Principles of responsible AI: FIPs, OECD AI Principles, ethics by design, and trustworthy AI
Every AI framework in this guide is built from a small shared set of principles. This topic traces where those principles come from, and pins down the three that are easiest to confuse: transparency, explainability and accountability.
Why this matters for the exam
Every AI framework in this guide is built from a small shared set of principles. This topic traces where those principles come from, and pins down the three that are easiest to confuse: transparency, explainability and accountability.
What you need to know
Responsible-AI principles are built on decades-old privacy principles. This topic traces the lineage, from the 1980 fair-information practices to the OECD's AI-specific set, and pins down the vocabulary that every framework shares.
The bedrock: Fair Information Practices (FIPs)
Long before AI, the 1980 OECD Guidelines established eight Fair Information Practices that still underpin privacy law, and by extension most AI governance principle sets built on top of them:
- Collection limitation — collect only what is necessary, lawfully and fairly.
- Use limitation — limit to specified uses absent consent or legal exception.
- Security safeguards
- Notice or openness
- Access or individual participation — people can see, correct or challenge data about them.
- Accountability
- Purpose specification
- Data quality and relevance
The OECD AI Principles
The OECD returned to these questions in the AI era with five principles written for AI systems specifically. They are a different list from the FIPs, though the two share vocabulary:
- Inclusive growth, sustainable development and well-being — trustworthy AI should advance these goals and encourage responsible stewardship.
- Human rights and democratic values, including fairness and privacy — respect the rule of law and diversity, with real safeguards for fairness and justice.
- Transparency and explainability — people should understand when they are engaging with AI and be able to challenge outcomes.
- Robustness, security and safety — function safely throughout the system's lifetime, continually assessing risk via traceability and a risk-management approach.
- Accountability — organizations and individuals who develop, deploy or operate AI are held responsible for its proper functioning.
Shared principles, consistent meanings
Across FIPs, the OECD AI Principles, and the many frameworks built on them, several principles recur with consistent meanings:
| Principle | Core question it answers | What it actually requires |
|---|---|---|
| Transparency | Do people know AI is involved, and roughly how it works? | Clear, easy-to-understand disclosure about a system's existence, development and deployment; making people aware they are interacting with AI. |
| Explainability | Why did the system produce this specific output? | "The capacity to describe an AI system and its expected impact and potential biases" — requires understanding how the system operates and what data trained it; supports challenging one particular decision. |
| Accountability | Who answers for the outcome? | Organizations and individuals are responsible for a system's proper functioning, irrespective of how many parties contributed to building it. |
| Fairness | Are outcomes equitable across groups? | Minimizing or eliminating bias that leads to unfair treatment. |
| Human-centricity | Does the system serve human agency rather than replace or control it? | AI that amplifies human agency and has a positive impact on the human condition. |
| Safety / robustness / reliability | Does it function correctly and resist failure over time? | Robust, secure and safe operation throughout the system's lifetime, with continual risk assessment. |
| Privacy and security | Is personal data protected and is the system resistant to intrusion? | Protective data practices plus strong defenses against exfiltration and model poisoning. |
Explainability is often used interchangeably with interpretability, but NIST distinguishes the two: explainability describes how the output was produced, while interpretability describes what the output means in context.
Three of these principles are especially easy to confuse. Transparency is about disclosure and general awareness. Explainability is about justifying one specific output after the fact. Accountability is about who is answerable. A system that can provide a rationale for a specific output is exhibiting explainability.
The scan-reading system from an earlier topic shows all three in one setting. The hospital tells patients that an algorithm helps read their scans: transparency. When a clinician asks why one scan was flagged, the system shows the factors behind that flag: explainability. And when a flag turns out to be wrong, the hospital answers for the outcome: accountability.
Ethics by design
Principles only matter if they shape how systems get built, which is where ethics by design comes in. Modeled on privacy by design, it holds that ethical considerations should be addressed at the outset, and re-evaluated at deployment, since risk profiles shift. Its six core principles:
- Respect for human agency
- Privacy and data governance
- Fairness
- Individual, social and environmental well-being
- Transparency
- Accountability and oversight
Trustworthy AI
The course wraps four of these principles into a single label. Trustworthy AI is AI that is human-centric, accountable, transparent and explainable, and privacy-enhanced. The label adds no new principle. The first three terms mean exactly what they meant above, and privacy-enhanced matches the privacy and security row, so there is no fifth framework to learn here. The bundle does add two details: accountable, in this setting, includes ensuring systems are safe, secure, resilient, valid, reliable and fair; and privacy-enhanced relies on privacy-enhancing technologies to guard against both intentional and accidental misuse.
- Human-centricity
- Education and literacy
- Privacy protection
- Security
- Fair competition
- "FAT": fairness, accountability, transparency
- Innovation
Next up: how AI systems fail, and who gets hurt when they do.
Remember
- Transparency is disclosure and awareness. Explainability is justifying one specific output. Accountability is who answers for the outcome. They are three different things.
- The five OECD AI Principles: inclusive growth and well-being; human rights and democratic values; transparency and explainability; robustness, security and safety; accountability.
- The eight FIPs (1980): collection limitation, use limitation, security safeguards, notice/openness, access/participation, accountability, purpose specification, data quality/relevance. A different list from the OECD AI Principles.
- Ethics by design has to be re-evaluated at deployment, since risk profiles shift. The design phase cannot anticipate every way real users will interact with a system.
- Trustworthy AI has four characteristics: human-centric, accountable, transparent and explainable, and privacy-enhanced.
Practise this topic
Domain I is free in the app, including its practice questions and flashcards, with progress tracking and no card details.
Previous: Why AI needs its own governance, and how to describe a system: the OECD's five dimensions
Next: How AI fails and who gets hurt: failure modes and the five levels of harm
Back to the AIGP study guide.
AI Governance Study is an independent study aid. It does not represent a government entity: it is not affiliated with, endorsed by or authorised by any government, government agency or regulatory authority, and it does not provide government services or legal advice. Laws and frameworks are described in our own words — the official texts are listed at official sources. It is also not affiliated with, endorsed by, or sponsored by the IAPP. The AIGP name is used only to identify the exam this material helps you prepare for.