Domain 6 of 6 · Chapter 1 of 2

Understand Responsible AI Principles

Unlock the complete study guide.

Bundled into the existing GitHub Copilot premium course — no separate purchase.

14-day money-back guarantee — no questions asked.

Included in this chapter:

  • Why responsible AI starts with fallible output
  • Limits that come from training data
  • Limits that come from context and generation
  • The six principles of responsible AI
  • Fairness: equal outcomes for similar groups
  • Reliability and safety: edge cases and every harm
  • Privacy and security: collect less, protect the rest
  • Inclusiveness: design for every user
  • Transparency: disclose, explain and keep a record
  • Accountability: people answer for AI outcomes
  • Mitigating harms: govern, map, measure, manage
  • Sensitive domains and consequential decisions
  • Example scenarios: principles, limitations and mitigations

Each responsible AI principle: what it requires, a documented mitigation, and a misreading

PrincipleWhat it requiresDocumented mitigationMisreading
FairnessNo different outcomes for similarly situated people or groupsReview training data, test with balanced demographic samples, adversarial debiasing, monitor by user segment, override controlsHigh overall accuracy proves the system is fair
Reliability and safetyWorks as intended in normal use, edge cases and unexpected input; minimizes physical, psychological and financial harmRed teaming (adversarial testing), safety filters plus human reviewNo crash and no injury means the output is safe
Privacy and securityConsent before collection, only the data needed, protected data and keysPseudonymization or aggregation, encryption in transit and at rest, controlled access to data, models and keysStorage encryption alone controls who can misuse the data
InclusivenessWorks for diverse users, abilities, languages, cultures and infrastructureAlternative interaction modes, language and cultural adaptation, low-resource or offline support, diverse participationOne interface for everyone is equal access
TransparencyUsers know AI is involved, what it can and cannot do, and why it behaved as it didDisclosure of purpose, capabilities and limits, interpretability, logging and auditingNaming the vendor explains the AI's role
AccountabilityPeople and organizations stay responsible; the model is not the final authorityHuman governance and review, especially for consequential decisionsResponsibility passes to the model provider

Decision tree

Similarly situated people orgroups get different outcomes?YesFairnesscompare segments, override controlsNoPersonal data or secrets collectedwithout consent, over-kept or exposed?YesPrivacy and securityconsent, minimize, protectNoSome users cannot use it or benefit(ability, human language, connectivity)?YesInclusivenessalternative modes, adaptationNoUsers cannot tell AI is involved,what it can do, or why it acted?YesTransparencydisclose, explain, logNoNo person answers for the outcome,or the model has the final say?YesAccountabilityhuman authority, recourseNoFails on edge cases or unexpectedinput, or causes unintended harm?YesReliability and safetyedge cases, red teaming, reviewNoNo match? Consider a Copilot limit or the govern, map, measure, manage processillustrative, not an exhaustive test
Cheat sheet

Cheat sheet

  • Hallucinations can sound plausible while being false
  • Training-data coverage affects output quality
  • Local context can hide system-level problems
  • Generated tests do not guarantee complete coverage
  • Custom instructions cannot guarantee identical compliance
  • Fairness requires equitable treatment of similarly situated groups
  • Training data can propagate existing bias and errors
  • Representative data reduces avoidable bias
  • Segment-level monitoring reveals hidden disparities
  • Human controls must be able to override unfair outcomes
  • Adversarial debiasing is a distinct fairness-mitigation technique
  • Reliable systems handle expected and unexpected conditions safely
  • AI safety includes physical, psychological, and financial harm
  • Adversarial testing exposes harmful and insecure behavior
  • Safety filters reduce risk but do not remove review duties
  • Sensitive domains require stronger safeguards and recourse
  • Personal-data collection requires notice and informed choice
  • Data minimization limits privacy exposure
  • De-identification reduces exposure of personal identities
  • Sensitive data requires protection in transit, at rest, and in access
  • Prompt context can expose sensitive information
  • Inclusive systems serve diverse users without disadvantage
  • Accessibility requires alternative modes of interaction
  • Language and cultural adaptation are inclusion requirements
  • Connectivity and computing constraints affect inclusion
  • Diverse stakeholder participation improves inclusive design
  • Transparency discloses an AI system's role and limits
  • Interpretability makes system behavior understandable
  • Logging and reporting make AI decisions auditable
  • People remain accountable for AI systems and outcomes
  • AI risks must be governed, mapped, measured, and managed

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. https://docs.github.com/en/copilot/responsible-use/inline-suggestions
  2. https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/gh-300
  3. https://learn.microsoft.com/en-us/training/modules/responsible-ai-with-github-copilot/
  4. https://docs.github.com/en/copilot/responsible-use/agents
  5. https://docs.github.com/en/copilot/concepts/prompting/response-customization
  6. https://learn.microsoft.com/en-us/training/modules/responsible-ai-with-github-copilot/2-manage-ai-risks
  7. https://learn.microsoft.com/en-us/training/modules/responsible-ai-with-github-copilot/3-six-principles-of-responsible-ai
  8. https://learn.microsoft.com/en-us/azure/machine-learning/concept-responsible-ai?view=azureml-api-2
  9. https://learn.microsoft.com/en-us/agents/design-guidelines/responsible-ai
  10. https://learn.microsoft.com/en-us/azure/foundry/concepts/evaluation-evaluators/risk-safety-evaluators
  11. https://docs.github.com/en/copilot/concepts/context/content-exclusion