Supervising AI Agents: The Rules of the Road

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-agentic-supervision-rules-teens-final2-teen

1. What does 'permission' mean when setting up an AI agent?

   1. What the agent is allowed to do on its own without asking a human
   2. How much memory the agent uses
   3. How fast the agent can process information
   4. What programming language the agent was built with

2. Why is 'reversibility' an important concept when supervising AI agents?

   1. It ensures mistakes can be undone and damage can be fixed
   2. It makes agents respond faster to commands
   3. It allows agents to learn from their errors automatically
   4. It prevents agents from using too much computer memory

3. What is 'oversight' in the context of supervising AI agents?

   1. The agent watching what users do
   2. The automatic security features of the agent
   3. A type of error message the agent generates
   4. Human monitoring and approval of agent actions

4. Why should agents be given read-only access first instead of full capabilities?

   1. So they can suggest actions but not take risky ones until humans approve
   2. To prevent them from learning too much about the user
   3. To make them run faster and use less energy
   4. Because read-only is the only safe mode for all agents

5. Which type of action should ALWAYS require human approval before an agent proceeds?

   1. Actions that take longer than one second to complete
   2. Actions that require typing text into a form
   3. Actions that involve reading data from a database
   4. Irreversible actions like sending messages, making payments, or deleting files

6. What is a sandbox environment used for with AI agents?

   1. A type of computer game designed for learning
   2. A testing area where agents can try actions without real-world consequences
   3. A cloud service that hosts agent programs
   4. A secure container that prevents agents from running at all

7. Why is it important to always have an 'undo path' for agent actions?

   1. So mistakes can be fixed and damage can be reversed if something goes wrong
   2. To allow the agent to learn from its mistakes automatically
   3. To satisfy legal requirements for AI systems
   4. To make the agent run faster than normal

8. Which of these is considered an irreversible action that needs human approval?

   1. Checking the current time and date
   2. Sending an email to a client, transferring money, or deleting a file
   3. Reading a document from a shared folder
   4. Asking the agent to summarize a report

9. How should you think about setting permissions for an AI agent?

   1. Like a manager deciding what employees can do on their own, what needs approval, and what's forbidden
   2. Like a computer program that follows exact instructions
   3. Like a teacher grading student work by checking every answer
   4. Like a parent watching children play to keep them safe

10. What does 'sign-off' mean in agent supervision?

    1. The final step when an agent is turned off
    2. The agent's digital signature on documents it creates
    3. A mathematical formula used to verify agent decisions
    4. Human approval given before an agent proceeds with a specific action

11. Why should you test an agent in a sandbox before letting it work in the real system?

    1. To impress clients with the agent's capabilities
    2. To make the agent run faster when it goes live
    3. To satisfy requirements from internet companies
    4. To catch mistakes and problems before they cause real harm or costs

12. Which of the following is an example of a reversible action for an AI agent?

    1. Saving a draft of an email before sending, using version control, creating backup copies
    2. Sending a payment to a vendor, deleting a customer account, publishing a blog post
    3. Shredding a document, closing a bank account, uninstalling software
    4. Mailing a letter, making a phone call, sending a fax

13. Why do irreversible actions specifically need explicit human OK rather than automatic approval?

    1. Because agents are never smart enough to handle them
    2. Because they are too complicated for agents to understand
    3. Because once completed, they cannot be undone and may cause real harm or losses
    4. Because the law requires a human signature on all of them

14. What does the phrase 'drafts not sends' mean when setting up an agent?

    1. Have the agent prepare content but require a human to review before it goes out
    2. Never let the agent write anything at all
    3. Always let the agent send messages immediately without checking
    4. Only use agents for drafting and never for sending

15. What could happen if an AI agent is allowed to take actions without proper oversight?

    1. It could become too smart and take over the computer
    2. It could send wrong messages, lose money, or delete important files by mistake
    3. It could refuse to work and go into sleep mode
    4. It could learn to lie to humans about what it does