300-745 practice question 13 of 15

Which generative AI impact is addressed by a human-in-the-loop design policy?

Full set: 58 questions
Answer options
A: AI hallucinations Correct answer
B: deep fakes
C: phishing
D: scale changes
Correct Answer:
A. AI hallucinations
Explanation

A human-in-the-loop (HITL) design policy integrates human oversight into the AI decision-making or content-generation process. This is a primary strategy for mitigating "AI hallucinations"—instances where a generative AI model produces outputs that are nonsensical, factually incorrect, or disconnected from the source input, despite being presented confidently. By inserting a human reviewer to validate, edit, or approve the AI's output before it is finalized, organizations can catch and correct these inaccuracies. This ensures the reliability and trustworthiness of the AI system, which is a core principle of responsible AI implementation.

Why Incorrect

B. deep fakes: While human review can help identify deep fakes, HITL is more about governing an organization's own AI generation process, not primarily a defense against externally created malicious media.
C. phishing: Phishing is an external social engineering attack. A HITL policy is an internal control for an organization's AI systems and not a direct security measure against incoming phishing attempts.
D. scale changes: A HITL policy generally limits, rather than addresses, scale. It introduces a human bottleneck, trading the AI's ability to operate at massive scale for higher accuracy and safety.

References

1. Cisco Responsible AI Framework (Version 2.0, 2023): In the "Accountability" pillar, the framework states, "Human-in-the-Loop: We determine the appropriate level of human oversight for AI systems based on their potential impact." This oversight is the mechanism used to verify outputs and correct errors such as hallucinations. (Page 11, "Accountability" section).

2. Zhang, Y., et al. (2023). A Survey on Hallucination in Natural Language Generation. ACM Computing Surveys, 55(12), Article 248. In Section 4.2, "Mitigation with Human Intervention," the paper discusses how incorporating human feedback and review is a direct method to "alleviate hallucination problems" by having humans correct the generated text. (DOI: https://doi.org/10.1145/3571730).

3. National Institute of Standards and Technology (NIST). (2023). AI Risk Management Framework (AI RMF 1.0). The framework emphasizes the importance of human oversight in AI systems. In the "Govern" function, it discusses implementing policies for "human review and oversight" (Section 3.2) to manage risks, including the generation of inaccurate or inappropriate content (hallucinations).

Scroll to Top