UNESCO Recommendation on the Ethics of Artificial Intelligence

What it is

The Recommendation on the Ethics of Artificial Intelligence was adopted by UNESCO’s General Conference at its 41st session in November 2021, with the support of all 193 member states. It was the first global standard-setting instrument on the ethics of AI.

It is a broad document. It sets the values and principles that should guide AI, then translates them into concrete policy action areas for governments. UNESCO supports implementation through two tools, the Readiness Assessment Methodology (RAM) for countries and the Ethical Impact Assessment (EIA) for specific AI systems, and through its Global AI Ethics and Governance Observatory.

Who it applies to

The Recommendation is addressed to member states, which are asked to apply it through laws, policies and institutions. It also speaks to all AI actors, including companies, researchers and civil society, across every stage of the AI lifecycle. It is not binding on anyone, but it shapes national AI strategies and public procurement, especially in countries that have run the RAM with UNESCO.

What it requires

Four core values

• Human rights and human dignity
• Living in peaceful, just and interconnected societies
• Ensuring diversity and inclusiveness
• Environment and ecosystem flourishing

Ten principles

1. Proportionality and do no harm. AI use should not go beyond what is needed to achieve a legitimate aim, and risks should be assessed.
2. Safety and security. Unwanted harms and vulnerabilities to attack should be avoided and addressed.
3. Right to privacy and data protection, protected throughout the AI lifecycle.
4. Multi-stakeholder and adaptive governance and collaboration.
5. Responsibility and accountability, with oversight, impact assessment, audit and due diligence mechanisms.
6. Transparency and explainability, appropriate to the context.
7. Human oversight and determination. AI systems should not displace ultimate human responsibility.
8. Sustainability, assessing AI’s effect on environmental and social goals.
9. Awareness and literacy, so the public understands AI and data.
10. Fairness and non-discrimination.

Eleven policy areas

The Recommendation sets out policy action in eleven areas. These are ethical impact assessment, ethical governance and stewardship, data policy, development and international cooperation, environment and ecosystems, gender, culture, education and research, communication and information, economy and labour, and health and social well-being.

The implementation tools

The Readiness Assessment Methodology is a country-level tool. It helps governments understand how prepared they are to govern AI ethically and where institutions, rules or capacity need strengthening.

The Ethical Impact Assessment works at the level of a single AI system. It has two aims, checking whether a specific system aligns with the Recommendation’s values and principles, and improving transparency by making information about the system available. It follows the AI lifecycle with requirements before and after deployment, and opens with scoping questions such as whether automation is the right solution for the problem. It is aimed especially at public bodies procuring AI, but the questions apply to any organisation.

Using it in practice

Organisations rarely adopt the Recommendation wholesale. They use it in three ways. It serves as a values reference when writing an AI ethics or acceptable use policy, to check that the policy addresses proportionality, human oversight, fairness, privacy and sustainability. It works as a procurement lens, asking vendors the kinds of questions the Ethical Impact Assessment asks, such as what data a system was trained on, who could be harmed and how decisions can be challenged. It also acts as a check on impact assessments, making sure environmental effects, gender and effects on vulnerable groups are considered alongside the more familiar privacy and security risks. Public bodies in countries that have worked with UNESCO on the RAM may also find its language in national AI policy.

Key dates

The Recommendation was adopted in November 2021. UNESCO published the Readiness Assessment Methodology and the Ethical Impact Assessment in 2023, and supports their use through technical assistance to member states.

How it relates to other frameworks

UNESCO’s Recommendation covers much the same ethical ground as the OECD AI Principles but extends further into social policy, with explicit attention to gender, culture, the environment and labour. Its Ethical Impact Assessment is one of several system-level impact tools, alongside the Council of Europe’s HUDERIA methodology and the international standard ISO/IEC 42005. Organisations that run impact assessments under one of those can usually map their work to the EIA’s questions with little extra effort.

For most organisations, the Recommendation is best used as a values reference for AI policy and procurement, with operational detail drawn from a management system standard or risk framework.