Computing Bias FRQ Practice

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

Question 1 of 5

A credit-limit algorithm trains on customers who shopped at specialty stores. Customers who shop mainly at discount stores receive lower limits despite similar incomes.

(a) Identify a training-data limitation.

(b) Who is mis-served?

(c) Explain the connection.

Question 2 of 5

A reading app recommends books using past clicks from one grade level. Advanced readers in other grades get too-easy books.

(a) What group is overrepresented in the click data?

(b) Who is poorly served?

(c) Why does that create biased recommendations?

Question 3 of 5

An insurance risk model is validated as follows.

Validation set composition Group later reporting high error
Drivers under 25 from one metro area Older rural drivers

(a) What does the table suggest about who was left out of validation?

(b) Who is mis-served later?

(c) Why can a model look accurate in validation yet fail for others?

Question 4 of 5

A captioning system trains on sports commentary audio. Lecture captions for quiet seminar rooms have more missing words.

(a) Identify the dataset mismatch.

(b) Who is mis-served?

(c) Suggest one concrete dataset fix.

Question 5 of 5

A spam filter trains on workplace email. Small-business owners who use informal texts see more false spam labels on real customer messages.

(a) What training context is overrepresented?

(b) Who is mis-served?

(c) How does context mismatch create bias?