The six responsible AI principles
AI systems make decisions that affect people, so Microsoft defines six principles to guide how they are built and used. AI-900 asks you to match a short scenario to one principle, so you need each definition clear.
The six principles
| Principle | Core idea | Typical example |
|---|---|---|
| Fairness | Treat all people equitably | A hiring model must not favour one gender |
| Reliability and safety | Work consistently and safely, even in unexpected conditions | Testing a self-driving system in bad weather |
| Privacy and security | Protect personal data and resist attack | Encrypting stored patient data, limiting access |
| Inclusiveness | Empower and engage everyone | Captions and voice options for users with disabilities |
| Transparency | People understand how and why it works | Showing which factors drove a prediction |
| Accountability | Humans stay responsible | Governance board, human review of decisions |
Fairness, reliability and safety, and privacy and security are the principles that most directly shape outcomes. Inclusiveness, transparency and accountability are the foundation under them.
How to recognise each one
Look at the verb or the risk in the scenario:
- Different outcomes for different groups: fairness.
- The system fails, crashes or gives dangerous output: reliability and safety.
- Leaked, misused or unprotected personal data: privacy and security.
- Some users cannot use the solution: inclusiveness.
- Users do not know why a result was produced, or that AI was used: transparency.
- Nobody owns the result or no review exists: accountability.
Tools and practices
Azure Machine Learning has a responsible AI dashboard, with model explanations (transparency), error analysis, and fairness assessment. Foundry also provides content safety filters for generative AI. You do not need the details, only that such tools exist to support these principles.
Typical good practice includes validating models before release, testing with diverse data, keeping a human in the loop for high-impact decisions, and telling users when they are talking to an AI system.
Exam traps
- Transparency vs accountability: transparency is understanding the system, accountability is who is responsible for it.
- Fairness vs inclusiveness: fairness is equal treatment in outcomes, inclusiveness is making the product usable by people of all abilities.
- Reliability and safety vs privacy and security: the first is about correct, safe behaviour, the second about protecting data.
- Hiding that AI is being used breaks transparency. It is never a responsible practice.
- An Agile process or faster delivery is not a responsible AI principle.
Practice questions
Choose an answer, then review the explanation to check your understanding.
1. A loan approval model approves fewer applications from one age group than from others with the same financial profile. Which principle is at risk?
Show answer
Answer: A.Fairness means AI treats all groups equitably and does not give different outcomes to similar people.
2. A bank must be able to explain to customers which factors led to a credit decision. Which principle does this support?
Show answer
Answer: B.Transparency means people understand how a system reaches its output and what its limits are.
3. You design a voice app so that it also works for users with speech impairments and hearing loss. Which principle are you following?
Show answer
Answer: A.Inclusiveness means designing AI that empowers everyone, including people with disabilities.
4. A company creates a review board that signs off on AI models before release and owns the outcome. Which principle is this?
Show answer
Answer: C.Accountability means people are responsible for AI systems and put governance and oversight in place.