Skip to content

Quiz: AI and Government

Test your understanding of how artificial intelligence is transforming government operations, democratic processes, civil liberties, and the challenge of governing emerging technologies with these review questions.

1. Algorithmic bias in government AI systems refers to what problem?

  1. Systematic errors in AI outputs that produce discriminatory results against particular groups, often because training data reflected historical inequalities
  2. The tendency of government algorithms to favor certain political parties over others in voter registration systems
  3. The deliberate programming of AI systems by government contractors to disadvantage minority applicants for benefits
  4. The inability of AI systems to process languages other than English, excluding non-English-speaking citizens
Show Answer

The correct answer is A. Algorithmic bias emerges when AI systems trained on historical data reproduce and sometimes amplify patterns of discrimination embedded in that data. In government contexts, examples include risk-assessment tools in criminal sentencing that score Black defendants as higher risk, benefits algorithms that disadvantage non-English speakers, and facial recognition systems with higher error rates for women and people of color. The bias is typically not deliberately programmed but is a structural consequence of training on historically unequal data.

Concept Tested: Algorithmic Bias

2. Deepfake technology poses a particular threat to election integrity because it can do what?

  1. Automatically change voting records in electronic election management systems without leaving a traceable log
  2. Create highly realistic synthetic audio and video content showing candidates saying or doing things they never actually said or did
  3. Intercept and alter ballots transmitted electronically between precincts and central counting facilities
  4. Disable voting machines through cyberattacks timed to coincide with peak voting hours
Show Answer

The correct answer is B. Deepfakes use generative AI to synthesize realistic video and audio of real people, including political candidates, saying things they never actually said. This technology threatens election integrity by enabling the rapid creation and viral spread of disinformation at a scale and realism previously impossible. A realistic deepfake of a candidate making a damaging statement, released hours before an election, may spread faster than fact-checkers can rebut it. This is a qualitatively new challenge compared to traditional text-based disinformation.

Concept Tested: Deepfakes

3. The EU AI Act, adopted by the European Union, establishes a regulatory framework based on what organizing principle?

  1. A flat prohibition on all government use of AI in any decision that affects individual rights
  2. A voluntary certification program with no enforcement mechanism, relying entirely on industry self-regulation
  3. A licensing regime requiring all AI developers to obtain government approval before releasing any product
  4. A risk-based tiered system classifying AI applications by potential harm, with stricter requirements for higher-risk uses
Show Answer

The correct answer is D. The EU AI Act (2024) organizes AI regulation around a risk hierarchy. "Unacceptable risk" applications (such as social scoring systems or most real-time biometric surveillance in public spaces) are banned. "High risk" applications (such as AI in critical infrastructure, education, employment, and law enforcement) face mandatory conformity assessments, transparency requirements, and human oversight. "Limited risk" systems have transparency obligations. "Minimal risk" systems face no additional requirements. This tiered structure aims to enable innovation while protecting fundamental rights.

Concept Tested: EU AI Act

4. Predictive policing tools use AI to forecast where crimes may occur or flag individuals statistically likely to offend. What is a primary civil liberties concern about their use?

  1. They may target communities based on race or past over-policing rather than actual risk, creating self-reinforcing feedback loops and raising due process concerns
  2. They analyze fingerprint databases in real time, and the concern is that fingerprint evidence is unreliable
  3. They monitor communications of known criminal suspects, and the concern is that they violate attorney-client privilege
  4. They automate the writing of police reports to reduce officer paperwork, and the concern is that errors in reports are harder to detect
Show Answer

The correct answer is A. Predictive policing systems use historical crime data, geographic patterns, and sometimes individual risk factors to allocate police resources or flag individuals. Civil liberties advocates raise two main concerns: if training data reflects historical over-policing of communities of color, the algorithm will recommend more policing of those same communities, creating a self-reinforcing feedback loop; and acting on predicted (not actual) criminal behavior raises Fourth Amendment and due process concerns about surveillance without individualized suspicion.

Concept Tested: Predictive Policing

5. Under current Fourth Amendment doctrine, how have courts generally approached government use of facial recognition technology in public spaces?

  1. Courts have unanimously held that facial recognition in public spaces requires a warrant because it constitutes an unreasonable search
  2. Courts have uniformly ruled that facial recognition in public spaces is always constitutional because people have no expectation of privacy in their public appearance
  3. The legal landscape is unsettled, with some courts applying the Carpenter v. United States (2018) principle that pervasive digital surveillance may require a warrant even for public movements
  4. Congress has passed comprehensive legislation clearly defining when facial recognition requires a warrant, leaving little room for judicial interpretation
Show Answer

The correct answer is C. Fourth Amendment doctrine on facial recognition is actively developing and unsettled. In Carpenter v. United States (2018), the Supreme Court held that long-term surveillance of cell phone location data requires a warrant even though the data was voluntarily shared with a third party, because its comprehensive nature reveals intimate details of life. Many scholars and some lower courts argue this reasoning extends to persistent facial recognition surveillance. No comprehensive federal facial recognition law exists as of 2025, and court decisions have been inconsistent across jurisdictions.

Concept Tested: AI and Fourth Amendment

6. The Biden administration's AI Bill of Rights Blueprint (2022) outlined five principles for responsible AI use. Which of the following is NOT one of those five principles?

