Prepare and test your skills
Prepare and test your skills
Free sample question, answered and explained below. The Generative AI Leader practice bank has 2,191 more like it. Try 10 free: instant score, no account.
An enterprise human resources department is implementing a generative AI assistant to help summarize candidate resumes and draft recruitment materials. During pre-deployment audits, the project team observes two distinct fairness concerns:
Which option correctly categorizes these two forms of bias and identifies an effective strategy to reduce discriminatory AI outputs during deployment?
This option attributes the fairness issues to architectural capacity constraints—specifically a knowledge cutoff date and token exhaustion—and suggests expanding the context window and raising the model's temperature parameter.
Historical bias occurs when existing societal inequities, stereotypes, or past discrimination are captured in training data and subsequently mirrored or amplified by the AI model. Even if data is accurately collected from the past, it reflects an unfair reality. Representation bias occurs when certain groups, populations, or characteristics are underrepresented or missing in the training and evaluation datasets, leading to degraded performance or skewed generalizations for those specific groups.
Correctly diagnosing the root cause of algorithmic bias ensures targeted remediation. Pairing behavioral steerability (system instructions) with output enforcement (safety filters) offers the most robust mitigation framework for generative AI applications without requiring expensive model retraining from scratch.
This option categorizes the hiring disparities as prompt injection security vulnerabilities and hallucination errors, proposing the use of Sensitive Data Protection (SDP) de-identification templates as the primary mitigation.
This option reverses the definitions of historical and representation bias and suggests disabling responsible AI safety filters to prevent legitimate candidate summaries from being blocked accidentally.
While tuning filter thresholds may be necessary to minimize false positives, completely disabling safety mechanisms in a production hiring tool violates core responsible AI governance principles.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated