Ruffini, G.

In the Algorithmic Information Theory of Consciousness (KT), algorithmic agents use compressive models inferred from world data to plan actions that maximize their objective function. What are the structural and dynamical consequences of tracking natural data generated by simple world models? To address this, we first propose a formalization of the concept of a generative model using the language of symmetry---Lie group theory. Then, using a generic neural network model as an agent system, we show that data tracking constrains the agent\'s dynamical repertoire, forcing it to mirror the symmetry of the generative world model. This narrows down both the space of potential parameters of the agent system and its dynamical repertoire, endowing both with structure inherited from the world. Based on these insights, we examine the link between data-tracking and the manifold hypothesis, which posits that natural high-dimensional data can be compressed into a reduced number of parameters due to the presence of a low-dimensional invariant manifold within the high-dimensional phase space. This work offers a new perspective for identifying neural correlates of agenthood and structured experience in natural agents and for developing AI and computational brain models.