Is expertise really just knowing more than others?

The standard mental model of expertise is "knowing more about X than other people." On this model, expertise is a stock of domain knowledge plus the ability to apply it. Replace the human knowledge stock with a larger AI knowledge stock and the model predicts AI replacement of experts.

The model under-describes expertise in a load-bearing way. Real expertise is also a role — a social position with timing, context, and audience awareness built into its performance. The expert is not just someone who knows, but someone who knows when to speak, when to listen, when to insist, when to defer, when to translate technical content for a non-technical audience, when to drop the translation because the audience is technical. The role-performance is what makes expertise effective in any specific situation. Without it, the same knowledge produces nothing useful.

This is why two experts with identical domain knowledge produce systematically different outcomes in the same situation. One is better at the role. They read the room. They time their interventions. They know which 5% of their knowledge is relevant here and now and they suppress the other 95%. The role-performance is at least as much of expertise as the underlying knowledge stock — possibly more, since the knowledge stock is increasingly available to anyone via AI.

AI cannot perform the expert role. It does not know the audience as a specific group with specific concerns. It does not know the timing — when this conversation is, what came before, what is being implicitly negotiated alongside the explicit content. It does not know which subset of available knowledge is relevant here. It produces, on demand, the available knowledge in fluent form, which is what the knowledge-stock model of expertise predicts but is precisely not what the role-performance model of expertise requires.

The diagnostic implication: AI is most disruptive to roles where the expertise is mostly knowledge stock and least disruptive to roles where the expertise is mostly role-performance. The expertise that survives AI is the expertise that was always more than its content — the expertise that was constituted by the situational judgment that AI cannot perform. Can AI replicate the communicative work experts do? is the closest companion claim; this adds the situational-role dimension that audience-anticipation alone does not capture.

The strongest counterargument: agentic AI with persistent context and user modeling will eventually perform the role. Possibly at the limit, but the role requires being-in-time with the audience, which Can AI attend to someone across the time between turns? — the role-performance is constituted by a temporal mode AI does not have.

Source: Knowledge Custodians