GCP Generative AI Leader Practice Question 29: Hallucination Human in the Loop
Google Cloud Generative AI Leader · Test 3 · question 9 of 10
A hospital pilots a gen AI assistant that drafts summaries of patient history to help clinicians prepare for appointments. During testing, the assistant occasionally states a medication or allergy that does not actually appear anywhere in the patient's record, stated with the same confident tone as the facts that are correct. Leadership must decide how to deploy the assistant safely.
What is the most appropriate control given the stakes involved?
Show answer and explanation
Answer:
Gen AI models can hallucinate: producing plausible sounding but factually wrong content, and current models offer no way to eliminate this risk entirely. In a high-stakes domain such as clinical care, the appropriate control is human-in-the-loop review: a qualified person checks the AI's output against the source of truth before it affects a decision, so an error is caught before it can cause harm. This lets the organization capture the assistant's time savings while keeping accountability with a human.
Option B is an overcorrection. Gen AI is being used broadly in healthcare-adjacent workflows precisely because human review, not outright avoidance, is the standard mitigation for hallucination risk.
Option C is insufficient for a high-stakes use case. A generic disclaimer does not stop an incorrect medication or allergy from reaching a clinician who may reasonably trust a confidently worded summary; the record still needs to be checked.
Option D is a factual error. Lowering temperature toward zero makes output more deterministic and often more conservative, but it does not guarantee grounded, hallucination-free responses; a model can state an ungrounded fact confidently even at low temperature.
Reference: Responsible AI