The Three Delegations — Summary
When Sarah asked the AI to handle her mother’s Medicare Advantage plan selection, three distinct things happened that looked like one. The AI performed cognitive delegation: comparing formularies, weighing network breadth against premium cost, exercising judgment about which factors matter given Margaret’s specific pattern of care. Then it performed execution: populating form fields and clicking submit. And it absorbed burden: the contradictory plan descriptions, the fine print about prior authorization, the complexity that exists because it serves institutional interests by discouraging comparison shopping. These are three different delegations. We collapse them into a single word, “help,” and obscure what is actually happening and what it costs.
Cognitive delegation carries a specific risk: the gap between composition and recognition. If Maria never drafts her own appeal letters, she loses the ability to judge whether the AI’s version is faithful to her situation. Writing requires organizing your own thinking, discovering what you believe through articulation. Reading someone else’s version of your argument requires only recognition. The capacity to advocate for yourself atrophies. You may not notice until the AI is unavailable or gets it wrong.
Execution is not purely mechanical either. The hospital administrator who delegates discharge planning to an AI stops encountering the specific friction that planning navigates — the missing ride, the unstable housing, the prescription nobody filled because the pharmacy is forty minutes away and the bus doesn’t run after six. The execution has been delegated; the awareness that came with execution goes with it. Dashboard metrics look fine. The people the metrics describe may not be.
The sharpest implication involves burden. Administrative complexity is not an accident — it defers costs and manages utilization through confusion. Human suffering under that complexity is information, a signal that systems are failing. When AI absorbs the burden, the signal disappears. The system’s operators see improved completion rates rather than lines of frustrated people. AI that manages imposed complexity without addressing its source sustains the very system it appears to oppose. When the AI absorbs the hostility, who remembers that the hostility was wrong?
The same technology produces fundamentally different moral situations depending on who uses it. For populations already navigating intersectional disadvantage, when cognition, execution, and burden are all delegated simultaneously, a form of systemic literacy — however painfully acquired — is lost. The knowledge, shared in waiting rooms and church parking lots, that the system is hostile, could become the basis for collective demand. When the AI absorbs the hostility, the knowledge does not form. Helped into helplessness is still helplessness.