Algorithm Review Boards

Description

An algorithm review board, also called an AI ethics board or AI review board, is a standing group from different disciplines that examines proposed AI systems or AI research projects before they proceed. It differs from an Institutional Review Board (IRB), which protects the human subjects of research, because it also weighs harms to the people a system will affect and to society. Many boards first sort projects by risk, and for higher-risk work the project team submits a written account of the system's purpose, data, foreseeable harms and planned mitigations, which members with technical, legal, ethical and domain backgrounds question and iterate on with the team. Where it has the power to decide, the board approves, approves with conditions such as a monitoring plan or a re-review date, or rejects; where it only advises, it recommends one of these. It keeps a dated record of its decision, reasons and conditions that later reviews and audits can check against what was built, and some organisations share that record with an oversight body or publish it.

Example Use Cases

General

A hospital group's algorithm review board examines a sepsis early-warning model before a ward pilot. It approves the pilot on condition of a monitoring plan for alert rates and missed cases, names who may switch the model off, and sets a re-review once six months of local performance data are in, before any wider roll-out.

A bank's board reviews a proposed credit-limit model before launch. It asks the team for its disparate-impact analysis by protected group and for evidence that applicants can contest a decision, and withholds approval until both are supplied, so the decision to deploy rests with a body outside the development team.

A university AI research programme releases grant funding only after each project has passed an ethics and society review, in which researchers state the risks their work poses to society and to particular groups, commit to mitigations, and revise them with an interdisciplinary faculty panel.

Transparency

A government department publishes the board's decision record for a benefits-triage tool, giving the reasons for approval, the conditions attached and the re-review date, so that its oversight committee and the public can see what was examined, by whom and when, and check later that the conditions were met.

Limitations

Review adds time and cost to every project that goes through it, and organisations under commercial pressure may route work around the board or narrow its remit, so the board's coverage, not just its existence, needs to be shown.
Its judgements are only as good as its membership. A board without the relevant technical, domain or community expertise will miss harms that fall outside what its members know, and a board drawn mainly from inside the team's own reporting line is open to group-think.
Independence is hard to keep inside an organisation. Practitioners report tension between profit and the cost of responsible-AI review, and a board can come under pressure to approve projects that leadership has already committed to, which weakens it most where review matters most.
Without authority to stop or delay a project, and a way to check that its conditions are met, a board's recommendations can be ignored. Evidence from a board is strong only when its decision record shows what it could block and what happened to its conditions.
There is no agreed measure of whether a board works, and review can become box-ticking. A record that a project was reviewed shows that a process ran, not that the review was substantive or that its concerns were addressed.

Resources

Research Papers

How to design an AI ethics board

Jonas Schuett, Ann-Katrin Reuel, and Alexis Carlier•Apr 1, 2025

The development and deployment of artificial intelligence (AI) systems poses significant risks to society. To reduce these risks to an acceptable level, AI companies need an effective risk management process and sound risk governance. In this paper, we explore a particular way in which AI companies can improve their risk governance: by setting up an AI ethics board. We identify five key design choices: (1) What responsibilities should the board have? (2) What should its legal structure be? (3) Who should sit on the board? (4) How should it make decisions? (5) And what resources does it need? We break each of these questions down into more specific sub-questions, list options, and discuss how different design choices affect the board’s ability to reduce societal risks from AI. Several failures have shown that designing an AI ethics board can be challenging. This paper provides a toolbox that can help AI companies to overcome these challenges.

Designing Artificial Intelligence Review Boards: Creating Risk Metrics for Review of AI

Sara R. Jordan•Nov 1, 2019

Discussions of the need for a review committee dedicated to ethical oversight of artificial intelligence (AI) research have not yet advanced to organization design discussions. What are the essential components for an AI research review board? Based upon precedent for composition and conduct of high-technology review boards, one essential component is a risk-adjusted review mechanism that assists review boards in the categorization of research needing review. In this paper, risk assessment tools are developed based upon lessons learned from other review boards, such as Institutional Biosafety Committees.

