Computing Bias FRQ Practice

Responses receive AI feedback against a rubric after submission. Scores are practice estimates.

Question 1 of 5

A hospital triage chatbot is trained on chats from English-speaking adults aged 30 to 50. Teen patients and patients who mix languages report more irrelevant advice.

(a) Identify a limitation in the training data.

(b) Who is mis-served as described?

(c) Explain how the data limitation connects to that outcome.

Question 2 of 5

A face unlock system is tested mainly on users without facial hair. Bearded users unlock less reliably after launch.

(a) What was missing from the test set?

(b) Who is mis-served?

(c) Why can biased testing hide problems until deployment?

Question 3 of 5

A city builds a predictive policing map.

Data used to train Neighborhoods heavily represented Neighborhoods rarely represented
Past arrest records from 2012 to 2018 Districts A and B District C

(a) Using the table, which district may be underrepresented in training?

(b) How might residents of that district be mis-served or unevenly treated?

(c) Why does historical arrest data risk encoding past bias?

Question 4 of 5

An automated essay scorer trains on essays written in one academic style. Students who use storytelling structures common in their communities score lower despite strong arguments.

(a) Identify the dataset bias.

(b) Who is mis-served?

(c) State one way a designer could check for this bias.

Question 5 of 5

A grocery demand model trains on weekday shopping logs. Weekend shoppers who buy different staples see empty shelves for their usual items.

(a) What pattern was overrepresented in training?

(b) Who is poorly served?

(c) How does leaving weekend patterns out create bias in outcomes?