Risks & safety
AI Ethics
Also known as: responsible AI
The study and practice of building and using AI in ways that are fair, transparent, accountable and respectful of people's rights.
AI ethics is about using AI in a way that is fair and does right by people. It asks questions like: Is this system biased against certain groups? Is it clear how it makes decisions? Who is responsible when it gets something wrong? Does it respect privacy? As AI is used in more decisions that affect real lives, getting these answers right is as important as making the technology work.
AI ethics is the field concerned with the responsible development and use of AI. It overlaps with technical AI safety but focuses on human, social and moral questions.
Common principles
Most ethics guidelines converge on a handful of themes: - Fairness: avoid unjust bias and discrimination. - Transparency: be clear about how systems work and when AI is used. - Accountability: someone is responsible for outcomes. - Privacy: protect people's data. - Human oversight and autonomy: keep meaningful human control.
From principles to practice
The hard part is turning broad principles into concrete engineering and policy, testing for bias, enabling explainability, labelling synthetic media (deepfakes), and complying with law such as the EU AI Act.
AI ethics examines the normative dimensions of AI, fairness, transparency, accountability, privacy, autonomy, safety and societal impact, and their translation into practice ("responsible AI"). A cross-analysis of global guidelines found broad convergence on a core set of principles but substantial divergence in interpretation and implementation (Jobin et al. 2019), highlighting the persistent gap between high-level principles and operational engineering/governance.
Concrete tensions include incompatible formal fairness criteria (forcing explicit value choices; see bias), transparency versus IP/security, and utility versus privacy. Operationalisation draws on impact assessments, disaggregated evaluation, explainability, documentation (model/data cards), provenance for synthetic media, and human-oversight requirements, increasingly codified in regulation (EU AI Act, sectoral rules) and standards (NIST AI RMF, ISO/IEC 42001). It is distinct from but interdependent with technical alignment and safety: ethics sets the objectives and constraints; safety and governance supply mechanisms to meet them.