Big Idea 5 · Topic 5.3 · Exercise 2
The Bias Reduction Clinic
25 minutes - 4 cases - LO IOC-1.D
Back to the Topic 5.3 lessonAll CSP topics
How this page works
The writing boxes below are yours. What you type stays in this browser, is saved here as you go, and is never sent to us or to your teacher. Hand the written work in the way your teacher asked for it.
The graded check at the bottom is the part that records to your teacher's gradebook.
The cases, from your handoutNot recorded
The same cases as your handout, in the same order.
Scenario 1 · Foundational
A translation app renders the gender-neutral sentence 'my friend is a doctor' as 'he is a doctor,' and 'my friend is a nurse' as 'she is a nurse' - every time. The app learned translation patterns from millions of published books and articles spanning several decades.
Channel + level (algorithm or data; where in development it entered):
Reduction action + justification (the deliberate step and why it combats the bias):
Limit to layer (enrichment - what your fix cannot repair):
Scenario 2 · Core
A scholarship-matching site contains an eligibility filter its developers wrote by hand: applications without a permanent home address are automatically set aside 'to reduce fraud.' Counselors notice that students in temporary housing - often the students who most need scholarships - never appear in any match list. No machine learning is involved.
Channel + level (algorithm or data; where in development it entered):
Reduction action + justification (the deliberate step and why it combats the bias):
Limit to layer (enrichment - what your fix cannot repair):
Scenario 3 · Stretch
A city pilots sidewalk-monitoring cameras that flag 'suspicious loitering.' The system passed every accuracy check - all run on daytime footage from the business district, where the pilot team works. Deployed citywide and overnight, it generates false alarms overwhelmingly in a few residential neighborhoods. The vendor's response: 'give it time - every month of new footage makes the model smarter.'
Channel + level (algorithm or data; where in development it entered):
Reduction action + justification (the deliberate step and why it combats the bias):
Limit to layer (enrichment - what your fix cannot repair):
Scenario 4 · Challenge
A wellness-chat app for teenagers escalates messages to a counselor when it detects 'crisis language.' It was framed, designed, coded, and tested by one small, homogeneous team; its phrase list came from transcripts of that team's own role-played conversations. Reviewers find it over-escalates slang from some communities as 'aggression' while missing genuine distress phrased in dialects the team never wrote. The company asks you for a complete reduction plan - not a patch.
Channel + level (algorithm or data; where in development it entered):
Reduction action + justification (the deliberate step and why it combats the bias):
Limit to layer (enrichment - what your fix cannot repair):
No graded check on this page yetNot recorded
About the line in your handout
Your handout says this exercise is available online and auto-graded. The exercise IS here, and it is the same work in the same order, but the auto-graded half is not written for this topic yet. Nothing on this page is scored or sent to your teacher.
Hand this in the way your teacher asked for it. Topic 1.1 has its graded check today, and the rest follow.