Complete AI Training
Sign inGet my AI kit

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

Video course · 14 chapters · 55 min · certificate

Build & Govern AI: Fix Bias and Unequal Errors

Find and fix bias in AI systems: fairness and decision thresholds, proxies and correlation, unequal error rates by group, their consequences and causes, and case studies that prepare you for mitigation.

Become a memberAlready a member? Sign in

What you'll learn

  • Explain fairness in data and models
  • Use a confusion matrix to compare groups
  • Spot proxy variables with correlation
  • Measure unequal error rates
  • Identify the four main causes of unequal errors
  • Prepare for bias mitigation in development

Chapters

14 chapters · 54:35
  1. 2:41 01Intro Members Getting started Setting up the session on bias in AI development.
  2. 4:19 02Resources Members Resources and case studies Links and five case studies used throughout the session.
  3. 1:34 03Focus Members Today's focus: unequal errors The theme: biases in data, specifically unequal error rates, following the OECD AI lifecycle.
  4. 1:33 04Concepts Members Fairness, proxies and the confusion matrix Key concepts: fairness, proxies and the confusion matrix.
  5. 8:48 05Thresholds Members Fairness and decision thresholds How decision thresholds affect fairness across groups.
  6. 1:56 06Proxies Members Proxies and correlation heat maps Using correlation heat maps to find variables that act as proxies.
  7. 3:49 07Errors Members Unequal error rates Introducing unequal error rates with false positives and false negatives by group.
  8. 3:35 08Examples Members Gender and age examples Worked examples of error rates by gender and age in a dataset of 1,470 people.
  9. 4:03 09Trade-offs Members Consequences and trade-offs Equalizing error rates involves trade-offs, including overall accuracy.
  10. 4:26 10Causes Members Causes of unequal errors Why unequal error rates happen, and what error rate to aim for.
  11. 6:14 11Case studies Members Case studies: under-representation and majority Concrete cases of groups being underrepresented or dominated by the majority.
  12. 5:28 12Causes Members Four reasons, part 1 The four main reasons unequal error rates happen, with concrete examples.
  13. 4:26 13Causes Members Weak inputs and biased labels When signals are weaker for a group, like pulse oximeters, or labels were measured worse.
  14. 1:43 14Next steps Members Wrap-up and mitigation preview Knowing what to correct before mitigation, and a preview of the next session.

Jobs this course suits

Our AI checked this course against 500 jobs; these get the most out of it. Each job links to its learning path.

Diversity and Inclusion ManagersData ScientistsHR AnalystsMachine Learning EngineersCDOs (Chief Digital Officers)Chief Digital Officers (CDOs)CIOs (Chief Information Officers)Human Resources DirectorsVP of Human ResourcesData AnalystsStatisticiansAI Engineers