody Problem and Scientific Trust
There is a cautionary shadow here. Nature also reports that a sleuth has identified dozens of studies that used the wrong antibody, corrupting years of biomedical research downstream. The lesson is uncomfortable: biological data is hard to verify, and errors compound silently. If AI models are now being trained on living tissue, the quality-control problem does not go away. It scales. The dream of AI-accelerated biotech runs directly into the oldest problem in science: garbage in, garbage out, except now the garbage is alive. Eugenia Kuyda's framing of software as a garden you raise rather than a product you ship feels newly literal when the training data has a pulse.