Physical AI Scientist

Overview

Physical AI Scientist is Medra's term for a robotic system that executes diverse biological experiments, logs process data, and continuously learns from the results. The concept is defined by three properties: modularity (can address varied experimental workflows), continuous learning (integrates experimental results back into model training), and scalability (throughput can be increased without fundamental redesign). It is distinct from both traditional lab automation and from purely software-based AI scientists.

The concept draws on a structural insight: the physical-digital interface — the point where a robotic system interacts with real-world instruments and biological samples — is the durable competitive moat in autonomous science (per Samuel Stanton of Anthropic/Coefficient Bio). Software agents can reason and plan, but cannot generate the proprietary experimental data needed to train domain-specific models without physical execution capability.