The Mechanics of Agency

Summary

Defines genuine agency through three essential criteria: (1) Embeddedness (existing within/interacting with environment, boundary conditions, continuous information/energy exchange), (2) Predictive Modeling (internal representations, simulation/forecasting, outcome evaluation), (3) Intentional Biasing (preferences/goals, influencing outcomes toward preferred states, non-random goal-directed behavior). Distinguishes genuine agents from purely reactive/mechanical systems. Identifies Minimal Viable Agent (MVA): simplest system fulfilling criteria—minimal reinforcement learning agents emerge as promising candidates. Explores biological agency: single-celled organisms satisfy criteria through chemotaxis, habituation, goal-directed resource seeking.

Key Insight: Genuine agency goes beyond stimulus-response, exhibits internally driven goal-directed behavior, modifies actions based on anticipated (not just current) states.

Tags

• agency
• models
• biology
• evolution
• cybernetics

Cross-References

• Related: The Physics of Agency, Part 4: The Law of Control Work — Agency Costs Energy
• Related: MVA/MTA spectrum

Notes

• Foundational definition used throughout archive
• MVA concept bridges minimal/maximal agency spectrum
• Biological examples ground abstract theory
• Sets up later AI alignment work (what systems count as agents?)