AI often acts before the decision: sorting, prioritization, recommendation. Who can reach the form when algorithmic pre-forms alter practical standing?
The case meets the pre-form — the answer reaches only the decision, not the pre-form.
Artificial intelligence in administration is often discussed between two extremes. One side expects faster procedures, better forecasts, relief for staff and more consistent decisions. The other side fears opaque automation, discrimination, surveillance and loss of human control. Both perspectives see part of the problem. Yet the decisive question lies earlier: where exactly does AI become a form through which administration acts?
AI does not have to decide a case in order to matter. It can sort, prioritize, summarize, flag risks, suggest text, recognize patterns, identify anomalies or prestructure files. Its power often lies before the visible decision. It changes the path on which a case reaches human decision-making. It can determine what appears urgent, plausible, risky, incomplete or atypical.
This is why AI in administration needs responsive capacity. A human signature at the end is not enough if the operative form has already shaped the file. Nor is it enough to promise transparency in general. Those affected, and the administration itself, must be able to reach the form that produced the sorting, recommendation or signal.
The normal case may appear harmless. A system helps distribute applications, recognize missing documents, summarize extensive files, detect fraud patterns or support allocation of scarce resources. In many contexts such assistance can be reasonable. Administration is overburdened, and digital tools can increase consistency. But wherever AI affects practical standing, its consequences must remain returnable.
A risk flag, for example, may not be a final decision. It can still alter attention, burden of explanation, processing time or tone of review. A prioritization does not necessarily refuse a service. It can nonetheless move some cases forward and others back. A generated summary may not replace the file. It can still shape what the decision-maker reads first. These are form effects before the decision.
Responsive capacity begins with visibility. Which system was used? What function did it have? Did it sort, recommend, summarize or flag? Which data entered the process? Which result influenced the procedure? Those questions cannot be reserved for specialists. They must be translatable into an intelligible answer for those affected and for the bodies that review administrative action.
Then comes objection. An affected person must not be reduced to challenging only the final decision if the relevant form effect occurred earlier. If AI generated a risk signal, categorized a case or suggested a route, the objection must be able to reach that form. Otherwise the answer remains too late and too shallow.
Justification also changes. It is not enough to say that a human made the final decision. The relevant question is whether the human decision remained practically open. If the AI form influenced attention, selection, prioritization or evidentiary burden, reasons must be able to address that influence. Otherwise human oversight becomes a pseudo-answer.
Finally, AI systems require revision. Individual correction is not enough where a model repeatedly generates the same distortions. Then the form itself must be reviewed: data base, training logic, thresholds, deployment context, organizational embedding, human review, monitoring. A system that learns operationally must also be reachable institutionally.
The point is not hostility to AI. Public administration may use technical tools. It may automate where the task is suitable and safeguards are real. But the more AI moves into pre-decision structures, the more necessary it becomes to build answerability at that level. The form must not be allowed to disappear behind the claim that the final decision remains human.
AI in administration needs responsive capacity because it can act before it decides. It produces relevance, routes attention and shapes practical standing. Where it does so, the answer must reach the operative form. Otherwise modernization becomes efficient, but not accountable.