Active Inference and Intentional Behavior.

Area of Science:

• Theoretical Biology
• Computational Neuroscience
• Artificial Intelligence

Background:

• Basal cognition and sentient behavior may emerge from in vitro cell cultures and neuronal networks.
• Neuronal networks exhibit structured behaviors when interacting with information landscapes.
• The free energy principle offers a framework for understanding self-organization as self-evidencing.

Purpose of the Study:

• To characterize self-organization in neuronal networks through the lens of the free energy principle.
• To define and distinguish between reactive, sentient, and intentional behaviors within active inference.
• To investigate emergent behaviors in simulations of in vitro experiments and machine learning benchmarks.

Main Methods:

• Utilized active inference to model agents that predict action consequences.
• Introduced a formal account of intentional behavior driven by preferred endpoints in latent states.
• Simulated in vitro neuronal cultures playing a simplified Pong game to observe self-organization.
• Applied machine learning benchmarks (grid world navigation, Tower of Hanoi) to study adaptive behavior.

Main Results:

• Neuronal cultures modulated activity to improve gameplay, demonstrating nested, free energy minimizing processes.
• Simulations distinguished between reactive, sentient, and intentional behaviors.
• Intentional behavior was formalized using inductive inference.
• Adaptive behavior emerged rapidly and efficiently under inductive active inference.

Conclusions:

• Emergent properties of neuronal networks, such as basal cognition and sentience, can be understood via self-evidencing and active inference.
• The study provides a formal distinction between reactive, sentient, and intentional behaviors.
• Active inference, particularly inductive forms, facilitates the emergence of efficient adaptive behaviors in artificial agents.