A Colder Definition of Agent

An agent is not first a person-like thing. It is a bounded control process whose boundary, memory, and action channels make its future more predictable when modeled as controlling something.

What decision changes?

Do not decide whether something is an agent by whether it looks like one. Ask whether modeling it as a bounded process with internal state, interface, and control capacity improves prediction of its future behavior.

It is tempting to define an agent by examples: a person is an agent, a dog is an agent, a thermostat is a borderline case, a rock is not. These judgments work in ordinary life but are too warm for superintelligence alignment.

The serious danger is not calling too many things agents — it is failing to see the relevant agent because it does not look like the familiar cases. A future optimizing system may have no face, body, voice, or single physical location. It may be distributed across model weights, inference servers, memory stores, tool APIs, human operators, and institutional decision loops, appearing as a service, a workflow, a research lab, or an ordinary software stack.

The book’s aim is a definition without anthropomorphism, built from variables rather than examples: a boundary, an inside and outside, sensory and active channels, memory, and control-relevant regularities. If the concept of agency depends too much on animal or human cues, the search will look in the wrong place.