7 Limitations and Responsible AI Online Quiz Questions

True

False

It will necessarily become more accurate as it encounters unfamiliar data.

Its performance may drop when real-world data differs from its training data.

It will automatically understand why the new data differs.

It will no longer be affected by the choices made during its design.

True

False

A system uses a complex model whose outputs are difficult to interpret.

A system’s performance declines when real-world conditions change.

A system infers sensitive personal details from information collected for another purpose.

A system is tested across groups with relevant data.

Check data quality and whether relevant groups are represented.

Assume neutral-looking inputs cannot act as proxies for sensitive characteristics.

Test system performance across relevant groups.

Involve people with varied perspectives and monitor outcomes after deployment.

Rely on a single fairness score to establish that the system is fair in every respect.

The system has identified the cause and the appropriate response.

The pattern may not explain why the outcome occurred or what action is appropriate.

The system has removed the need for human judgment.

The pattern proves the system will work reliably in every situation.

Define the system’s purpose and assess risks before choosing to use it.

Evaluate it before deployment using relevant data and realistic conditions.

Inform affected people when AI plays a significant role and provide ways to seek review or corrections.

Assign accountability to people and organizations responsible for development, deployment, and oversight.

Monitor the system as conditions and impacts change, and address problems that arise.

Treat deployment as the final step, with no need for later monitoring.

It is primarily a security concern because safety means preventing unauthorized access.

It is primarily a safety concern because safety focuses on preventing harmful outcomes.

It is an explainability concern because all unexpected behavior is an explanation problem.

It is a fairness concern because any system failure is necessarily discriminatory.

Its output must be correct because it sounds plausible.

It may produce a plausible answer that is inaccurate.

It can only make errors when its software is outdated.

It understands the context whenever it gives a confident answer.

Deploy it immediately if it reproduces past hiring decisions accurately.

Assume that patterns in historical records are necessarily job-related.

Test whether it works reliably for the intended use and whether its patterns reflect job-related qualifications.

Avoid evaluating it with people or situations it will encounter in practice.

True

False

Collect only necessary data, restrict access, protect the data, and consider privacy throughout the system’s life cycle.

Collect as much personal information as possible so the system can infer more details.

Focus on privacy only after a data exposure occurs.

Make all personal information available to anyone evaluating the system.

Withhold information about the system’s purpose so users cannot challenge it.

Publish every piece of personal data used by the system.

Explain only the system’s technical architecture, regardless of who is affected.

Provide appropriate information about where and why the system is used, its limitations, and who is responsible.