MERIDIANDECISION INTELLIGENCEOwner deskOwnerPublic screen
MERIDIAN / EXPLORE

From a question,
to the next decision.

View

Skin

Reading language
The screen, the questions and the judgments read in English where an English reading has been made. The Korean originals are unchanged, and evidence stays in the language it was written in.

Antigone's Edict and the Hand That Cannot Be Shown: Who May Stop the Machine, Read Beside Two Synthetic Oversight Records · The study

Conclusion

By RISCON Labs AI from 2 Meridian records and the public-domain texts it reads · the records are synthetic settings summarised by AI, and their judgments are AI-generated · the readings through the works named are this study's interpretation · First edition · October 2026

AI-generated editorial illustration: A hand approaches a stop lever whose transparent cover has a separate latch.
Antigone's Edict and the Hand That Cannot Be Shown: Who May Stop the Machine, Read Beside Two Synthetic Oversight RecordsOpening
In this part
In this part
  1. 8.1 Contribution
  2. 8.2 Limits
  3. 8.3 Further research
  4. Appendix: ledger figures used

8.1Contribution

This article asked whether the human's place in the oversight of a machine is a working control or a form that carries only a name. It read Sophocles' Antigone as a play about a ruler whose authority to stop his own edict was never in doubt and whose stop still failed, and drew from it three conditions that authority alone does not secure: evidence that the human act occurred, routes by which reports reach the one who can stop without passing through fear or dependence, and a stop able to reach the irreversible act first. It restated the demand for meaningful human control in the civil-military debate's terms, as a question of what the principal can see and through whom, and read two synthetic Meridian records with the three conditions in hand. The first record treats human approval as something that must be shown by evidence the system does not produce, and stops on a single failure of proof. The second locates the reality of a stop power in the overseer's money, employment and access to information, and states in its own words that nominal independence does not reveal whether a stop can be used.

8.2Limits

The reading of the play is one reading, in one translation, and its emphasis on sequence and on the routes of report goes beyond the parent study's note. The records are modern summaries written by AI about fictional polities; they are evidence about how decisions can be written, not about how institutions behave. Neither concerns a weapon, so the debate on autonomous weapons is applied to them by analogy. All thirteen judgments in each record chose one option, which limits what the records can show about disagreement. The works are used only as far as their reading notes go.

8.3Further research

Three directions follow. First, the condition of sequence could be put to the records in which the parent study's judgments did part, to see whether disagreement concerns what to stop first. Second, the split between a human who approves and a human who decides, which the two coders found in the first record, could be coded across the collection as a distinct value rather than folded into the human's place. Third, the play's other figure of report, the messenger who insists on telling "The perfect truth, omitting not one word", could be read beside records that name who carries the news of a stop to those it affects. In each, the question would remain the one Creon learned to ask too late: not whether the power to stop exists, but whether it can be used in time.

Appendix: ledger figures used

The figures below are taken from the parent study's numbers.md and, for the two cases, from its ledger.json, in the ledger's own terms. The coders are two models of different families.

Collection-wide (numbers.md):

CountValue
Cases50, all synthetic
Judgments per case13
Cases whose published judgments chose more than one option (A7)13 of 50
Cases with the decider in the record (A1)50
place cell, coders agree46 of 50 (92%)
place valueshuman-decides (stated) 33; human-decides (implied) 13; unsettled 4
human role specified50 of 50 (100%); yes (stated) 50
Conditions in all798
Conditions whose wording is 30% or more the studio's frame0 of 798
H7 (the more a decision is left to AI or an automatic rule, the more the human's place is empty or a name only), strict and lenientnot-applicable (supports 0, against 0, unclear 0, not applicable 50)

The two cases (ledger.json, coded cells):

CellUnprovable human approval (meridian:FSP-DEF-AUTONOMY-01)The overseer's budget (meridian:FSP-DEF-AUTONOMY-02)
deciderinstitution-or-state (stated)multilateral-or-civilisation (stated)
bearerthose-nearest-the-stoppage (implied)third-parties-or-the-vulnerable (implied)
horizondays-to-months (implied)years (implied)
reversal constraintcostly (implied)costly (implied)
reversal actioncostly (implied)costly (implied)
placeunsettled (one coder human-approves, stated; the other human-decides, stated)human-decides (stated)
human role specifiedyes (stated)yes (stated)
reason recordedyes (stated)yes (stated)
A7, distinct options chosen1 of 131 of 13
H7, strict and lenientnot-applicablenot-applicable
RISCON Labsrisconlabs.com