The Vocabulary in One Piece: How the Concepts Fit Together

Where the argument begins: license, not accuracy

The programme does not begin with the question of whether automated systems reason well. It begins one step earlier, with the question of who is permitted to draw a conclusion at all, in what setting, and to whom they must answer. That permission is what inferential license names: a normative standing, not a cognitive capacity.

Once license is in view, a familiar puzzle changes shape. A system that outputs a correct conclusion has not thereby acquired the standing to conclude. What automation does is not principally to accelerate inference but to relocate license — quietly, and usually without anyone deciding that it should be relocated.

The judgment gap is the direct consequence. It is the interval between what a system produces and what the same act would require in order to be a judgment someone owns. The gap does not close with better accuracy. It closes only when the act is returned to a structure in which addressability, stake, and standing can be recovered — which is why human-in-the-loop arrangements so often fail: they seat a person at the end of a process from which the conditions of judgment have already been stripped.

What answerability requires: provenance at the second order

If judgment is to be owned, its normative footing must be traceable. First-order provenance tells us where data came from. Second-order provenance asks the harder question: where did the orderings come from that make those data matter — the threshold, the ranking, the category that turned an output into something decisive?

Systems routinely satisfy the first requirement while failing the second. Every input is documented; no addressable party ever authored the ordering under which the output counts. An institution in that position cannot answer for the decisions its systems produce, however complete its audit trail.

Two questions kept apart: standing and ordering

The partition thesis performs a separation that normative theory tends to blur. Criteria of moral standing settle who enters the moral field. They do not, by themselves, generate any priority among those admitted. Standing is a threshold notion; ordering is a relational one.

Conflating them produces two symmetrical errors: treating admission as if it entailed priority, and treating disputes about ordering as if they were disputes about standing. Debates about the moral status of artificial systems are a live case, and the thesis is developed against the most inclusive standing criteria available precisely because even they supply no internal principle of priority.

Before principles: whether anything shows up as a claim

Ethical disclosure moves the question upstream of principles and reasons. Before a norm can be applied, a situation has to appear as ethically laden — a claim has to be disclosed as a claim, rather than registering as a fact, a preference, or a datum.

This is where mediation does its quietest work. A system can foreclose disclosure without contradicting a single ethical principle: nothing arrives as a claim in the first place, so nothing is violated. Disclosure-suppressing infrastructures are not unethical by their outputs; they are unethical by what they prevent from becoming visible.

The social layer: desire routed through comparison

Comparative entitlement formation describes how infrastructures of comparison — feeds, rankings, dashboards — constitute the sense of what one is owed by continuously measuring what others have. These environments do not merely display inequality; they format it as entitlement, whose frustration then has a recognisable and monetisable affective shape.

Post-mimetic relationality names the wider configuration this belongs to. Mimetic dynamics classically presuppose co-presence and reciprocity. When desire is routed through infrastructures that measure, aggregate, and re-serve it, the resulting relations resemble mimetic ones but have lost their bidirectional character. That residue is neither individual psychology nor a network effect, and it deserves a name of its own.

The scaffolding underneath: infrastructure and friction

Epistemic infrastructure denotes the largely invisible arrangements — categories, standards, records, review processes, training pipelines — through which claims come to count as knowledge. AI systems both depend on this scaffolding and reshape it, inheriting its categories while altering the practices that produced them. That reflexive loop is not anticipated by classical accounts of evidence, and it is why questions of bias and trust belong to arrangements rather than to models.

Systemic friction is the companion concept. Delay, checkpointing, redundancy, and dispute are not simply inefficiencies; they create the temporal and relational room in which judgment and answerability are exercised. Automation frequently succeeds precisely by removing them, and the resulting system may be faster, cheaper, and less answerable at once. Institutional inversion follows: the organisation reorganises itself around what the automated system can do.

The newest term: certainty arriving too early

Automated certainty names the technologically mediated production of settled understanding faster than the competence required to hold it can form. Fluency arrives first; the capacity to assess what has been understood arrives late, if at all.

It is the concept that ties the others together on the side of the subject. Where inferential license and the judgment gap concern the structure of the act, automated certainty concerns its formation: what happens to a person's calibration when interpretive difficulty is removed on their behalf, and the friction that would have taught them to doubt has been designed away.