Efficiency for Whom? Human-Rights Tests for Britain’s Automated Public Services

Automated decision-making arrives in public services dressed as common sense. A council wants to triage housing applications faster. The Department for Work and Pensions wants to spot fraud sooner. Police forces want to allocate resources more efficiently. The Home Office wants to screen immigration cases at scale. In each case, the promise is the same: better outcomes, less waste, more innovation.

The language is seductive because it is not obviously political. Who opposes efficiency? Who opposes innovation? But the Oxford Amnesty Lectures archive offers a necessary corrective. Its digital-rights lectures return to a simple claim: a technical system is never neutral. It encodes choices about who is risky, who is credible, who deserves help and who can be ignored. Those choices are political, and they are human-rights questions.

The efficiency trap

Efficiency is not a neutral measure. It depends on what you count. A benefits algorithm that flags more claimants as suspicious may look efficient if it reduces payments. It looks very different if it generates errors, debt, destitution and appeals. A policing tool that predicts where crime will happen may look efficient if it sends officers to certain neighbourhoods. It looks different if it deepens over-policing and erodes trust.

The archive’s lectures on data and democracy make a further point: automation often moves discretion rather than removing it. A human decision-maker may be replaced by a score, a threshold or a risk category. The discretion does not disappear; it migrates into design, procurement, training and the interpretation of outputs. That is why human-rights scrutiny must begin before a system is bought, not after it has harmed someone.

Three places where the stakes are highest

Benefits

In social security, automated tools are used for fraud detection, debt recovery and claim triage. The human-rights stakes include the right to social security, the right to a fair hearing, non-discrimination and data protection. The DWP has a legal duty to get decisions right, not merely to process them quickly. When an algorithm generates a suspicion, the claimant should not bear the burden of proving the system wrong without clear information and a real right of appeal.

Policing

Predictive policing, facial recognition and risk-assessment tools raise questions under Articles 8, 10 and 11 of the European Convention on Human Rights, as well as the Public Sector Equality Duty. The archive’s lectures on surveillance stress that proportionality is not a box-ticking exercise. It requires evidence that a measure is necessary, that less intrusive alternatives have been considered, and that affected communities have been heard.

Immigration

The Home Office uses digital tools for identity checks, asylum screening, age assessment and case prioritisation. Here the consequences can include detention, removal or refusal of protection. The right to seek asylum, the prohibition of refoulement and the right to an effective remedy all require that automated outputs are explainable, contestable and subject to meaningful human review.

What the digital-rights lectures add

The Oxford Amnesty Lectures archive does not offer a single template for algorithmic accountability. It offers something more useful: a set of habits. Ask who designed the system and for what purpose. Ask what data it uses and what data it ignores. Ask who is most likely to be misclassified. Ask where a human being can intervene. Ask what remedy exists when the system is wrong.

These habits challenge the language of innovation. Innovation is not a human-rights justification in itself. A new system can be innovative and unlawful. It can be efficient and discriminatory. It can be modern and opaque. The relevant question is not whether a tool is new, but whether it is lawful, necessary, proportionate and fair.

Seven human-rights tests for 2026 public-sector algorithms

1. Lawful basis and necessity. Has the public body identified a legal power for the automated process? Can it show that the tool is necessary and proportionate, not merely convenient?
2. Equality before the algorithm. Has an equality impact assessment been published? Does the system produce different outcomes for disabled people, ethnic minorities, women, children or people in poverty?
3. Explanation that means something. Can an affected person understand why a decision was reached? A generic statement that “the system” made a decision is not an explanation.
4. A real human review. Is there a named official with the authority and time to overturn an automated output? Human review is meaningless if it is a rubber stamp.
5. Data minimisation and provenance. What data is used? Is it accurate, relevant and up to date? Are sensitive categories being inferred? How long is data retained?
6. Contestability and remedies. Can a person challenge the decision quickly, affordably and with legal aid where necessary? Is there an appeal route that does not depend on understanding the algorithm’s code?
7. Transparency by default. Has the body published a Data Protection Impact Assessment, an equality assessment and, where applicable, an Algorithmic Transparency Recording Standard entry? Procurement contracts should not be secret simply because they involve software.

How to respond to local consultations

Local consultations are where these tests can be applied before a system becomes embedded. Start by identifying the public body: a council, a police force, an NHS trust, a combined authority or a central government department. Then ask for the documents that should already exist.

• Request the Data Protection Impact Assessment and the equality impact assessment. If they have not been done, say so publicly.
• Ask whether the system appears in the Algorithmic Transparency Recording Standard. If not, ask why.
• Submit a Freedom of Information request for procurement documents, pilot evaluations and error rates.
• Ask councillors and scrutiny committees to call the responsible officer to give evidence.
• Raise the issue with your MP, especially where a national system is being piloted locally.
• Work with affected groups: claimants, tenants, disabled people’s organisations, racial justice groups and advice agencies. Their testimony should shape the questions, not decorate them.

Consultations are often designed to elicit support for a decision already taken. Human-rights scrutiny can change that. It can force publication, delay procurement, secure a pilot evaluation or win a commitment to human review. The goal is not to oppose every use of technology. It is to insist that public-sector automation is judged by the rights it protects, not only by the money it saves.

The algorithm is not neutral. Neither is the question of who gets to decide what it does.