Human-in-the-loop
Human-in-the-loop is a way of designing AI systems so that a person stays involved in their decisions: checking outputs, approving actions or stepping in when something goes wrong. It keeps a named human accountable for outcomes, with the level of involvement matched to the risk and reversibility of what the AI is doing.
In more detail
Human-in-the-loop, often shortened to HITL, describes any AI workflow in which a person is deliberately placed to check, approve or intervene in what the system does. The principle behind it is accountability: an AI system cannot be held responsible for an outcome, but an organisation and its people can. Most workflows follow one of two collaboration modes. In the first, AI recommends and a human decides, so nothing happens until a person acts. In the second, AI executes and a human monitors, sampling the work, handling exceptions and able to stop or reverse actions; this is sometimes called human-on-the-loop.
Choosing between them
The right mode depends on how much harm a wrong decision could cause, whether the action can be undone, how many decisions are involved and how reliable the AI has proved on the task. Many teams start a new agent in the first mode and move defined categories of work to the second as evidence builds. The most common failure is a reviewer who is not really deciding: people approving hundreds of recommendations a day tend to accept them without scrutiny, a pattern known as automation bias.
How AND Digital approaches it
AND Digital holds that both collaboration modes must keep a human accountable for the outcome, and its approach to building with AI embeds strict human-in-the-loop governance.
Ask about this
Related terms
Related capabilities
- 01
AI, intelligent systems & agentic operations
We help you realise and execute the potential that lies in AI by combining human ingenuity with intelligent systems and AI agents to unlock new possibilities for your business. AI can transform how your organisation creates products, makes decisions and gets work done. But often, AI changes quicker than your organisation can deliver.