This video features rare, in-depth conversations between Denise Holt, Karl Friston, and Gary Marcus, recorded at the International Workshop on Active Inference 2025 in Montreal, Canada, October 15-17, 2025. #ActiveInference #KarlFriston #GaryMarcus

Filmed on site during the conference, this episode brings together two of the most influential thinkers in AI, neuroscience, and cognitive science to discuss where artificial intelligence is today and where it must go next.

In an extended one-on-one interview, Karl Friston explores Active Inference, the Free Energy Principle, agency, uncertainty, and why current generative AI systems lack true understanding. Friston explains why intelligence must be grounded in generative models of the world, how uncertainty and curiosity drive real intelligence, and why Active Inference represents a first-principles approach to building adaptive,autonomous systems.

The episode also includes a candid interview with Gary Marcus, one of the most prominent critics of large language models. Marcus discusses the limitations of LLMs, the dangers of hallucination and overconfidence, the need for explicit world models, and why symbolic reasoning and probabilistic structure must return to the core of AI systems. He alsoaddresses governance, energy efficiency, and why scaling data and compute alone will not lead to reliable intelligence.

Finally, the video features a joint on-stagediscussion with Karl Friston and Gary Marcus, moderated by Tim Verbelen at the International Workshop on Active Inference, where they debate the future of AIbeyond deep learning, the role of uncertainty, world models, agency, and what it will take to move from pattern generation to genuine understanding.

This episode was recorded live at IWAI 2025, where researchers from around the world gathered to advance Active Inference as a foundational framework for intelligence in neuroscience, robotics, and artificial systems.

Topics covered in this video include:

·     Karl Friston on Active Inference and the FreeEnergy Principle

·     Karl Friston on Yann LeCun

·    Gary Marcus on the limits of large languagemodels

·     Why uncertainty is essential for intelligence

·     Agency, curiosity, and information-seekingbehavior

·     World models vs generative AI

·     Energy efficiency and sustainability in AI

·     Why scaling deep learning is not enough

·     The future of adaptive, autonomous AI systems

This conversation is essential viewing for anyone interested in the future of AI, neuroscience, cognitive science, and the next generation of intelligent systems.

Recorded at the International Workshop on Active Inference 2025 in Montreal, Canada.

Learn more about Active Inference AI at https://aix.us.com/

#ActiveInference #KarlFriston #GaryMarcus #AIX #SeedIQ #IWAI

Episode ID: 1000744622881

GUID: c63d3605-b795-46fb-9301-db8d61a1a6db

Release Date: 11/01/2026, 02:16:43