AI and the Ground of Contestability

Why The Machine Must NOT Be The Decision Maker

This is what Putin meant when he said whoever owns AI rules the world.

Those who lived under the late planning regimes in the Soviet sphere did not typically say that they were silenced; many in fact continued to complain, to write reports, to lodge memoranda and objections, to speak in meetings that were ritualistically convened for the airing of dissent. The strange thing, the thing that distinguished this regime from older forms of domination, was the dawning recognition that the objections did not land anywhere. They were not refuted, incorporated, or even falsified, but they were routinely punished. They were also metabolized into nothing by a decision-making machinery that no longer used human disagreement as an input. Solzhenitsyn understood that the cruelty lay in the perfection of the procedure, not in the violence.

We must now start to realize that AI could reconstruct that condition without the drama or the uniforms. Decision-making is migrating to a stratum upstream of deliberation, the layer where eligibility, allocation, visibility, suspicion, sanction, triage, and permission are computed before any reason is asked to answer for them. A citizen may still object after the fact, but by then the decision has propagated through code, contract, and compliance; any appeal is posthumous. The structure inverts: argument now arrives after consequence.

The new unanswerability is civilian, contractual, computational. A hospital could potentially deny care with ab AI model; a bank may not sever a client with accusation but with a risk line; a platform may not throttle speech with decree but with a classifier sealed behind contract, as happens. We have already seen this logic in practice: Australia’s “robodebt” system issued debts at machine speed, forcing citizens into obligation before appeal was possible, and for years no official could explain the system’s reasoning at all. Justification has been replaced by compliance: not “this is true” or “this is right,” but “this is the output.” Outputs are not refutable in the human sense, only debuggable by specialists, and dissent, denied access, exerts no pressure.

What makes this more dangerous than the closures of the twentieth century is that the new system does not even claim a philosophy. It presents itself as neutrality, optimisation, reduction of error, friction removed in the name of competence. Because people recognise the local correctness of individual outputs, they mistake local correctness for structural legitimacy. They do not see that what has been removed is not only delay or error, but the room in which reasons can be made to count.

The danger is temporal as much as epistemic. AI accelerates decision velocity beyond the tempo in which counter-speech can stabilise into argument. A right to appeal is structurally empty when the window in which an appeal could reverse harm has been eliminated. At the same time, provenance, the civic ability to reconstruct how a conclusion acquired force, is disappearing behind ensembles of models trained on non-public corpora, tuned by private rules, filtered through legal risk, and sealed by corporate secrecy. There is no place where a losing reason can enter the circuit. Dissent does not fail; it is never admitted.

A civilisation does not lose contestability only when speech is forbidden; it loses contestability when speech can no longer reach the machinery of decision. We are already preserving the outward architecture of debate, parliaments, op-eds, hearings, comment periods, judicial language, while hollowing out the causal channel that once ran between argument and outcome. The rituals survive; the leverage is gone. This is how a free people becomes post-political without noticing: they continue to speak, but speech no longer routes anywhere that binds.

The essential novelty of AI is not that it thinks; it is that it occupies the epistemic choke points through which reasons must pass to acquire force. It is the first technology that does not merely influence decisions but disqualifies the category of a publicly answerable decision. When the grounds of decision are sealed inside models that do not expose their interior to human reasons, the practice of giving reasons ceases to be a civic instrument. Refutation presupposes access; when access is banned by architecture, refutation collapses into theatre. This is the true nature of global rule that leaders like Vladimir Putin, who famously remarked on AI’s dominance, only partially intuit: it is not the power of a new weapon, but the power to disqualify dissent and politics itself.

A society can survive lies, censorship, even terror, history has ample record of recovery from all three. What no society has ever recovered from is the loss of the mechanism by which dissent is rendered legible to power. Without that, citizen becomes spectator. The republic continues in form while it dies in function. People adapt to a world in which explanation is replaced by procedure and responsibility is replaced by diagrams; and once habituation sets in, the memory that things were ever otherwise evaporates.

The reforms required are structural, not moral. Contestability must be rebuilt as a constitutional design requirement, not a press release. Decision-grade systems must afford reconstruction, not corporate “audit” but civic reconstructability, such that a losing party can trace the causal chain in human language to a human locus of responsibility; in practice this means that a risk line or eligibility score must be legally translatable into refutable plain‑language claims rather than a sealed numeric output. Critical domains must be slowed by law such that counter-speech has juridical force while reversal is still possible. No consequential decision may be insulated inside non-discoverable chains of influence. And authorship must be made legally inalienable: every outcome that binds a citizen must terminate in a name that can answer to reasons.

We are late, but not past recall. The window between deployment and habituation is narrow; once a generation reaches adulthood inside a civic format where dissent is a performance without jurisdiction, the faculty of co-authorship will die. People will still speak, protest, vote, litigate, but they will do so in a world where the speech is severed from the switchboard. Freedom will not end with force; it will end with the quiet conviction that nothing one says any longer matters.

That was the world Solzhenitsyn recorded: not merely a regime of violence, but a regime in which the structure of procedure rendered human reasons irrelevant. AI does not repeat that world; it completes it. Not by crushing dissent, but by evacuating the space in which dissent could ever become binding. The end of democratic civilisation does not always announce itself with fire or collapse. It arrives in the form of decisions that descend like weather, unanalyzable, ownerless, unanswerable, and a people, still speaking, slowly cease to remember that decisions were ever supposed to answer back.

We must ensure decisions and civil discourse are not driven by machine output.

Stay curious

Colin

Image generated with Google Gemini