True or false: A computing system can produce biased results even when no individual involved in creating it intends to discriminate.
5. Bias in Computing and Artificial Intelligence Online Quiz Questions
Use this free practice quiz with 30 questions to review 5. Bias in Computing and Artificial Intelligence, test your knowledge, and prepare for your next test or exam.
A survey about access to public services is conducted only online. Which type of bias is most directly illustrated if people with limited internet access are underrepresented?
- A
Historical bias
- B
Sampling bias
- C
Measurement bias
- D
Labeling bias
A reviewer accepts an AI-generated risk score without checking contradictory evidence. This behavior is called .
True or false: If an AI system has high overall accuracy, it is necessarily fair across all demographic groups.
- A
True
- B
False
What is the specific two-word term for a reviewer trusting a computer-generated score too much?
Which practices can help reveal whether an AI system performs differently for relevant groups? Select all correct choices.
- A
Compare false-positive and false-negative rates across groups
- B
Evaluate the system only on its training data
- C
Test performance on separate validation data
- D
Conduct intersectional analysis
A health-risk model uses medical spending as a proxy for health needs, even though spending differs because people have unequal access to care. Which type of bias is most directly involved?
- A
Representation bias
- B
Labeling bias
- C
Measurement bias
- D
Missing-data bias
The NIST AI Risk Management Framework organizes continuing risk-management work into four functions: , , measure, and manage.
What two-word category describes bias arising from institutions, policies, infrastructures, and social practices that advantage some groups and disadvantage others?
Which statements accurately describe limitations or risks in bias mitigation and evaluation? Select all correct choices.
- A
A ZIP code may act as a proxy for a sensitive characteristic
- B
A high overall average may hide severe errors for a smaller group
- C
Deleting the race field guarantees that the system cannot discriminate
- D
A model may perform poorly when used with a different population
A company is considering an AI hiring tool trained on records of past hires. Which plan best addresses the material's concerns about historical bias, proxy variables, and accountability?
- A
Deploy the model immediately because automation removes human subjectivity
- B
Remove every sensitive attribute and stop evaluating group outcomes
- C
Use only the overall accuracy score to approve the system
- D
Define job-related criteria, audit historical labels, test relevant groups, monitor outcomes, and provide human review and appeal
Why is fairness in an automated decision-making system an ongoing governance and evaluation responsibility rather than a one-time technical certification? Explain how an organization should address this responsibility throughout the system life cycle.
True or false: A model validated on one population can be assumed to perform equally well when deployed to a substantially different population.
- A
True
- B
False
Which safeguards are recommended when deploying an AI system that can make high-impact decisions? Select all correct choices.
- A
Keep the system's purpose and limitations secret to prevent gaming
- B
Provide understandable information about the system's purpose and limits
- C
Give affected people a meaningful way to appeal
- D
Monitor for drift and disparate impacts after deployment
A company trains a hiring model on records from a workforce that was historically dominated by men. The model learns that characteristics associated with men predict hiring success. Which type of bias best describes this problem?
- A
Representation bias
- B
Historical bias
- C
Missing-data bias
- D
Sampling bias
A developer removes a race field before training a model but leaves ZIP code and school attended as features. What is the most accurate conclusion?
- A
It guarantees that the model cannot discriminate.
- B
It makes subgroup testing unnecessary.
- C
It may leave proxy variables that encode sensitive characteristics.
- D
It ensures that all groups receive identical outcomes.
Which statement best explains why overall accuracy is not sufficient to establish that an AI system is fair?
- A
A model can be accurate on average while still being unfair to a minority group.
- B
A model is fair whenever its overall accuracy is high.
- C
Accuracy and fairness always measure the same property.
- D
Fairness can be determined without considering the system's purpose or context.
True or false: Adding a human reviewer to an AI decision-making process automatically removes bias from the system.
- A
True
- B
False
True or false: A large dataset is automatically a fair dataset.
- A
True
- B
False
What term describes a reviewer's tendency to trust a computer-generated score too much?
How many continuing risk-management functions does the NIST AI Risk Management Framework identify? Enter the number as a whole number.
