MIT Technology Review Narrated AI materials discovery now needs to move into the real world
6 snips
Dec 24, 2025 Innovative startups are leveraging AI to revolutionize materials discovery, but challenges remain. Despite the excitement around AI breakthroughs like AlphaFold, solving real-world synthesis issues is essential. Experts discuss the slow pace of material testing and the need for automation. Trials with autonomous labs show promise, yet commercial success is still elusive. There's potential for AI to shorten lengthy research cycles, but the industry is wary, waiting for tangible outcomes from this technology.
AI Snips
Chapters
Transcript
Episode notes
AI-Run Sputtering Lab Demo
- Lila Sciences runs a sputtering instrument where an AI agent determines recipes and varies element combinations to make films.
- A human currently approves steps while AI suggests experiments and analyzes results.
Making Beats Imagining
- The biggest bottleneck in materials discovery is making and testing real-world samples, not ideation or simulation.
- Simulations frame problems but cannot fully replace lab experiments for real-world validation.
Simulations Miss Real-World Conditions
- DeepMind's large-scale computational predictions exposed the simulation-to-reality gap when many proposed crystals fail at lab temperatures.
- Predicting atomic minima at absolute zero doesn't guarantee stable, useful materials at real temperatures.