CONCEPT
Goal-Directed Agency
The biological capacity to maintain internal goal states, perceive the world to assess progress toward goals, and act autonomously to reduce discrepancies—the feature Tomasello argues current AI systems lack and thermostats possess.
Goal-directed agency, in Tomasello's 2025 analysis, is what separates biological agents from stimulus-driven devices, regardless of computational sophistication. A thermostat is a simple goal-directed agent: it has a goal state (the target temperature), it perceives the environment (current temperature), it compares perception to goal, and it acts autonomously to reduce the discrepancy (turning heating or cooling on or off). A large language model, despite extraordinary linguistic and reasoning capabilities, is stimulus-driven: it responds to prompts, generates outputs optimized during training, but does not maintain goals it pursues through autonomous perception and action. This distinction is not about complexity but about architecture. The thermostat, trivial in computational terms, is closer to a biological agent than the LLM because it instantiates the feedback structure that evolution built and that defines agency in the biological sense.
