把任务交给AI以后,责任最终属于谁? / Who Is Responsible After a Task Is Delegated to AI?

Judgment in the Age of AI · Article 50

AI can advise, generate and execute, but it cannot bear legal or professional duties, compensate injured people or justify value choices as a person or institution can. “The system did it” describes a process; it does not answer who is accountable.

Ask along every link in the responsibility chain

  1. Selector: Who decided to use AI for this task?
  2. Configurer: Who chose data, model, permissions and thresholds?
  3. Reviewer: Who can inspect and stop mistakes?
  4. Decision-maker: Who turned the recommendation into action?
  5. Operator: Who monitors outcomes and responds to anomalies?
  6. Remedy owner: Whom can affected people approach for correction, appeal and redress?

Responsibility cannot stop at the interface

A provider is responsible for product design and claims; a deploying organisation for purpose and controls; and named people for their authorisations and decisions. Responsibilities can coexist. A long chain must not allow every participant to say “not me”.

Complete one final check before delegation

  • Is the objective legitimate and clearly stated?
  • Are permissions proportionate to risk?
  • Can results be inspected, stopped and corrected?
  • If it fails today, who will handle it tomorrow morning?

The essential judgement of the AI age is not refusing all delegation. It is retaining a complete path of human responsibility after delegation.

References


Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.