A health-risk model uses medical spending as a proxy for health needs, even though spending is affected by unequal access to care. What specific type of data bias is this?
What evaluation approach would compare outcomes for older women with disabilities as a combined group rather than examining age, gender, and disability only one at a time?
Complete the sequence of NIST AI Risk Management Framework functions: govern, map, , manage.
A résumé-screening model trained on applicants for one technical role is later used to rank applicants for a different role with a different population. Which problem is most directly illustrated?
- A
Labeling bias
- B
Missing-data bias
- C
Historical bias
- D
Deployment bias
A speech-recognition system is trained almost entirely on recordings from speakers with one regional accent. It performs much less accurately for speakers with other accents. Which type of bias best describes this problem?
- A
Historical bias
- B
Representation bias
- C
Measurement bias
- D
Automation bias
A health-risk model uses a person's past medical spending as a proxy for the person's health needs. The model systematically underestimates the needs of people who have had less access to medical care. Which problem is most directly illustrated?
- A
Sampling bias
- B
Historical bias
- C
Measurement bias
- D
Missing-data bias
A risk model is 95% accurate overall, but its error rate is much higher for a smaller demographic group. Which conclusion is best supported by the material?
- A
High overall accuracy proves that the system is fair.
- B
Fairness can be evaluated without identifying affected groups.
- C
A model is fair whenever it uses the same threshold for everyone.
- D
A model can be accurate overall while still being unfair to a minority group.
A recruiter treats an AI-generated applicant ranking as objective and stops examining qualified applicants who were ranked lower. Which concept best explains the recruiter's behavior?
- A
Automation bias
- B
Sampling bias
- C
Historical bias
- D
Measurement bias
A screening system uses employment records. People who live in areas with unreliable transportation have more gaps in those records, and the system penalizes applicants with such gaps. Which NIST category most directly describes the underlying unequal transportation condition?
- A
Computational bias caused only by an incorrect algorithm
- B
Human bias caused only by an annotator's personal judgment
- C
Systemic bias arising from unequal social conditions
- D
Representation bias caused only by a small training set
Related study packs and tools
AP Computer Science A: Integrated Java Problem Solving
A practical review of the integrated Java skills, debugging habits, algorithm reasoning, and free-response strategies needed for AP Computer Science A.
Fault Tolerance and Redundant Systems
A progressive guide to designing, operating, and testing systems that continue useful service despite component faults, overload, data loss, and recovery challenges.
How the Internet Works
A medium-difficulty quiz on how independently operated networks, protocols, addresses, naming systems, transport services, and routing mechanisms work together to form the Internet.
Java ArrayList and Collections: Operations, Traversals, and Algorithms
Learn how to create, modify, traverse, and analyze Java lists with ArrayList, wrapper types, standard algorithms, and the Collections utility class for AP Computer Science A.
Java Boolean Logic and Selection
Build a clear understanding of Java Boolean expressions, logical operators, selection statements, nested branching, equivalent logic, and systematic code tracing for AP Computer Science A.
Java Foundations and Program Design
A practical introduction to Java program structure, data types, operators, input and output, problem-solving strategy, testing, and debugging for AP Computer Science A.
Java Inheritance and Polymorphism
A practical guide to Java inheritance, constructor chaining, method overriding, polymorphism, access control, and object behavior for tracing and writing AP Computer Science A programs.
Java Objects, Classes, and the String API
A focused guide to Java classes, object references, constructors, methods, strings, wrapper classes, and API usage for AP Computer Science A.
Java Primitive Types and Expression Evaluation
A focused guide to Java primitive types, expression evaluation, numeric conversion, floating-point behavior, integer overflow, and common methods in the Math class for AP Computer Science A.
Java Primitive Types and Expressions
A focused review of Java primitive types, expression evaluation, numeric conversion, arithmetic behavior, Math methods, randomness, and tracing techniques for AP Computer Science A.
Packets, Links, and Network Performance
A progressive guide to how packet-switched networks encapsulate data, move it across links, and manage delay, throughput, queues, and congestion.
Parallel and Distributed Computing
A structured guide to designing parallel and distributed systems through decomposition, communication, coordination, state management, scalability, and fault tolerance.