Human-in-the-Loop Agents Quiz

Answers and explanations appear as you take the quiz.

1. What defines a human-in-the-loop agent compared to a fully autonomous one?
   • It pauses for human review or approval before consequential actions
   • It runs faster than autonomous agents
   • It never uses any machine learning
   • It can only operate offline
2. What is a 'confidence-based escalation' pattern in a HITL system?
   • The agent always asks a human for every action
   • The agent only interrupts a human when its certainty falls below a set threshold
   • The agent ranks humans by their confidence
   • The agent boosts its own confidence over time
3. How is a human checkpoint typically implemented inside an agent's control loop?
   • By deleting the agent's memory
   • By running two agents in parallel
   • As an interrupt or tool-call gate that suspends execution and requests review
   • By disabling the agent's network access
4. In 'active learning,' what happens to the cases a human resolves?
   • They are discarded immediately
   • They are sent to other users
   • They permanently shut down the agent
   • They become training data to improve future predictions
5. For which systems does EU AI Act Article 14 specify human-oversight requirements?
   • High-risk AI systems within the Act’s scope, with oversight proportionate to risk, autonomy, and context
   • Every AI system in every application
   • Only low-risk systems that generate text
   • Only systems that use an explicit approval button
6. Which scenario best illustrates a sensible HITL approval gate?
   • An agent sends every email automatically with no review
   • An agent drafts a large refund and waits for a manager's sign-off
   • An agent ignores all human input
   • An agent randomly approves transactions
7. What does Article 14 of the EU AI Act say human oversight should aim to do for high-risk AI?
   • Guarantee all outputs are correct
   • Replace all technical safeguards
   • Prevent or minimize risks to health, safety, or fundamental rights
   • Require a human to approve every AI output in every context
8. What should effective human oversight under Article 14 help overseers do?
   • Accept every system recommendation without review
   • Approve every action regardless of risk
   • Replace the system’s logs with a human summary
   • Recognize automation bias, interpret outputs, and disregard, override, or interrupt them when appropriate