What is responsible AI?
By Nick Jang, founder of MINERVA Studio ·
The short answer
Responsible AI is the practice of designing, building and using AI so that it benefits people and avoids harm: it’s fair, reliable and safe, respects privacy, can be explained, and someone is accountable for it. The most widely adopted statement of it is the OECD AI Principles, adopted in 2019, updated in May 2024 and backed by 47 countries, Canada among them. Principles are the easy part. Responsible AI shows in the practices behind them: inventories, risk assessments, testing, approvals, monitoring and records.
The principles people mean
Most lists say much the same thing in different words. The OECD’s five values-based principles:
| Principle | In plain terms |
|---|---|
| Inclusive growth, sustainable development and well-being | AI should benefit people and the planet. |
| Human rights and democratic values, including fairness and privacy | AI should respect the law, rights, fairness and privacy, with safeguards such as human oversight. |
| Transparency and explainability | People should know when they’re dealing with AI and be able to understand and challenge its results. |
| Robustness, security and safety | AI should work reliably and safely throughout its life, and be stopped or fixed when it doesn’t. |
| Accountability | The people and organisations behind AI are responsible for it working properly. |
Company versions are similar. Microsoft’s six, for example, are fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. The Government of Canada has its own guiding principles for the federal public service’s use of AI.
From principles to practice
A principle only matters if something changes because of it. A useful test is to ask, for each one, what you’d show someone who doubted it:
| Principle | What you’d show |
|---|---|
| Fairness | Bias test results by group, the fairness measure chosen and who chose it. |
| Reliability and safety | Validation results, monitoring thresholds, and what happened when a threshold was crossed. |
| Privacy and security | The data each system uses, the privacy assessment, red team findings and their fixes. |
| Transparency | Notices to customers, explanations given for decisions, and the published list of AI uses where required. |
| Accountability | A named owner for each system, and who approved it, when, on what evidence. |
That evidence comes from a small set of practices: an AI use case inventory, an impact assessment for systems that affect people, bias testing, independent validation, monitoring, an acceptable use policy for staff, and approvals by someone other than the builder.
Responsible AI, AI governance and AI ethics
The terms overlap and get used loosely. Roughly: AI ethics asks what’s right; responsible AI is the commitment to act on it; [AI governance](/answers/ai-governance) is the structure that makes it happen and proves it did, with the policies, roles, approvals and records. A company can have a fine responsible AI page and weak governance. A regulator looks at the governance.
Why it matters for banks and insurers
For a Canadian financial institution, responsible AI isn’t optional. OSFI’s E-23 brings AI models under model risk management from 1 May 2027, with expectations on explainability, bias and oversight; Quebec’s AMF guideline on AI expects fair treatment of clients; and privacy law, including Quebec’s Law 25 on automated decisions, applies throughout. Each of them asks for evidence, not a statement of values.
MINERVA is where that evidence is kept: every model and AI system with its owner, rating, tests, findings and approvals, decided by a person other than its builder and recorded so it can be checked. See AI governance software.
Related questions
What are the principles of responsible AI?
Commonly: fairness, reliability and safety, privacy and security, transparency and explainability, and accountability. The OECD’s five principles add inclusive growth and well-being, and respect for human rights and democratic values.
What is the difference between responsible AI and AI governance?
Responsible AI is the commitment to use AI in a fair, safe, transparent and accountable way. AI governance is the set of policies, roles, approvals and records that puts it into practice and proves it.
Has Canada adopted the OECD AI Principles?
Yes. Canada is one of the 47 adherents to the OECD AI Principles, which were adopted in May 2019 and updated in May 2024.
What is an example of responsible AI in practice?
A lender testing a credit model for bias across groups before launch, having it validated and approved by people independent of its developers, explaining declines to applicants, and monitoring the model’s outcomes afterwards.
Is responsible AI required by law?
Not as such, but parts of it are: anti-discrimination and privacy law apply to AI decisions, and financial regulators such as OSFI and the AMF expect fairness, explainability and oversight for AI models.
Sources
- OECD — AI Principles (adopted 2019, updated 2024)
- Microsoft — Responsible AI principles
- Government of Canada — Responsible use of artificial intelligence in government
New answers are posted on MINERVA on LinkedIn. Follow along, or send the question you’d like answered next to minerva@minerva.studio.
A plain-language explanation, not legal advice. The regulator’s own text is the source; read it, and take advice on how it applies to your institution.