Step by Step
H
Human-in-the-loop (HITL) — human approves every decision
A human must explicitly approve every single AI decision before it takes effect — the highest level of oversight, but also the slowest and most costly to operate.
Example: a human loan officer reviewing and explicitly approving every single automated loan recommendation before any decision is finalized.
O
Human-on-the-loop (HOTL) — AI acts, human monitors
The AI system acts autonomously in real time, while a human monitors its behavior and retains the ability to intervene if something goes wrong.
Example: an autonomous trading system executing trades in real time, with a human monitor watching dashboards and able to halt trading if something looks wrong.
O2
Human-out-of-loop — fully autonomous
The AI system operates fully autonomously with no human oversight in the loop at all — appropriate only for very low-stakes decisions that have undergone extensive prior testing.
Example: a spam email filter operating fully autonomously, since the stakes of an occasional misclassification are relatively low and the system has been extensively tested.
M
Model cards and algorithmic auditing — documentation and oversight
Model cards provide standardized documentation of a model's capabilities, limitations, intended use, and performance across different demographic groups; algorithmic auditing involves third-party evaluation for bias, safety, and accuracy.
Example: a model card explicitly documenting that a facial recognition system has a higher error rate for one demographic group than others, informing users of this limitation before deployment.
Applied Walkthrough
1
A company is deploying three different AI systems: one for high-stakes medical diagnosis decisions, one for real-time fraud detection, and one for spam email filtering.
2
For the high-stakes medical diagnosis system, human-in-the-loop is appropriate, given the severity of potential errors — a human should explicitly approve every decision.
3
For the real-time fraud detection system, human-on-the-loop might be more practical, letting the AI act quickly while a human monitors and can intervene if needed.
4
For the low-stakes spam filter, human-out-of-loop is appropriate, since the system has been extensively tested and the consequences of occasional errors are relatively minor — illustrating how the appropriate automation level scales with the stakes involved.
Exam Application
Exams test whether you can match the appropriate automation level (human-in-the-loop, human-on-the-loop, human-out-of-loop) to a described scenario based on its stakes, and whether you understand model cards and algorithmic auditing as key documentation and oversight tools for responsible AI deployment.
⚠ Common Trap
The most common trap is assuming full automation (human-out-of-loop) is always the most efficient, desirable end goal. The appropriate level of human oversight scales specifically with the stakes involved — high-stakes decisions warrant human-in-the-loop despite being slower and more costly, precisely because the cost of an error is so much higher.
✓ Quick Self-Check
1. What does human-in-the-loop (HITL) require?
A human explicitly approves every single AI decision before it takes effect.
Tap to reveal / hide
2. What does human-on-the-loop (HOTL) involve?
The AI acts autonomously in real time, while a human monitors and can intervene if needed.
Tap to reveal / hide
3. When is human-out-of-loop (fully autonomous) operation appropriate?
Only for very low-stakes decisions that have undergone extensive prior testing.
Tap to reveal / hide
4. What does a model card document?
A model's capabilities, limitations, intended use, and performance across different demographic groups.
Tap to reveal / hide
5. What is algorithmic auditing?
Third-party evaluation of an AI system for bias, safety, and accuracy.
Tap to reveal / hide