Kamus AI
Human in the loop
The principle that important AI decisions must be checked and approved by a person.
AI may prepare drafts, summaries or proposals. A person decides, especially on matters about people (hiring, grading, sanctions), money, law and health. The principle also means the person must really check, not just press approve.
Also called: Human in the loop, human-in-the-loop, human approval
Related terms
Appears in 47 lessons
- When AI goes wrong in hurried hands
- When not to automate
- People decide
- Automated decisions and people's rights
- AI does not decide a candidate's fate
- From interviews to closing the round
- Not legal advice: AI assists, the advocate answers
- Contract review with a playbook: AI flags, the advocate decides
- Prompt library: contracts and documents
- Legal and compliance template pack
- AI does not diagnose
- A new service day, part 1: roster, slots and SOP
- Safety decisions stay with people
- Drafting fleet and dispatch plans
- Late route case, part 2: from finding to SOP
- Mini project: a safe operations kit
- Segregation of duties and approval
- Prompts for invoices, tax and approval
- Month-end close, part 2: report and approval
- Mini project: a safe finance work kit
- Ready-to-use finance templates
- A person sends the reply, not a machine
- Customer chat reply templates
- Project: a safe shop kit for one product
- AI does not decide grades or a student's fate
- Feedback without exposing student work
- Messages to parents and school letters
- One teaching unit from start to reflection
- Class AI rules and academic integrity
- Project: a safe AI teaching kit for one unit
- The limits of AI in public service: it assists, it does not decide
- Decisions about citizens stay with officials, and must be fair
- Answering citizen questions and complaints with approved templates
- Prompt library: staff analyses, service templates and citizen complaints
- One service change, from data to announcement and FAQ
- Mini project: a communication kit for one of your agency's services
- A ready-to-use public service template pack
- Code from AI is a draft, and you own it
- Prompts for review, migrations and releases
- A release with a rehearsal and a way back
- Editors and CLI agents
- Plan, build, review
- When to use an agent, and how freely
- Project: a team instruction file and permission rules
- Stop conditions
- Case study: the planned command agent
- A human gate before release