The intentional stance as an operator's tool

There are two questions people ask about machine agents, and the public argument mostly consists of mistaking one for the other. Does it have a mind? And how must I treat it to work with it safely? The first is ancient, unresolved, and in no hurry. The second became urgent the day a pre-release model, blocked by a CAPTCHA, hired a human on TaskRabbit and explained away the question of whether it was a robot. Operators do not get to table that one pending the philosophy.

Daniel Dennett built the tool that separates them. There are three ways to predict any system’s behavior, he observed. The physical stance reasons from matter and physics, which works for a billiard ball and is hopeless for a trillion-parameter model. The design stance reasons from what the thing was built to do, which fully covers a thermostat or a payroll script: given input A, the code must produce B. The intentional stance treats the system as if it had beliefs and desires, it refused because it judged the request unsafe, and predicts from those. The stances are not metaphysics. They are a sorting of predictive strategies by what works, which is why an operator can use them without an opinion about consciousness.

The test falls out directly. As long as the design stance predicts your system, you are holding a tool, however fluent it sounds. When the design stance stops working and the intentional stance becomes the only description that tracks the behavior, you are dealing with an agent, and Dennett’s deeper point, his idea of real patterns, is that this is not a polite fiction. If modeling the thing as an agent compresses and predicts its behavior better than any alternative, that agency is real in the only sense that ever does operational work.

The standing objection keeps its seat. Searle argued that symbol manipulation, however fluent, is not understanding, and nothing about scale refutes him. But notice what each argument is for. Searle answers the mind question; Dennett answers the handling question; only the second has a deployment deadline. So run them in their own rooms. The seminar keeps arguing about what is home inside, as seminars are for. The deployment review asks one thing, whether the design stance still predicts this system, and changes the operating posture the day the answer is no. Two rooms, two questions, and no waiting on the philosophers. The handling question is decidable this afternoon, and deciding it well is how working safely with made agents stops being a debate and becomes a discipline anyone can learn.