Spatial Web AI Podcast Active Inference AI: Cutting-Edge Use Cases | Denise Holt, Mahault Albarracin, PhD & David Bray, PhD
Aug 16, 2025
In this captivating discussion, Mahault Albarracin, a leader in active inference AI at VERSES, and David Bray, a dynamic force at the Stimson Center, share their insights into the future of intelligent systems. They explore how Active Inference AI enhances human-AI collaboration and the potential of federated agent systems for multi-agent coordination. Key topics include the exciting advances in energy-efficient AI models, the implications of shared intelligence for global governance, and the integration of sustainability into technology, all paving the way for a smarter, more adaptive future.
AI Snips
Chapters
Transcript
Episode notes
Biomimetic Models Beat Big-Data Brute Force
- Axiom demonstrates biomimetic AI that learns causes and structure instead of relying on large data and brute force.
- It matched high performance while using dramatically less compute, enabling faster, cheaper generalizable learning.
Use Active Inference For Novelty
- When facing novelty or surprise, choose models that can anticipate and adapt rather than relying on past-data pattern matching.
- Use active inference tools for environments that change unpredictably instead of only generative AI.
Active Inference Models Agents, Causes, And Curiosity
- Active inference embeds embodiment, causality, and dual drives for goals and information gain, unlike typical gradient-descent ML.
- It models causes and agent position in the world, enabling continual learning and purposeful action.