FREE STUDY NOTES · AI-901

The six Responsible AI principles, with examples

Fairness, reliability and safety, privacy and security, inclusiveness, transparency and accountability, and how to tell them apart in a scenario.

From Ultra Transcenders AI-901 by Tony Rough (publishing soon)

Microsoft describes responsible AI through a short, fixed list of principles. Learning the list, and what each principle is about, lets you classify almost any scenario you meet.

The Microsoft Responsible AI Standard defines six principles of Responsible AI:

  1. Fairness
  2. Reliability and safety
  3. Privacy and security
  4. Inclusiveness
  5. Transparency
  6. Accountability

The principles are about the people AI affects, not about the AI itself or the organisation that builds it.

Common trap: Knowledgeability, decisiveness, opinionatedness, “protecting developers’ interests” and “placing AI in the public domain” sound plausible but are not Responsible AI principles. Autoscaling and Agile are not principles either; they are engineering practices.

What each principle means

The table below gives the meaning of each principle and the kinds of scenario that point to it. When you read a scenario, ask “what is this requirement protecting, and for whom?” and match it to the row.

Principle Means Typical scenarios
Fairness Treat everyone fairly; similarly situated groups not affected differently Checking outputs for demographic bias; assessing outcomes across gender, ethnicity, age and reducing gaps (Responsible AI dashboard fairness assessment); representative training data
Reliability and safety Perform as intended across conditions, respond safely to unanticipated input, resist harmful manipulation, fail safely Testing under varied conditions; handling unusual or missing values; declining to predict when key fields are missing; validating the model in software review
Privacy and security Protect personal and business data; notice and consent; resist attacks Encryption and restricted access; personal data visible only to staff involved; giving consumers controls over their data; not training on data obtained without permission
Inclusiveness Empower and engage everyone: accessibility, language variation, differing technical proficiency Multi-language and screen-reader support; assistive technology; finding barriers that exclude users (speech impairments, strong accents)
Transparency People know when AI is used, what it can and can’t do, and how/why it produces outputs Explaining a loan decision; recording the decision process so each quote can be explained; documentation that helps developers understand the system; disclosing training data sources
Accountability People are answerable for AI; clear human oversight Human in the loop reviewing decisions; humans able to override AI; naming the responsible team; risk governance committee (legal, risk, privacy)

Telling similar principles apart

Several principles overlap in everyday language, so it helps to know where the boundaries sit.

Common trap: Declining to predict when inputs are unusual or missing is sometimes mistaken for privacy and security. Nothing is being protected; the system is avoiding an unreliable answer, which is reliability and safety.

Common trap: When one scenario lists several requirements, map each requirement on its own rather than picking one principle for the whole scenario. For an insurance-quote system: quotes that can be explained = transparency; personal data visible only to the staff involved = privacy and security; screen-reader support = inclusiveness.

Responsible AI is ongoing

Responsible AI doesn’t stop when a model goes live: quality, safety and fairness are monitored in production, and issues found there feed back into the design.

Get the whole book

This note is one section of Ultra Transcenders AI-901: Microsoft Azure AI Fundamentals, an independent study guide that explains every topic the exam covers by technology, with comparison tables, diagrams and the common traps, plus a glossary linked to Microsoft Learn.

Amazon.co.ukKindle: coming soonPaperback: coming soon
Amazon.comKindle: coming soonPaperback: coming soon

Publishing soon on Amazon in Kindle and paperback editions.

About the book · Free AI-901 glossary · All AI-901 study notes

More AI-901 study notes