Descriptions:
Joseph Krauss, CEO of Radical AI, joins Latent Space hosts Brandon and RJ to explain why materials science — specifically inorganic structural metals and alloys — requires a fundamentally different AI approach than biology or small-molecule drug discovery. The core problem: materials cannot be characterized by string representations like SMILES because critical performance factors (microstructure, processing method, post-manufacturing treatment) are not captured at the compositional level. No single model, Krauss argues, can one-shot a new material that ends up in an iPhone or on a Starship.
Radical AI’s response is the self-driving lab: a closed-loop system that autonomously runs synthesis and characterization experiments, applies computer vision and human-in-the-loop analysis to results, stores everything in a structured database, and feeds that data back to an AI scientist that designs the next experimental campaign. Krauss distinguishes this approach from competitors including Laya, CUSP, and Periodic by emphasizing experimental ground truth over purely generative discovery — a thesis the company has been building toward for two and a half years.
The conversation also covers timelines for aerospace and defense applications (three to five years for defense and space systems, longer for manned flight), how the active learning loop handles negative results and compounding errors, and why chip supply constraints are creating delays across the entire materials industry. One of the more technically substantive AI podcast episodes available for anyone tracking the intersection of AI and scientific discovery.
📺 Source: Latent Space · Published June 17, 2026
🏷️ Format: Podcast
