Algorithmic accountability turns on decisions people can challenge
Published on 10 November 2026 at 00:44 GMT
By Editorial Team SDG16
An automated decision can affect a person without that person understanding how the system contributed to it. AlgorithmWatch, a non-profit organisation based in Berlin and Zurich, investigates algorithms and artificial intelligence in relation to rights, democracy and sustainability. Its work places accountability within the decisions people encounter, rather than treating a system’s technical performance as a complete account of its public effects.
The organisation describes an approach combining scientific studies, investigative journalism and policy analysis with advocacy. Its areas include fundamental rights, public discourse and the distribution of power around technology. These are its stated priorities. The existence of an investigation does not establish that every automated system causes the same harm or that every recommendation follows from an identical set of findings.
AlgorithmWatch argues for transparency, oversight and routes to justice where automated systems affect significant decisions. These are attributed positions held by the organisation. They raise questions about how people can understand and contest a decision, while the legal obligations of any particular provider or public authority depend on the applicable rules and circumstances.
Explanation and remedy are different functions
A description of a model can disclose how a system is intended to operate. An explanation of an individual decision concerns what happened in a particular case. A route to challenge that decision concerns the institution able to reconsider it. These forms of transparency and accountability can be related without one automatically supplying the others.
A published document, for example, can provide information while leaving a person uncertain about the procedure for correction. A complaint channel can exist while its scope and decision-making authority remain separate questions. The practical account concerns what information is available, who can act on it and what happens when someone disputes an outcome.
The United Nations Educational, Scientific and Cultural Organization (UNESCO) adopted its Recommendation on the Ethics of Artificial Intelligence in November 2021. It sets out an ethical framework including human rights, fairness, transparency and human oversight. It is a recommendation rather than a directly enforceable court remedy for every individual decision involving a system.
The European Commission describes the European Union AI Act as a risk-based regulatory framework. Its phased application means that entry into force and the application of particular provisions are different dates. That distinction matters when discussing accountability in 2026: the existence of the regulation does not mean that every provision applies identically to every system or use.
The organisation using a tool remains part of the decision
A developer and a deploying institution can have different roles. A system may produce a score while a public body or employer uses it within a wider process. Understanding that process involves the rule applied to the score, the information available and the authority to make or revise a decision. Technical documentation alone does not describe every institutional choice.
Aggregate performance also differs from an individual result. A measure across a dataset can describe the cases included in that evaluation. It does not establish that every affected person receives a correct outcome, or that performance is identical outside the assessed setting. A claim about fairness or accuracy consequently concerns the population, method and comparison being used.
AlgorithmWatch links its research to policy advocacy. That relationship makes attribution important: a finding about a system, an interpretation of its consequences and a demand for legal change are different statements. Reporting the organisation’s position does not require presenting its recommendation as an independent verdict or treating a campaign objective as an existing legal duty.
The connection to SDG 16 (peace, justice and strong institutions) concerns accountable decision-making and access to justice. The goal provides a frame for institutional questions without implying that every use of an algorithm is a public decision or that a single transparency measure resolves all questions of accountability.
A correction to data and a reversal of a decision are also different events. An institution can acknowledge that information was inaccurate while a separate process determines the consequence for an outcome. A route to contest automated decision-making consequently involves more than access to a technical description: it concerns the specific authority able to act on a complaint and the scope of that authority’s decision.
The work described by AlgorithmWatch follows the relationship between a technical system and the institutions using it. Its public-interest focus concerns what happens when a person seeks to understand or contest an outcome. A model disclosed, a decision explained and a remedy available describe different points in that process, each requiring evidence about the actual arrangements rather than the promise of accountable technology alone.
Written by a human author, edited with AI assistance.
Further information:
• AlgorithmWatch organisational overview. Establishes its non-profit status, locations and public-interest remit.
• AlgorithmWatch approach. Provides its research methods and attributed accountability positions.
• UNESCO ethics recommendation. Provides the original 2021 ethical framework.
• European Commission AI Act overview. Explains the risk-based framework and phased application.
• United Nations Goal 16 report. Provides the accountable institutions and justice context.