  1. Safe and effective systems
  2. Algorithmic discrimination protections
  3. Mandatory criminal penalties for AI developers whose systems cause harm
  4. Data privacy
Show Answer

The correct answer is C. The White House Office of Science and Technology Policy's AI Bill of Rights Blueprint (October 2022) identified five principles: (1) Safe and Effective Systems, (2) Algorithmic Discrimination Protections, (3) Data Privacy, (4) Notice and Explanation (you should know when AI is being used and how it affects you), and (5) Human Alternatives, Consideration, and Fallback (you should be able to opt out of automated systems). The Blueprint was a non-binding advisory document—it did not establish mandatory criminal penalties for AI developers.

Concept Tested: AI Bill of Rights Blueprint

7. Congressional oversight of AI presents unique challenges compared to oversight of traditional government activities. Which challenge is most significant?

  1. The Constitution prohibits Congress from regulating private technology companies
  2. Members of Congress are constitutionally prohibited from consulting technical experts outside the legislative branch
  3. AI oversight requires unanimous consent of all 50 states, making federal regulation impossible
  4. AI systems change and improve rapidly, while the legislative process is slow, creating a persistent gap between deployed capabilities and governing rules
Show Answer

The correct answer is D. The "pacing problem"—where technology advances faster than legislation and regulation—is especially acute for AI. Congressional hearings on social media in the 2010s demonstrated how legislators struggled to understand rapidly evolving platforms; AI presents even greater technical complexity. By the time Congress passes comprehensive AI legislation, the technology may have changed substantially. This creates pressure for more flexible regulatory approaches (like agency rulemaking with periodic updates) rather than rigid statutory rules that quickly become obsolete.

Concept Tested: Congressional AI Oversight

8. Autonomous weapons systems raise significant questions about the laws of war. Which is the most fundamental ethical concern about fully autonomous lethal systems?

  1. Autonomous weapons are more expensive than human soldiers, creating unsustainable defense budgets
  2. Removing human judgment from life-and-death targeting decisions may violate international humanitarian law requirements for proportionality, distinction between combatants and civilians, and human accountability for violations
  3. Autonomous weapons are unreliable in bad weather, creating military vulnerabilities
  4. International law already comprehensively bans autonomous weapons, making their development a straightforward treaty violation
Show Answer

The correct answer is B. International humanitarian law (the laws of armed conflict) requires that lethal force decisions satisfy principles of distinction (between combatants and civilians), proportionality (civilian harm proportionate to military advantage), and military necessity. Critics argue that current AI cannot reliably apply these contextual, morally loaded judgments in complex battlefield environments. Additionally, if an autonomous weapon commits a war crime, accountability is diffuse—between the manufacturer, programmer, military commander, and deploying state. No comprehensive international treaty banning autonomous weapons yet exists.

Concept Tested: Autonomous Weapons

9. AI-generated disinformation at scale poses a qualitatively new challenge to democratic elections that traditional media literacy approaches may be insufficient to address. Evaluate this claim by considering both sides.

  1. The claim is false; media literacy programs have always been effective and AI changes nothing about the fundamental challenge
  2. The claim is partially supported; existing debunking tools and platform policies can effectively counter AI disinformation if adequately resourced
  3. The claim is strongly supported; the combination of hyper-realistic synthetic media, algorithmic amplification, and targeting capabilities creates disinformation threats that outpace individual media literacy skills, requiring systemic technical and regulatory responses
  4. The claim is irrelevant; citizens have always ignored political information they dislike, and AI disinformation will have no measurable effect on election outcomes
Show Answer

The correct answer is C. Prior disinformation required human effort to create and spread. Generative AI enables the production of thousands of realistic fake articles, images, audio clips, and videos at near-zero cost, which can then be micro-targeted to specific vulnerable audiences using platform data. Individual media literacy—while valuable—operates person by person and cannot scale at the same speed as automated disinformation. Experts broadly argue that effective responses require a combination of technical detection tools, platform transparency requirements, pre-bunking at scale, and regulatory frameworks—not individual skill alone.

Concept Tested: AI-Generated Disinformation

10. Design a regulatory framework for government use of AI in criminal sentencing and pretrial risk assessment that balances efficiency with constitutional due process requirements. Which framework best addresses all relevant concerns?

  1. Ban all AI use in criminal sentencing and pretrial decisions because any algorithmic influence on liberty interests is unconstitutional
  2. Allow unrestricted AI use in sentencing because judges retain the final decision, making any constitutional concerns moot
  3. Require AI developers to obtain a government license before selling risk assessment tools to courts, with licensing decisions made entirely by the judiciary without public input
  4. Require that AI risk tools be validated for accuracy and fairness across demographic groups, that defendants receive full disclosure of the algorithm's inputs and methodology, that they can challenge the tool's output, and that a human judge makes the ultimate decision with AI as one non-determinative input
Show Answer

The correct answer is D. Due process requires meaningful notice and an opportunity to be heard. In State v. Loomis (2016), the Wisconsin Supreme Court upheld use of the COMPAS risk tool but noted the judge did not rely on it as the sole basis for sentencing. A comprehensive framework must satisfy: (1) scientific validity—the tool must be demonstrably accurate and fair across race, gender, and age; (2) transparency—defendants must know what factors are used; (3) contestability—defendants must be able to challenge the tool's output; and (4) human primacy—the algorithm informs but never determines the sentence. Outright bans ignore genuine efficiency benefits; unrestricted use ignores real constitutional risks.

Concept Tested: Algorithmic Decision-Making