Seven Stages of Action

In the AI Story

The seven-stage model gave Norman's two gulfs their precise location within human cognition. Stages two through four — where the person translates intention into action — constitute the entire Gulf of Execution. Stages five through seven — where the person translates system feedback into evaluation — constitute the entire Gulf of Evaluation. The model made both gulfs analytically tractable and identified exactly where design interventions could narrow them.

The AI interaction compresses the user's role from seven stages to approximately three: form a goal, communicate it, evaluate the outcome. Stages two through four have been absorbed by the system. The intermediate cognitive work that used to happen between intention and action — the translation, specification, and execution — now happens inside the machine rather than inside the user. This absorption is the structural feature that The Orange Pill identifies as the collapse of the imagination-to-artifact ratio.

The compression is not symmetric in its consequences. The absorbed stages were not only execution work; they were also comprehension work. The programmer who specified and executed code was simultaneously learning about the code she was producing. Removing those stages removes that learning. The user now arrives at stage five (perceive the system state) without having passed through the understanding that stages two through four used to produce. She must evaluate what the system did without the internal model that execution would have built.

Norman's later work acknowledged that the seven-stage model was idealized — real cognition is more distributed, iterative, and opportunistic than a clean sequence suggests. But the framework's analytical power remains in the AI era precisely because it specifies where in the cognitive chain the transformation has occurred and what the consequences of that transformation must be. The evaluation stages are now harder than they have ever been, not because they are more complex in the abstract, but because they must be performed without the preparation the execution stages used to provide.

Origin

Norman introduced the seven-stage model in User Centered System Design (1986) and formalized it in The Design of Everyday Things (1988). The model drew on action theory developed by Norman and colleagues at UCSD, including work with David Rumelhart and Donald Gentner.

Contemporary extensions of the model to AI systems appear in Norman's later writings on complex sociotechnical systems and in the emerging HCI literature on human-AI interaction, where the asymmetric compression of the stages has become an active research area.

Key Ideas

Seven stages map the full action cycle. Goal → intention → specification → execution → perception → interpretation → evaluation. Each stage is a potential site for design intervention.

The gulfs live between stages. Execution gulf spans 2–4; evaluation gulf spans 5–7. Stage 1 (goal) and the bridge back from 7 are internal to the user.

AI compression is asymmetric. Absorbing stages 2–4 eliminates both execution and the comprehension execution produced. The person now jumps from stage 1 to stage 5 without the preparation stages 2–4 provided.

Stage 7 without stages 2–4 is structurally harder. Evaluation requires understanding. Understanding came from execution. Removing execution while preserving evaluation transfers the cognitive load to the stage least equipped to bear it.