Computing Bias MCQ Practice

Question 1 of 10

A hiring tool is trained only on resumes from past employees who were mostly recent graduates from three universities. Applicants with strong trade-school credentials are ranked lower.

Which bias source is most clearly illustrated?

  1. A. Training data that underrepresents some qualified backgrounds.
  2. B. Perfectly balanced data that cannot create ranking differences.
  3. C. Bias that appears only when no training data exists.
  4. D. Trade-school applicants being overrepresented in the training set.

Question 2 of 10

A skin condition app is trained and tested mostly on light skin tones. Users with darker skin receive more incorrect "benign" labels.

Who is most clearly mis-served?

  1. A. Only the developers, who are unaffected by labels.
  2. B. Users with darker skin tones underrepresented in training and testing.
  3. C. Only users with light skin tones, who receive all errors.
  4. D. No users, because medical apps cannot show bias.

Question 3 of 10

A voice assistant is trained on adult speech samples and fails more often for young children.

Which explanation best fits computing bias?

  1. A. Voice systems ignore training data composition entirely.
  2. B. Bias requires that adult voices also fail equally.
  3. C. Training data left out enough child voices, so children are poorly served.
  4. D. Children always speak clearer than adults to assistants.

Question 4 of 10

A loan model is tested only on urban borrowers. Rural applicants later see higher false denials.

Dataset stage Who is included heavily
Training Urban borrower histories
Testing Urban borrower histories
Deployment Urban and rural applicants

Which conclusion is best?

  1. A. Urban-only testing guarantees fair rural outcomes.
  2. B. Deployment never reveals gaps missed in testing.
  3. C. Rural false denials prove urban borrowers were excluded from training only by accident that cannot relate to bias.
  4. D. Leaving rural cases out of training and testing can mis-serve rural applicants at deployment.

Question 5 of 10

A classroom seating app recommends groups using past project scores from one honors section. Students new to the school with strong skills but no local score history are rarely recommended as leaders.

Which claim is NOT supported?

  1. A. Using only local score history cannot create unfair recommendations.
  2. B. Students without local history may be under-selected as leaders.
  3. C. Training on one section can omit other capable students.
  4. D. Missing score histories can bias who is labeled leadership-ready.

Question 6 of 10

A photo tagging system trained on sunny outdoor images mislabels indoor night photos more often.

Which group of images is poorly represented?

  1. A. Only text documents, which the system tags perfectly.
  2. B. Indoor night photos underrepresented in training.
  3. C. Sunny outdoor photos that never appear in training.
  4. D. All photos equally, with identical error rates by design.

Question 7 of 10

A translation model is trained mainly on formal news text. Users writing informal dialect messages get more nonsensical outputs.

Who is most clearly mis-served?

  1. A. No users, because translation models cannot reflect training mix.
  2. B. Only users who write formal news, who are never served well.
  3. C. Writers of informal dialect underrepresented in the training text.
  4. D. News writers who provided most of the training data and always get the worst outputs.

Question 8 of 10

A fraud detector is trained on credit-card patterns from one age group. Older adults with different spending rhythms trigger more false fraud freezes.

Which statement is best supported?

  1. A. False freezes prove the model was trained equally on every age group.
  2. B. Fraud detectors ignore who appears in training data.
  3. C. Older adults cannot be affected by computing bias.
  4. D. Age groups left out of training patterns can be mis-served by false freezes.

Question 9 of 10

A search ranking learns from clicks by users in one city. Local businesses in other cities rarely appear on page one.

Which bias pathway fits?

  1. A. Training signals from one city clicks underrepresent other cities businesses.
  2. B. All cities clicks were equally represented by definition.
  3. C. Search ranking cannot reflect geographic training skew.
  4. D. Page-one rarity proves other cities dominated training clicks.

Question 10 of 10

A wheelchair-navigation app is tested only on campus paths with curb cuts. Users in older neighborhoods with missing curb cuts get routes that are impassable.

Which evaluation is most accurate?

  1. A. Navigation apps cannot show dataset-related bias.
  2. B. Test data that omits difficult real sidewalks can leave those users mis-served.
  3. C. Campus-only testing guarantees citywide route quality.
  4. D. Impassable routes prove curb cuts were the only training examples needed.