- Nothing a system knows affects its survival until it acts. Action is the only causal chain from knowledge to consequence.
- The agency interface has three working parts: beliefs that settle enough to commit on, frameworks that compress an intractable world, and plans that sequence decisions.
- Real agents are bounded — they satisfice, not optimize — so effective agency is a balance: beliefs stable enough for action, adaptable enough for learning.
#The claim
Action against reality is the only causal chain toward survival. A system can perceive perfectly, model beautifully, and know truly — and none of it touches the world until it does something. And doing is physical: the arrow from knowing to acting is a powered arrow, paid for in compute and energy. Agency is the groundwork on which knowledge earns its consequence, and in economic terms it is where value is created or captured: the point where knowing converts into doing.
The conversion runs through the agency interface: the beliefs we hold, the frameworks we reason within, and the action plans we commit to in making sequential decisions. Belief is the linchpin. Peirce defined it as the settled state that guides action — the cessation of doubt that permits commitment.Peirce (1877), “The Fixation of Belief.” Belief as the settled state that appeases doubt and guides conduct. An agent that cannot settle cannot commit, and an agent that cannot commit cannot act.
#The mechanism
How does an agent actually hold the world in its head well enough to act on it? Not completely. Formally, agents maintain belief states — probability distributions over possible world states, updated through observation and action.Kaelbling, Littman & Cassandra (1998). Belief states: acting on a probability distribution over worlds you cannot fully see. Practically, no real agent computes over the full distribution, because rationality is bounded: cognitive actors satisfice rather than optimize, relying on frameworks that compress an intractable world into actionable models.Simon (1978), Nobel lecture. Bounded rationality: agents satisfice under real limits of attention and computation.
The mind’s own architecture shows the compression at work. Deliberate System 2 reasoning produces judgments that, through repetition, compress into System 1’s automatic intuitions — yesterday’s slow analysis becoming today’s fast reflex.Kahneman (2011), Thinking, Fast and Slow. Deliberate judgment compressing, with repetition, into automatic intuition. A framework, in this sense, is cached deliberation: the expensive computation done once, amortized over every decision that reuses it.
#Why it matters
Effective agency is a balance the loop must hold: belief systems stable enough for action but adaptable enough for learning. Too rigid, and the agent keeps executing a plan the world has invalidated. Too fluid, and it never commits long enough for consequences to teach it anything.
This is also where the evolution interface cashes out. The conjectures that survived criticism do not act on the world by themselves — they become beliefs and frameworks, and those convert into committed plans. Evolution learns; agency acts; and the trajectories of experience that acting generates are the raw material for the next round of learning. Where that conversion physically lives — in what kind of system — is the next factor’s subject.
Action against reality is the only causal chain toward survival. Agency is the groundwork on which knowledge earns its consequence. Knowledge becomes actionable through the agency interface, through the beliefs we hold, the frameworks we reason within, and the action plans we commit to in making sequential decisions. In economic terms, this is where further value is created or captured. Belief is the settled state that guides action,1 the cessation of doubt that permits commitment. Agents maintain belief states,2 probability distributions over possible world states, updated through observation and action. Rationality is bounded,3 so cognitive actors satisfice rather than optimize, relying on frameworks that compress an intractable world into actionable models. The mind operates through a dual architecture,4 in which System 2’s deliberate reasoning produces judgments that, through repetition, compress into System 1’s automatic intuitions. Effective agency depends on maintaining belief systems both stable enough for action and adaptable enough for learning. The evolution interface is mediated into action precisely here, where the beliefs and frameworks refined through iteration are what convert knowledge into committed plans.
References
- Peirce, C. S. 1877. “The Fixation of Belief.” Popular Science Monthly 12: 1–15.
- Kaelbling, L. P., M. L. Littman, and A. R. Cassandra. 1998. “Planning and Acting in Partially Observable Stochastic Domains.” Artificial Intelligence 101 (1–2): 99–134.
- Simon, H. A. 1978. “Rational Decision-Making in Business Organizations.” Nobel Memorial Lecture.
- Kahneman, D. 2011. Thinking, Fast and Slow. New York: Farrar, Straus and Giroux.