Post #6684

🧬Google, DeepMind & MIT build AI Scientist for biomarker discovery A joint team from Google Research, Google DeepMind, and MIT introduced CoDaS, a multi-agent system that can run the full biomarker discovery pipeline autonomously from raw wearable sensor data to hypothesis generation, statistical validation, and final manuscript writing. • Discovered and named new biomarkers with no human guidance, including “late-night doomscrolling” as a predictor of depression severity. • Built-in adversarial checks rejected flawed but misleading results, like using glucose² to predict insulin resistance. • AI-written papers achieved an 86% acceptance rate in blind expert reviews. • Rival AI science agents saw rejection rates of 85% to 100%. • Compressed 37 person-days of expert research into 6–8 hours. AI is moving beyond copilots and into autonomous scientific discovery. Source. @aipost🏴