The Global Risks Forum (GRF)
Governments use algorithms to assess benefit claims, detect fraud, allocate inspections, set risk scores in policing and migration, and support judges and caseworkers. These decisions affect rights and livelihoods at scale, and errors repeat across thousands of cases before anyone sees them. This page sets out how algorithmic public decisions go wrong and how safeguards work, the records that show them, and who decides whether and how a system is used.
GRF · In development · service level R0
Algorithmic public decisions are decisions by public bodies that are made or shaped by automated systems, in areas such as welfare, tax, policing, migration and justice. Their risks are scaled errors, bias, nominal human review and opacity. Ministers, data protection authorities, ombudsmen and courts hold the decisions.
For algorithmic public decisions, in GRF's situational-awareness reading, the map opens on population density and night lights with night lights tonight from NASA GIBS, framed on East Asia; 4 further NASA layers wait in the list.
The interactive map needs JavaScript; the layer list and sources are below.
Layers 2
Connected activity, for this page
Keyboard, with the map focused
Hold Ctrl or ⌘ and scroll to zoom
Latest available
NASA GIBS / EOSDIS · ASTER GDEM (METI, NASA) · Blue Marble (NASA) · © OpenStreetMap contributors (ODbL) · Natural Earth
Loading the latest NASA imagery.
Drag to move. Ctrl or ⌘ and scroll, pinch or the buttons to zoom; hover for the values under the pointer; reset returns to the region.
For algorithmic public decisions, in GRF's situational-awareness reading, the map opens on population density and night lights with night lights tonight from NASA GIBS, framed on East Asia; 4 further NASA layers wait in the list.Satellite and model products as NASA publishes them, each with its own date; indicators, not assessments of any place; claim ceiling INTEGRATED.
Layers and dates
The working screen behind this subject: its structure, terms and states. Names are placeholders and values illustrative.
Algorithmic public decisions · Algorithm register · Quarter 1
| Benefit fraud flagging | Social agency | High | Missing | Partial |
| Permit triage | Planning | Limited | Done | Yes |
| School placement | Education | High | Done | Yes |
| Chat assistant | Tax office | Minimal | Done | n/a |
Contestability and transparency. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
A flawed rule or model applies to every case it touches. Without monitoring of outcomes and appeals, a systematic error can wrongly deny or penalise many people before it is detected.
Historical data reflect past decisions and enforcement patterns. Proxies such as postcode or nationality can produce unequal outcomes even when protected characteristics are excluded.
Officials asked to review automated outputs under time pressure tend to accept them. Meaningful review needs time, authority to overrule and information about how the output was produced.
People cannot contest a decision they do not know was automated or cannot understand. Notice, reasons and access to an appeal decide whether errors are corrected.
Each term as this page uses it. Where a standard-setting body defines the term, the wording follows that body.
| Term | Definition |
|---|---|
| Automated decision-making | A decision taken by technological means without meaningful human involvement. |
| Algorithm register | A public list of algorithmic systems used by a government body, with their purpose and safeguards. |
| Automation bias | The tendency of people to over-rely on automated outputs and accept them without adequate scrutiny. |
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Label |
|---|---|---|---|---|
| Public algorithmic systems in a transparency register | count by agency | on update | Algorithm registers | DECLARED |
| Algorithmic impact assessments published | count | annual | Government publications | DECLARED |
| Appeals against automated decisions | count and share upheld | annual | Agency and tribunal records | OFFICIAL |
| Outcome disparity across groups | ratio of adverse outcome rates | per audit | Independent audits | ESTIMATED |
| Systems suspended or withdrawn after review | count | annual | Government and court records | OFFICIAL |
| Ombudsman complaints about automated decisions | count | annual | Ombudsman reports | OFFICIAL |
The topic as a computational graph: named sources feed the methods, an evidence review gates the outputs, and the decisions stay with the authorities that hold them.
Fig. 02Nodes carry their type, two metric rows and a state; connectors show what feeds what; dashed frames name the owners; the timeline is the cadence. Design of the pipeline, not a record of a run.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Source | What it gives | Licence class |
|---|---|---|
| UK Algorithmic Transparency Recording Standard records | Published records of algorithmic tools used by public bodies | Open Government Licence |
| Amsterdam and Helsinki algorithm registers | City registers of algorithmic systems in use | Open data, terms per city |
| Canada Directive on Automated Decision-Making and Algorithmic Impact Assessments | Mandatory impact assessments for federal automated decisions | Open Government Licence, Canada |
| Ombudsman and audit office reports | Investigations of automated decision systems in public services | Public records, terms per office |
| Court judgments on automated public decisions | Rulings on lawfulness of algorithmic systems | Public records, terms per court |
AI and Digital Governance prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Who decides | The question | Evidence this page serves |
|---|---|---|
| Ministers and agency heads | Whether to automate a decision and with what safeguards | Impact assessments, audit evidence and lessons from comparable systems |
| Data protection authorities | Whether processing and automated decisions are lawful | Assessment records and complaints |
| Ombudsmen and audit offices | Whether to investigate a system and what to recommend | Appeals, complaints and outcome data |
| Courts and tribunals | Whether a decision or system breached rights or law | Records presented by parties; the page serves context only |
Algorithms trained on historical data can reproduce historical patterns of disadvantage. A proxy variable that seems neutral, such as postcode, may correlate with protected characteristics. Errors at scale differ from individual errors: a flawed rule applied to an entire caseload harms many people at once and in the same way, which can overwhelm complaint systems.
The burden of proof often shifts silently. When a system flags someone as high-risk, that person may need to prove a negative, without knowing what evidence the system used. Well-documented failures of automated benefits and fraud systems show that the greatest harm comes when automated flags are treated as findings rather than prompts for human investigation.
Listing systems used by public bodies and their purposes. Limits: registers are often incomplete and vary in detail.
Telling people when an algorithm was used and why it produced a result. Limits: explanations of complex models can be partial.
Measuring performance across groups, using machine learning methods. Limits: different fairness measures can conflict, and choosing among them is a value judgement.
Routes to human review, appeal and remedy. Limits: depend on accessible processes and adequate staffing.
The current service level is R0: information and enquiries. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.