AP CSP Topic 5.3 Guided Notes - Computing Bias
Big Idea 5: Impact of Computing · Topic 5.3 · Guided Notes (Student)
Computing Bias — Guided Notes
Fill these in during class or catch up here if you were absent. Print this page or work on paper — then check yourself with the CFUs on the Topic 5.3 page.
Print these notesTopic 5.3 lesson pageAll CSP topics
Today’s objectives
- Explain how computing innovations can reflect existing human biases — written into their algorithms or carried in the data they use (LO IOC-1.D)
- Locate where bias can be embedded across all levels of software development, and name the actions programmers take to reduce it (LO IOC-1.D)
Bell ringer
Riverview High buys AttendanceCast, software that predicts which students will skip class next semester. It was trained on exactly ONE dataset: last year's honors students. This fall it scores every student in the building.
Predict: which students will it WRONGLY flag as likely skippers — and what, in how the tool was built, causes those mistakes? List two groups and the cause.
01. Where Bias Comes From
Key Vocabulary (LO IOC-1.D)
| Term | Definition (write it) |
|---|---|
| Algorithmic bias | |
| Training data |
Innovations Mirror Their Makers
- An innovation can inherit human bias through two channels: rules typed by programmers, or .
- Bias in a computing innovation starts with , not with the machine.
- A system built on a lopsided sample treats that slice as .
Two Channels, One Symptom
Bias in the ALGORITHM
Bias in the DATA
- If the unfairness was typed straight into the decision rules, the channel is .
- If flawless code still produces skewed results, suspect .
AP TIP: One phrase gives the channel away — 'the developer wrote a rule…' points to the algorithm; 'the system was trained on…' points to the data.
Stop and think
- AttendanceCast wrongly flags a student who works a night shift. Name the channel the bias came through, and point to the exact build decision that created it.
- A teammate insists a hiring tool cannot be biased since 'the computer just follows math.' Explain the two ways existing human bias can still get in.
- In one sentence, describe how you would rebuild AttendanceCast's dataset so its predictions treat the whole school fairly.
Answer in complete sentences. Then check yourself with the matching CFUs on the Topic 5.3 page.
02. Every Level, and the Fix
Embedded Everywhere, Fought on Purpose
- Bias can enter at of building software, not just data collection.
- The CED expects programmers to bias, not wait for it to dilute.
- Teams with varied backgrounds help since members .
Diagnose the Bias
Five systems misbehave. For each: the channel, the level where bias entered (D.3), and one deliberate reduction action (D.2).
Complete the empty cells.
| Biased system | Channel | Level it entered | One reduction action |
|---|---|---|---|
| Résumé screener trained on ten years of past hires — mostly one group | Data collection (training set) | Rebuild the training set to mirror the real applicant pool | |
| Loan tool with a hand-written rule: certain ZIP codes score lower | Coding (the rule itself) | Delete the ZIP-code proxy; audit every rule for stand-ins | |
| Fitness app counts only steps — wheelchair athletes always score zero | Algorithm | Bring diverse users into design | |
| Face unlock tested almost entirely on lighter-skinned users | Data | Testing (skewed test set) | |
| AttendanceCast: honors-only training, rolled out to the whole school | Data | Deployment (used past its data) |
Watch out — “Just Collect More Data”
Myth: AttendanceCast's vendor has a fix ready: next year the tool will train on five times as much data. With enough rows, the bias will average itself out and the flags will become fair.
Explain why this is wrong:
Deep Dive · Beyond the AP Exam
Deep Dive: The Benchmark That Caught the Skew
- The 2018 benchmark exposed skew by scoring accuracy rather than overall.
- Vendors closed the gap only after made the skew visible.
Stop and think
- Pick one row of the diagnosis table. In two sentences: name its channel and level, then defend your reduction action as a deliberate step rather than a bigger data haul.
- A vendor claims its essay-grading tool 'removes human bias entirely — no teacher ever touches the score.' Give two reasons the claim fails.
- Explain why 'add five times more data' fixed nothing for AttendanceCast, but 'change which students the data comes from' would.
Answer in complete sentences. Then check yourself with the matching CFUs on the Topic 5.3 page.
Common AP Traps
Three ways Topic 5.3 loses points on the exam — one minute now, real points in May.
“Bias only enters via data” — in your own words:
“More data fixes bias” — in your own words:
“The algorithm is neutral” — in your own words:
Computing Bias, in One Slide
- Computing innovations can reflect existing human biases — through biases written into the algorithms or biases in the data used.
- Two channels, one test: was the unfairness typed into the rules, or learned from skewed inputs?
- Biases can be embedded at all levels of software development — framing, design, coding, data collection, testing, deployment.
- Programmers should take action to reduce bias — rebuilt datasets, rule audits, per-group testing — combating existing human biases.
- More data is never the fix — and diverse teams help keep bias out from the start.
Exit check — I can…
- ☐ explain how bias enters innovations through algorithms and through data (LO IOC-1.D)
- ☐ locate bias at any level of software development, not just data collection (LO IOC-1.D)
- ☐ name deliberate programmer actions that reduce bias — never 'more data' (LO IOC-1.D)
Get in Touch
Whether you're a student, parent, or teacher — I'd love to hear from you.
Just want free AP CS resources?
Enter your email below and check the subscribe box — no message needed. Students get daily practice questions and study tips. Teachers get curriculum resources and teaching strategies.
Message Sent!
Thanks for reaching out. I'll get back to you within 24 hours.
Prefer email? Reach me directly at [email protected]