Physical Grounds for Causal Perspectivalism.

Area of Science:

• Thermodynamics
• Causality
• Information Theory

Background:

• Causal relations are fundamental to understanding physical systems.
• Existing models often lack a grounding in physical processes.
• The nature of causality in open, irreversible systems remains an active area of research.

Purpose of the Study:

• To ground the asymmetry of causal relations in the physical states of a causal agent.
• To investigate the role of thermodynamics in learning causal relations.
• To demonstrate causal perspectivalism through a physical system.

Main Methods:

• Modeling a causal agent as an autonomous, non-equilibrium system with sensors, actuators, and a learning machine.
• Utilizing feedback mechanisms within the learning machine, driven by thermodynamic constraints.
• Analyzing the relationship between error minimization and power dissipation during learning.

Main Results:

• Causal relations are identified as learned probabilistic functional relations between internal physical states.
• Learning is driven by minimizing dissipated power, linking thermodynamics to causal inference.
• Learned causal relations are shown to be dependent on the agent's hardware and environment.

Conclusions:

• Causal relations can be physically grounded in the internal states of a causal agent.
• Thermodynamic principles dictate the learning of causal relationships.
• The dependence of causality on the agent's physical constitution demonstrates causal perspectivalism.