Skip to content
PlayOwl

Computers & Digital Literacy · Grade 9

Explainability, Limitations and Human Judgment

Learners explain models globally and locally: they compute permutation feature importance as the drop in performance when a feature is scrambled, and write counterfactual explanations for individual decisions, judging whether they are actionable. They then study what explanations cannot tell us (correlation is not causation; out-of-distribution inputs) and design human oversight with review queues, appeals and override logs that counter automation bias.

  • 4 lessons
  • About 110 minutes
  • Explanation and oversight check: 10 questions
  • Ways to answer: choose an answer, highlight a text, type a number, put in order, sort into groups and write a short answer
See plans Included with Family, Teacher and School plans. Open in PlayOwl

What learners will discover

  • Distinguish global and local explanations, and compute and interpret permutation feature importance as the change in performance when one feature is scrambled.
  • Write counterfactual explanations (the smallest change that flips a decision) for a points-based model and judge whether they are actionable.
  • Recognize the limits of explanations: model importance is not causation, and out-of-distribution inputs fall outside what a model has learned.
  • Design human oversight (review queues, appeals, override logging, audits) and identify practices that increase or reduce automation bias.

Lessons

  1. Global explanations and feature importance 8 steps5 practiceabout 28 min
  2. Local explanations and counterfactuals 7 steps4 practiceabout 26 min
  3. What explanations cannot tell you 6 steps4 practiceabout 26 min
  4. Human oversight and automation bias 6 steps4 practiceabout 30 min
  5. Explanation and oversight check 10 questions at the end

Keep exploring

More 9th grade computer skills

Guide for grown-ups