3.1Discussion
The eleven records read here are inventions, and the paper has treated them as such: a designed set of scenarios in which a studio imagines how a machine's judgment about a person's case might be corrected, or fail to be, between 2029 and 2030. Read that way, the set makes a consistent argument about correction. Whether a judgment can be put right depends less on whether anyone notices the error than on who keeps the clock. In almost every scene someone notices and asks: a student, a tracer, a welfare officer, an access-control officer, a nurse, a caseworker, an outreach worker, a counsellor. What differs is whether the asking starts anything with a date attached.
In the first branch the auditor finds fault often, but the faults take the form of a measurement someone else could repeat. In the second the auditor can only disclaim, and the reasons share one shape: something the auditor would need in order to check the judgment, often including the trigger for a recalculation or the route to a hearing, is sealed, undisclosed or not independent.
One record holds both halves of the argument in a single scene. The nurse who tests the system with a hypothetical drop in oxygen saturation puts, in this paper's reading, a counterfactual question to it, and the system answers. And the record shows that, without a trace of whether the counterfactual was computed, the answer cannot be distinguished from a reassurance. A form of contest that the debate offers as a way around opacity turns out, in this scenario, to depend on the same disclosure it was meant to spare.
Ramírez and Selin, Plausibility and probability in scenario planning (2014), propose moving past the choice between plausible and probable scenarios to a richer set of combinations. The ASI 2030 branches read best, in this paper's reading, as neither a forecast nor a story but as a controlled contrast: the same kinds of question about correction, asked by the same kinds of people in the same domains, set in two institutional worlds. The contrast does not make either world more likely, and the records do not ask that of it.
3.2Contributions
The paper adds three things. First, to the debate on contestability, it adds time. Beside design, information and justification, the records add, in this paper's reading, a further condition: a decision is contestable in substance only if the pace of its correction is set, and can be checked, by someone other than the party that made it. A contest with no date and no independent hearer has the outward form of contest and none of its effect.
Second, for the study of administrative discretion, the records sketch a role that neither the front-line decider nor the vanished official quite covers. The officer, the nurse and the counsellor stay at their posts, but what they do there is relay: they hold no power over the outcome, only the standing to put a question on someone's behalf. Whether that asking is worth anything depends on the clock. In the first branch the officer's question about one day of delay meets a standard and a threshold; in the second the counsellor's question about how many days, and whether the time already waited counts, meets a reply that the system is working as designed.
Third, on transparency, the reading narrows what the sealed branch hides to something concrete. The documents missing there are mostly the ones that would fix a time: when a ranking is recalculated, by when a review must come, whether days already lost are restored. Ananny and Crawford question the equation of seeing a system with governing it. The records suggest, in this paper's reading, one place where seeing still matters in a narrow and practical way: not to understand the system, but to know when it must look again.
3.3Limits
The limits are those of the evidence. The cases are made up, down to their percentages, offices and rules. They tell nothing about the speed at which actual administrations repair their errors or the patience asked of actual claimants, nothing about how deployed AI behaves, and nothing about whether or when general or superintelligent machines will come. The paper reads eleven records on one axis of one studio, chosen by a single rule, and its counts describe those records only. The opinions and the reopening lines are the studio's authored choices, so the pattern found is a pattern of design, not of behaviour. The two branches differ in many ways at once, and the paper cannot separate the effect of who keeps the clock from the effect of secrecy in general; it reads them together because the records do. The frame of repeated wording was measured only among the eleven. The works are read through their abstracts, and several readings set against the records are marked as this paper's own.
3.4Further research
Three lines follow. The seven trunk records on correctability, written before the branches part, could show whether a clock of correction exists before the scenario's audit institutions do. The same question of time could be put to the studio's other axes, especially the one on who may ask a question, to see whether the people who carry questions on someone else's behalf meet the same divide. And the records' practice of writing, for each judgment, the threshold or the day on which it should be revisited could be set beside real administrative decisions, to ask whether people who are told when a decision will be reconsidered contest it differently. That question can only be answered with real decisions and real people, which these synthetic records do not supply.
The contribution, stated once more at the close: in ASI 2030's correctability records, as this paper reads them, a judgment about a person is correctable in substance only when the clock of its correction is held where others can read it, and what the second branch withholds, above all, is the date.