2.1. The Six Principles of Responsible AI
💡 First Principle: Responsible AI is a set of design-time commitments, not a final compliance stamp. The six principles — fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability — are guardrails you build in throughout the AI lifecycle, because the failure modes they prevent are baked into how the systems learn and predict.
Why care? On the exam, a large share of the concepts-domain questions are short scenarios — "a hiring model rejects more applicants from one group," "users can't tell a chatbot's answer came from AI" — and your job is to name the principle at stake. Getting these right depends on crisp, non-overlapping definitions, because the wrong answers are almost always other principles that sound plausible. In practice, these principles also determine whether an AI system earns user trust and survives regulatory scrutiny.
The mental model is a checklist applied across the whole lifecycle, from data collection to deployment to monitoring. No single principle is "the security one" or "the legal one" — they overlap and reinforce each other. The table below is your fastest path to telling them apart; the subsections then go one level deeper on each.
| Principle | Core Question It Answers | Classic Scenario |
|---|---|---|
| Fairness | Does it treat all groups of people equitably? | Loan model approves one demographic far more often |
| Reliability & Safety | Does it behave predictably, even in unexpected conditions? | Self-driving system in weather it wasn't trained on |
| Privacy & Security | Is personal data protected and the system secured? | Training data containing customer records |
| Inclusiveness | Does it work for people of all abilities and backgrounds? | Speech feature that fails for certain accents |
| Transparency | Can people understand how and when it's being used? | Users unaware an AI made a decision |
| Accountability | Are humans answerable for the system's behavior? | "The algorithm decided" with no one responsible |
⚠️ Common Misconception: "Responsible AI is mainly a legal checkbox you complete at the end." It isn't. Each principle shapes choices made early — what data you collect, which model you choose, what you disclose to users. Treating it as end-stage paperwork is exactly how biased or unsafe systems ship.
