AI ethics asks what AI systems should and should not do, and who is responsible when they cause harm. It covers the choices made at every stage: which data is collected and how, what a model is built to optimize, where it is deployed, who can contest its decisions, and what happens to the people it affects.
A shared vocabulary
Many organizations have published AI ethics principles. When Jobin, Ienca and Vayena mapped these guidelines in 2019, they found convergence around five themes: transparency, justice and fairness, non-maleficence (avoiding harm), responsibility, and privacy. They also found wide disagreement about what those words mean in practice and how to implement them.
International bodies have since set common reference points. The OECD AI Principles were adopted in 2019 and updated in 2024. In November 2021, UNESCO’s 193 member states adopted a Recommendation on the Ethics of Artificial Intelligence.
From principles to practice
Principles are only useful once they turn into specific work. Some of that work has its own entries in this guide:
- Fairness and bias in AI: measuring whether a system treats groups differently, and tracing why.
- Explainability: giving reasons a person can inspect and challenge.
- Algorithmic accountability: clear ownership, audit trails and routes for redress.
- Privacy: limiting what data is collected and what a model can reveal about it.
How it relates to safety and governance
The boundaries blur, but a rough split helps. AI ethics asks what is right and whom a system serves. AI safety focuses on preventing systems from behaving in harmful or unintended ways, including as they grow more capable. AI governance covers the laws, standards and oversight that enforce both.