Legibility, answerability, and refusal: a humanities and social sciences framework for the age of automated judgment

In June 2026, Canada released a national AI strategy assembled largely without the  humanities and social sciences in the room. That same season, everywhere else, those  disciplines were being summoned. The frontier AI laboratories were hiring philosophers  at pace; the press was declaring the revenge of the humanities graduate; and the Pope’s  first encyclical, a document on artificial intelligence, drew its deepest arguments from  precisely the traditions our policy process had left in the lobby. The contradiction is  impossible to miss: the disciplines being recruited to solve AI’s hardest problems inside  the labs are still excluded from governing it in public. 

This editorial answers the exclusion with a contribution. It sets out a policy framework,  drawn from the social sciences and humanities, that centres what we call the structural  minimum of personhood under automation: legibility, answerability, and refusal. 

The framework is offered in deliberate kinship with Demos’s Verification, Deliberation,  Accountability (VDA) model for democratic health, which names its trio the structural  minimum of democracy. VDA asks what a democracy needs from its institutions; this  framework asks the adjacent question the AI era forces: what does a person need from  the systems that now judge them? The two questions can come apart, and that is the  danger. A society could pass every democratic test and still deliver its citizens, one by  one, into automated verdicts they cannot read, cannot contest, and cannot escape. The  failure is not the technology. It is that personhood was never on the checklist, and every  harm that follows gets minimized as a technical glitch rather than named as what it is: a  person, erased from the ledger of things worth verifying. 

What personhood needs under automation 

Rights on paper become substance only when a person facing an automated system  can answer three questions in the affirmative; a right you cannot see, argue, or exercise  is a right in name only. 

Legibility: can I read what reads me? 

Automated systems compose accounts of people, scores, bands, categories, inferences,  and act on those accounts. Legibility is the condition of being able to see the account: to  know that a system is operating, what it has inferred, on what basis, and in language a non-specialist can understand. It is verification turned inward: not “can society establish  what is true” but “can I establish what is being held true about me.” 

Legibility fails in two directions at once. Downward, when the person cannot see the  profile, the reasons, or even the system’s existence, judged at machine speed by  evidence they are never shown. And upward, when the system’s own operators cannot  explain it, the black-box alibi, so that opacity is claimed as technical fate rather than  admitted as design choice. When legibility is robust, a person can inspect the account  against their own life. When it collapses, the person becomes what one of us has  elsewhere called a digital subaltern: endlessly described, never shown the description. 

Answerability: can I answer back? 

Answerability is the condition of being able to respond: to reach a human being with the  authority to review and change an automated outcome, to contest an inference, to  correct the record, and to have the contest matter. It is accountability translated from the  institutional register to the personal one: not only “can the powerful be held to account by  society” but “can this decision, about me, be made to answer.” 

The classical tradition located judgment in the space of encounter, where the judged ones can  appear and speak; automated systems relocate judgment to the one venue where the  person cannot appear at all. Answerability is the demand that a door be built back into  that room. Its tests: an explanation legible to its subject, a reachable human with power,  a contestation route whose outcomes are consequential, and measured time-to resolution. A complaint box is often offered as redress, and an ombudsperson as  oversight; but a complaint box is not answerability, and an ombudsperson who cannot  alter the outcome is not a door. The test is whether answering back has ever changed a  result. 

Refusal: can I decline and still live? 

Refusal is the deepest condition, and the one most often missing. It is the capacity to  say no to a system, to decline the collection, exit the service, opt out of the inference,  without forfeiting participation in ordinary life. It is deliberation’s hard edge: a voice that  cannot say no is not a voice; it is a formality on the way to yes. 

Refusal is where the present order is most degraded, by design. Consent has been  engineered into theatre: the pre-ticked default, the unreadable terms, the enrolment that  takes one click and the exit that takes an afternoon, the “agreement” extracted at the  checkout of every necessity. And refusal is where the collective dimension becomes  unavoidable: for many communities, and Indigenous data governance states this with  the greatest clarity, consent is not an individual checkbox but a collective, ongoing  relation, and no individual click can waive what belongs to a people. A framework that  cannot handle collective refusal cannot govern data at all.

Figure 1: Personhood needs under automation

The three conditions hold only together, as Figure 1 shows. Legibility without  answerability is a window onto a verdict: informed, but unheard. Answerability without  legibility is an appeal argued blindfolded: heard, but without the information redress  requires. And both together, without a genuine right of refusal, amount to courtesies  inside a cage. The framework is therefore offered as a single standard, implemented in  tandem or not at all. 

The lineage: refusal has a history, and it is curriculum 

Here the humanities contribute what no compliance regime can: the knowledge that all  of this has been survived before, and how. Every community that has lived under hostile  systems of description and control has built what this framework names. Policy  habitually treats these histories as heritage: honoured, quoted, quarantined. This  framework treats them as the tested engineering of legibility, answerability, and refusal  under maximum adversity, the deepest expertise available on problems AI governance  calls novel. The counterpublics that built these tools reconstructed personhood from  below; ignoring their methods means relearning what the tradition already knows.

The case for the answerable society 

The wager is simple. The AI question has been treated as a technical question with  social side effects, to be managed by the disciplines that build, with the disciplines that  understand people invited to comment afterward. Every failure catalogued here, the  verdicts in absentia, the consent theatre, the strategies written without the humanities,  follows from that ordering. The correction is not hostility to the technical disciplines,  whose craft is necessary and often magnificent. It is the restoration of the older ordering,  in which we build answers to what we owe each other, and the disciplines whose  entire subject is what we owe each other hold structural, not ceremonial, authority over  the systems now composing our characters and dispensing our fates. 

Legibility, answerability, refusal: can I read what reads me, can I answer back, can I  decline and still live. Three questions, one framework: an audit instrument, a design  standard, and a public grammar for the age of automated judgment. A society whose  systems can answer all three has nothing to fear from its machines. 

AI use disclosure: the authors used an AI assistant for copy editing, and figure preparation; the  framework, arguments, and final text are the authors’ own.