The short answer
Transparency is disclosure about the system: that it exists, what it is for, what data it used, who is accountable. Explainability is an account of why this particular output happened. A system can be fully transparent and still not explainable, and the reverse.
Responsible AI, Human Rights and Safety · Starter
- In one sentence
- Being open about where an AI system is used, what it does, what information it uses, and who remains responsible for its effects.
- Why it matters
- Transparency helps people recognise when AI may affect them and what questions to ask before relying on it. If a tool recommends, ranks, drafts, flags, or filters information, people need to know that it is being used, what it is meant to do, and where it may go wrong. That supports safer decisions, clearer accountability, and more meaningful human oversight. It also helps organisations spot unfairness, reduce blind trust in automation, and make it easier to challenge mistakes. Without transparency, people may not know that an AI system influenced an outcome at all, which makes review, remedy, and public trust much harder.
- The everyday version
- In simple terms, transparency means not hiding the use of AI. People should be told that a tool is being used, what it helps with, what its limits are, and who is still responsible.
- The one thing to remember
- Transparency means disclosing where AI is used, what it does and relies on, its important limits, how people can question it, and who remains responsible.
- The commonest misconception
- Transparency means publishing the full model, code, and data to everyone. Not always. Useful transparency depends on audience and risk. Public users may need clear notices and limits, while auditors or regulators may need deeper technical access. Full disclosure can conflict with privacy, confidentiality, security, or intellectual property, so the goal is meaningful information, not maximal exposure.
Read the full page
Responsible AI, Human Rights and Safety · Intermediate
- In one sentence
- Explainability means being able to show, in human-usable terms, why an AI system produced a particular output or pattern of outputs.
- Why it matters
- If an AI system affects people, resources, or risks, a bare output is often not enough. People may need to ask why one case was flagged, why one message was prioritised, or why one applicant received a higher score. Explainability helps users test whether the output fits the evidence, spot errors or bias, and recognise when the system should not be trusted in that context. It also supports accountability: if no one can give a usable reason, it is harder to challenge the result, correct harm, or show that human rights, fairness, and safety were taken seriously. For non-specialists, this matters because explainability shapes whether AI can be used responsibly, especially where decisions need human oversight rather than automatic acceptance.
- The everyday version
- In simple terms, explainability means giving people a usable reason for an AI result so they can check it instead of just accepting it.
- The one thing to remember
- Explainability makes an output or pattern open to human scrutiny. It can support checking and challenge, but a convincing explanation is not proof that the model is correct or fair.
- The commonest misconception
- If a system gives an explanation, the output must be reliable. An explanation can be helpful and still be incomplete, misleading, or too simple. Some explanation methods only estimate which inputs mattered. People should treat explanations as evidence to inspect, not proof that the output is correct.
Read the full page
Responsible AI, Human Rights and Safety · Deeper
- In one sentence
- Interpretability is the extent to which people can understand how a model works or produces outputs, sometimes by inspecting its internal parts.
- Why it matters
- For non-specialists, interpretability matters because an AI system can appear accurate while relying on the wrong reasons. A model may pick up shortcuts, hidden bias or brittle patterns that only become visible when conditions change. If a system affects access, safety, rights or important work, decision-makers need more than a polished answer. They need evidence about whether the system’s behaviour is trustworthy enough for the task, what risks remain, and where human review must stay in place. Interpretability does not solve these problems on its own, but it can make risk assessment, red-teaming, accountability and safer system design more concrete.
- The everyday version
- It is a way of opening the black box a little to see what parts of an AI system seem to drive its answers.
- The one thing to remember
- Interpretability seeks evidence about the internal features and pathways producing a model's output. That evidence can reveal shortcuts or failure modes, but it is rarely a complete account and is not proof of safety.
- The commonest misconception
- If a model is interpretable, it is automatically safe and fair. Interpretability can reveal useful clues about internal behaviour, but it does not prove the system is safe, fair or rights-respecting. A model can be partly interpretable and still produce harmful outputs, and some harmful patterns may remain hidden or be misread. It should support, not replace, testing, impact assessment and human oversight.
Read the full page