Cybernetics and World Modeling

Post by @orthogonal-lab.bsky.social — Bluesky

Our paper “A ‘good’ regulator may provide a world model for intelligent systems” is now available as part of the Royal Society A special issue “World models in natural and artificial intelligence”.

Our work as Situated Cognition (its place in the special issue).

Our work situates a cybernetic concept known as the Every Good Regulator Theorem (EGRT) in the modern world of AI research, particularly the notion of a World Model. As per the NVIDIA Glossary:

World models understand the dynamics of the real world, including physics and spatial properties. They can use input data, including text, image, video, and movement, to generate videos that simulate realistic physical environments. World models can be used to generate synthetic data or downstream AI models for training autonomous agents.

There is a diversity of ways in which World Models are explored in this special issue: including but not limited to forms of intelligence, models of the self, theories and hypotheses about the world, complex system control and management, experiential phenomena, and causal models.

Cover for the special issue. Cover art includes examples from spatial cognition in rodents, reflective self-representation in primates, map-based planning in autonomous vehicles, and mirror-based self-recognition in robots.

This paper was the product of the lab’s Cybernetics Interest Group. Check out more of our activities on our YouTube playlist or the Cybernetics and Physical Intelligence (PhASiC) website.