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
- Selector: Who decided to use AI for this task?
- Configurer: Who chose data, model, permissions and thresholds?
- Reviewer: Who can inspect and stop mistakes?
- Decision-maker: Who turned the recommendation into action?
- Operator: Who monitors outcomes and responds to anomalies?
- 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.