Autonomy Levels and Human Control
How much should the agent do on its own?
Autonomy
Autonomy describes how independently an agent acts. Some systems only suggest next steps. Others can take actions automatically. Most real systems are somewhere in between, with human review at important points.
Common autonomy levels
| Level | Description |
|---|---|
| Low | The AI only recommends actions |
| Medium | The AI acts with human approval |
| High | The AI acts automatically within limits |
More autonomy means more responsibility
If an agent can take actions by itself, its mistakes can have larger consequences.
Human-in-the-loop
A common safety design is human-in-the-loop control, where a person reviews, approves, or corrects important actions. This is especially useful for financial, medical, legal, or security-related work.
What does human-in-the-loop mean?
Human-in-the-loop systems keep a person involved in the process.
Correct answer: A human reviews or guides the AI at important points
Why is high autonomy potentially risky?
Mistakes can spread faster when the system acts independently.
Correct answer: Because the agent can make and carry out more decisions without human review.
A useful design principle
A beginner-friendly rule is: give an agent only as much autonomy as needed for the task, and add human checkpoints for higher-risk actions.
AUTONOMY & CONTROL
The spectrum from human-driven to fully autonomous