Human-on-the-loop (HOTL) describes a level of human control over autonomous systems—particularly weapons systems, surveillance tools, and AI decision-support—in which a machine selects and executes tasks on its own, but a human operator monitors performance and retains the ability to halt or override actions. It sits between human-in-the-loop (where a human must approve each action before it occurs) and human-out-of-the-loop (where the system acts with no human ability to intervene).
The term is central to debates over lethal autonomous weapons systems (LAWS) at the UN Convention on Certain Conventional Weapons (CCW) Group of Governmental Experts, which has met in Geneva since 2017. Many states and the International Committee of the Red Cross (ICRC) argue that meaningful human control requires more than supervisory oversight, citing concerns about automation bias (operators deferring to machine recommendations) and the speed of machine decisions outpacing human reaction time.
Real-world examples often cited include:
- The Aegis Combat System and the Phalanx CIWS, naval defense systems that can engage incoming threats automatically while operators supervise.
- Israel's Iron Dome interceptor, which runs largely autonomously during salvos.
- Patrol drones and border-surveillance AI deployed by various states.
Critics, including the Campaign to Stop Killer Robots and scholars such as Noel Sharkey, argue HOTL is often nominal control: when systems operate at machine speed or across many simultaneous targets, a single supervisor cannot realistically exercise judgment over each engagement. Proponents counter that HOTL preserves human accountability while allowing operational tempo that human-in-the-loop control cannot match.
The distinction matters legally because international humanitarian law—principles of distinction, proportionality, and precaution under Additional Protocol I to the Geneva Conventions—presumes a human decision-maker capable of contextual judgment. Whether HOTL satisfies that requirement remains contested.
Example
During CCW discussions in Geneva in 2023, several delegations argued that defensive systems like Israel's Iron Dome, which operate under human-on-the-loop supervision, should be distinguished from offensive autonomous weapons in any future regulatory framework.
Frequently asked questions
In human-in-the-loop, a person must approve each action before the system executes it. In human-on-the-loop, the system acts autonomously and the human can only intervene to stop or override it.
Keep learning