Ethics and society review: Ethics reflection as a precondition to research funding

Michael S. Bernstein et al.•Dec 21, 2021

Researchers in areas as diverse as computer science and political science must increasingly navigate the possible risks of their research to society. However, the history of medical experiments on vulnerable individuals influenced many research ethics reviews to focus exclusively on risks to human subjects rather than risks to human society. We describe an Ethics and Society Review board (ESR), which fills this moral gap by facilitating ethical and societal reflection as a requirement to access grant funding: Researchers cannot receive grant funding from participating programs until the researchers complete the ESR process for their proposal. Researchers author an initial statement describing their proposed research’s risks to society, subgroups within society, and globally and commit to mitigation strategies for these risks. An interdisciplinary faculty panel iterates with the researchers to refine these risks and mitigation strategies. We describe a mixed-method evaluation of the ESR over 1 y, in partnership with an artificial intelligence grant program run by Stanford HAI. Surveys and interviews of researchers who interacted with the ESR found 100% (95% CI: 87 to 100%) were willing to continue submitting future projects to the ESR, and 58% (95% CI: 37 to 77%) felt that it had influenced the design of their research project. The ESR panel most commonly identified issues of harms to minority groups, inclusion of diverse stakeholders in the research plan, dual use, and representation in datasets. These principles, paired with possible mitigation strategies, offer scaffolding for future research designs.

Tutorials

ESR Process & Toolkit — Stanford Ethics & Society Review

The Ethics and Society Review is improving ethical reflection in research by bringing together a panel of ethics experts to help researchers identify and mitigate negative ethical or societal aspects of their research and elevate the positive.

Documentations

Looking before we leap: Expanding ethical review processes for AI and data science research

Dec 13, 2022

Expanding ethical review processes for AI and data science research

Related Techniques

Name	Description	Assurance Goals
Human Rights, Democracy and the Rule of Law Risk and Impact Assessment	HUDERIA is the Council of Europe's methodology for identifying and assessing the risks an AI system poses to human rights, democracy and the rule of law, and for planning how to prevent or reduce them, from design through deployment to retirement. It is non-binding guidance that public and private organisations may use or adapt in whole or in part; it was drawn up to support risk assessment in line with the Council of Europe's Framework Convention on Artificial Intelligence, and the human rights treaties its resources cite are illustrative, binding only the states that are parties to them. A multidisciplinary team first runs a context-based risk analysis (COBRA): it scopes the system's purpose, data and affected groups, works through risk factors in the system's application, design and deployment contexts, maps potential impacts on specific rights and on democratic and legal processes, gives each impact an initial rating, and triages the system, deciding whether further assessment is needed and whether the system should be built or deployed at all. Impacts are rated for severity, judged by scale (how serious the harm is), scope (how many people, which groups and for how long) and reversibility (whether those affected can be restored to their earlier position), and for probability; for systems that may pose significant risks, the team may engage affected stakeholders and then re-examines and re-rates each impact in a fuller risk and impact assessment. The output is a mitigation plan that prefers avoiding a harm to reducing, repairing or compensating for it, names who is responsible for each measure, outlines where appropriate the remedies open to affected people, and schedules repeat reviews as the system and its context change.	General
Human-in-the-Loop Safeguards	Human-in-the-loop safeguards hold an AI system's output at a checkpoint until a trained person has reviewed it. Designers set the rules that send a case to review, such as low model confidence (the system's own estimate of how likely it is to be right), a high-stakes outcome or a request to appeal. Some schemes also review a random sample of routine cases to audit the rest. The reviewer approves, changes or rejects the output, and the record of each review shows how often people overrode the system. Review protects only where reviewers keep using independent judgement, so the design also has to guard against over-trust and fatigue.	General Safety Transparency
UK Algorithmic Transparency Recording Standard Record	An Algorithmic Transparency Recording Standard (ATRS) record is a public account that a UK public sector organisation writes about one algorithmic tool it uses, on the UK government's standard template, and publishes in a central repository on GOV.UK. The template has two tiers: a short plain-language summary for the general public (what the tool is, why it is used and a contact), and eight further sheets for specialist readers such as journalists, researchers and civil society, covering who is accountable for the tool and which suppliers were involved, how it works and what it replaced, how its output feeds into decisions, when a person reviews that output, how members of the public can appeal, the system's and each model's inputs, outputs, architecture and performance (including any bias testing), the data used, and the impact assessments, risks and mitigations. Information that would be exempt under the Freedom of Information Act can be left out or summarised field by field, with the reason stated, and the record is cleared by the tool's team, its senior responsible owner (the role accountable for the tool) and the press office, checked for readability by the central ATRS team, and updated when the tool changes substantially or is retired. Publishing is mandatory for central government departments and for arm's-length bodies that deliver public or frontline services or deal directly with the public, for tools in beta, pilot or production that significantly influence decisions affecting the public or interact with it directly; other public bodies may publish voluntarily. Unlike a model card, written by a developer for people choosing a model, or a supplier's AI FactSheet for buyers, an ATRS record is written by the deploying public body for the public and describes the decision process, human review and appeal routes around the tool as well as the tool.	Transparency