Guide-linked quiz · Medium Level
AI in Jury Selection Quiz
Test AI support in jury administration and voir dire, including data quality, bias, court authority, Batson limits, and human verification of outputs.
Question 1 of 8
A model ranks jurors by predicted sympathy for a party. What does that score establish by itself?
Preview every question8 questions
Answers and explanations appear as you take the quiz.
- A model ranks jurors by predicted sympathy for a party. What does that score establish by itself?
- That the prospective juror cannot be impartial
- That the judge must excuse the highest-ranked person
- Only a model output based on its chosen data and target; it does not establish bias or a lawful basis for a strike
- That the score is independent of protected traits
- During voir dire, transcription software changes “not guilty” to “guilty.” What should counsel do before relying on the text?
- Check the passage against the audio and correct the transcript
- Use the generated transcript because it has a confidence score
- Ask a model to summarize the transcript without replaying it
- Treat the wording as immaterial because the question was clear
- A peremptory strike is challenged as race-based. Which principle from Batson is relevant?
- A demographic prediction makes any strike valid
- A software recommendation replaces counsel’s explanation
- Only judges may consider whether a strike was discriminatory
- Race cannot be the basis for a peremptory strike in the covered proceedings
- A public profile may belong to a different person with the same name. Which safeguard addresses this risk?
- Infer identity from a matching city alone
- Verify identity and the original source before treating the information as about a juror
- Use the model’s ranking to resolve identity uncertainty
- Retain the profile permanently in case it becomes useful
- A court’s local order restricts online research about prospective jurors. How should counsel use an AI search tool?
- Search anyway if the tool uses public pages
- Ask the vendor whether the restriction applies to models
- Follow the local order; an AI search does not create an exception
- Use a private browser to avoid reporting the search
- Which assessment is most useful for a model that classifies questionnaire responses?
- Measure only how quickly the system returns a label
- Compare its labels with its own training labels
- Count how often attorneys accept the recommendation
- Compare classifications with independent human annotations and examine disagreement and subgroup errors
- A ranking changes sharply when a non-case-related feature is removed. What does this suggest?
- The recommendation may depend on a feature whose relevance and fairness need scrutiny
- The model has proved that the feature is legally relevant
- The ranking is reliable because it is sensitive to inputs
- The removed feature can safely be used as a proxy
- Which practice best preserves human control over an AI-assisted jury workflow?
- Automatically strike everyone above a threshold
- Let counsel inspect the underlying answer, reject the suggestion, and document the permissible reason for a decision
- Let the vendor retain sole authority over corrections
- Hide the model’s sources to prevent anchoring
Keep testing yourself
More quizzes picked for your level.