Autonomy level and responsibility allocation must align
Aliases: responsibility alignment · authority accountability match · meaningful human control
What it is
Autonomy–responsibility alignment gives a role accountable for outcomes commensurate goal authority, information, intervention, time, and competence. A system cannot select and execute consequential action automatically, then assign all failure to a nominal supervisor.
Why it happens
Responsibility ordinarily follows capacity to foresee, influence, and prevent consequence. Autonomy alters that causal chain through speed, overridden input, and hidden evidence. Reusing manual-era responsibility tables separates nominal accountability from real controllability and suppresses feedback about system defects.
Studying it
Event scenarios can vary goal setting, choice, execution, visibility, and intervention time, measuring attribution and actual preventability. Log analysis should supplement moral judgement. Legal allocation varies by jurisdiction and requires qualified review.
Where it stops holding
Alignment need not place all responsibility on one party; manufacturer, deployer, manager, and operator control different risks. Human input can still causally contribute. Interface design supplies authority and evidence but does not itself determine law.
Applying it
- Map who sets goals, selects, executes, stops, and maintains each function against failure consequence.
- Verify that accountable roles can obtain information and prevent harm under worst conditions; otherwise reallocate authority or responsibility.
- Tabletop incidents should compare written responsibility, actual control chain, and logged evidence.