Google DeepMind: The Podcast

This episode explores invisible mathematical watermarking in AI, from tracing image and video provenance to protecting protein sequence design. It looks at SynthID, SynthID Bio, and what makes watermarking resilient without changing biological function.

Weather forecasting is the focus here, as Hannah Fry talks with DeepMind researcher Peter Battaglia about how machine learning is reshaping prediction. The episode looks at hurricane warnings, the limits of traditional physics-based models, probabilistic forecasts, and newer systems like GraphCast and WeatherNext 3.

AI uncertainty is the focus here, as Zoubin Ghahramani explores why today’s systems often sound certain even when they’re wrong. The episode looks at Bayesian thinking, real-world uses of uncertainty, and whether better calibrated confidence could be key to future AI progress.

This preview is based on the episode’s published show notes, not the full audio. It spotlights why interpretability matters, the techniques researchers use to inspect AI systems, and the limits of understanding models as we move toward safer, more trustworthy AI.

This preview, based only on the episode notes, looks at a 10-year retrospective on AlphaGo’s match against Lee Sedol. Hannah Fry, Pushmeet Kohli, and Thore Graepel revisit why that moment mattered and how its methods connect to today’s AI breakthroughs.