Contribution
It names three responses to the harms of data-driven government and ranks them. Access and inclusion, and transparency, “seek to remedy harms” while working “largely within the prevailing logic of algorithmic governance”. Data justice, abolition and refusal contest that logic. It treats the three as “a dialogue rather than a chronology”. In the record’s terms, its sharpest point is that access and transparency each run in two directions.
Key insight
Being included is being made visible. “Access is generally angled ‘up’: something is made available to those who otherwise ‘lack’ it. Access is however bidirectional, and access ‘down’ for powerful players is politically consequential.” The same holds for transparency: in Australia’s Robo-debt “No credible claim to transparency was made …, though transparency was arguably demanded of welfare recipients through the ‘reverse onus’”. So whether a remedy helps depends on who is made legible to whom, which a demand for more access or more disclosure does not settle.
Assumptions
- Its object is “algorithmic governance”: algorithms, and the data they rely on, “used within institutional government settings to assist in decision-making and service delivery” (after Danaher), with the focus on public-sector services and social welfare.
- Its normative starting point is data justice: begin from “impacted communities and social groups”, and treat datafication as political rather than technological (Dencik and Kaun).
- The evidence is illustrative cases from the literature and the authors’ research: LinkNYC kiosks, Australia’s Digital Transformation Strategy, Robo-debt, algorithmic impact assessments, Australia’s Smart Cities and Suburbs Program, blocksidewalk in Toronto, tech-worker refusals.
Key results
- Two theses. Responses drawing on “traditional public administration values are not well formulated for addressing the various harms”, and data justice and refusal “holds more promise than access, inclusion and transparency”.
- Access and inclusion. It does not resolve the exclusions that the digital divide follows. It brings “exposure to increased monitoring, social sorting and surveillance”, and it reduces non-digital alternatives, so that services can fail at a single point. LinkNYC’s free Wi-Fi is read, after Green (2020), as a means to collect and monetize data. It quotes Benjamin: “what does it mean to be included, and hence more accurately identifiable, in an unjust set of social relations?” It also states the paper’s charge against “horizontal neutrality”: it “exacerbates existing inequities precisely by treating everybody the same.”
- Transparency. In government “performance” it relies on “techniques of classification, measurement, evaluation and rank, which bring into view that which is to be rendered transparent and acted upon”; “Transparency is a tautological value which justifies itself: it is good to see that it is good to see.” Notice does not secure consent. Where comprehension and contestation cannot be assured, the limits “may render conventional notions of transparency trivial and incongruous”. The “procedural management of opacity” hides a second, contractual opacity: public–private partnerships and proprietary algorithms.
- Refusal. “the capacity to say ‘no’ to the application and development of data-driven systems”, drawn from Benjamin, the Feminist Data Manifest-No and Barabas. It is generative (“refusal and commitment”), collective, and “not reducible to voluntaristic forms of exit”: “Refusal is not absence.” Its cases are blocksidewalk, whose aim was to “put No on the table” and which outlasted Sidewalk Labs’ withdrawal in May 2020; the Google Walkout; UK councils that abandoned tools; and Swedish strikes against automating benefit decisions.
- Limits of refusal. “Those reliant on government services are not afforded many concrete opportunities to refuse.” Under Robo-debt it was “‘impossible to refuse’ the recovery of the assumed debt”. Refusal “does not entail success”, is also bidirectional, and may be co-opted, “where everyone could … ‘just say no’”. The authors “have yet to find widespread collective refusals” in social welfare.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Access and inclusion, and transparency, rest on public-administration values that tend to maintain existing structures, so they reproduce the inequities they aim to correct | weak to moderate | argument from the literature, illustrated by LinkNYC, Robo-debt and impact assessments; no comparison of outcomes |
| C2 | Access and transparency are bidirectional: inclusion gives powerful actors access to the included, and transparency can be demanded of the governed | moderate as a conceptual point | LinkNYC and smart-kiosk cases; Robo-debt’s reverse onus |
| C3 | Transparency without the capacity to comprehend and contest is trivial, and procedural transparency obscures contractual opacity | moderate | Robo-debt; public–private partnerships with proprietary algorithms |
| C4 | Refusal is a more transformative response, collective and generative, and not reducible to exit | weak | argued; blocksidewalk and other cases are illustrations, not tests |
| C5 | The capacity to refuse is unevenly distributed and weakest for those who depend on government | moderate | Robo-debt; the authors find no widespread collective refusal in welfare |
Concepts
- Algorithmic governance: algorithms and their data, used in government settings for decisions and service delivery; also the “new modes of social ordering” these produce.
- Response: a theoretical and policy stance toward algorithmic governance, characterized by its alignment to the system it would change. The three responses overlap and do not form a sequence.
- Bidirectionality: access, transparency and refusal each run both ways between governed and governing.
- Refusal: “an informed practice of ‘talking back’” (Garcia et al.), paired with a commitment to alternatives. Its two implications are the “distributed capacity to act on an objection” and its social iteration.
Connections
It sets itself beside Pasquale’s two “waves” of algorithmic accountability (improve the system, or ask whether it is needed), and builds on Dencik and colleagues’ data justice, Benjamin’s abolitionist tools, Couldry and Mejias’s data colonialism, Eubanks, Costanza-Chock’s design justice, and Barabas’s refusal by computational practitioners. It does not cite Scott, Espeland, or the epistemic-injustice literature.
Bearing on the record
- TERM-064. C2 is the term’s directedness, argued from the governed side. “Society legible to an institution” is LinkNYC’s access “down”. Robo-debt shows transparency demanded of the observed and withheld by the observer. The paper adds that the two directions can be traded against each other: a remedy offered as legibility of the institution can be delivered as legibility of its subjects. Its line on transparency through “classification, measurement, evaluation and rank” is the term’s regime in other words. A25’s “counterweight” is accurate: the paper asks whether one should be inside the regime at all. Its answer is not “never”. It is that the choice to be inside should be open, and that it often is not.
- CLAIM-403. The paper supports the claim’s first point from outside: admission to a system (“access and inclusion”) can widen while nothing else changes. It adds what the claim does not say: admission can itself be the harm, by making people “more accurately identifiable”. It has no ladder of recognition, credibility and uptake. Its refusal, together with design justice’s “the possibility to say no”, sits near the claim’s top rung, a part in governing the standards, but it is asserted rather than analysed.
- CLAIM-349 separates access, authority and action for an organization’s own signals. C3 is the governed’s version: being informed is not being able to act on it.
- CLAIM-348. “horizontal neutrality … treating everybody the same” is the paper’s form of a description adequate for administration that merges what members need told apart.
- THEORY-857 holds C2.
No instruction for machine-learning practice; nothing for the anthology.
Limitations
- No evidence of comparative outcomes. The claim that refusal “holds more promise” rests on argument and selected cases. No case is shown in which access or transparency failed where refusal succeeded, or the reverse. For blocksidewalk the paper itself reports that Sidewalk Labs withdrew “citing pandemic related business concerns”, so the campaign’s part in the outcome is not shown.
- The responses are characterized at their weakest. Transparency is discussed mostly in its procedural forms (notice, rights to explanation, impact assessments). Contestation built into transparency regimes is mentioned only as what they lack.
- Refusal’s operationalization is left open. The authors say so: they “offer suggestions” and note the risk of co-optation.
- Scope. The case material is Australian, UK, US and Canadian public services and smart cities.
Open questions
- When does a remedy offered as making the institution legible to the governed become, in practice, making the governed legible to the institution? The paper gives instances and no conditions.
- What does refusal require of an institution’s procedures, rather than of its opponents? That would connect it to CLAIM-403’s top rung.