Overview
This series asks whether delegating work to AI actually transfers responsibility. It explains why AI–human team framing can pull perceived responsibility toward a human, while severe oversight constraints can weaken the human’s perceived control. A four-part model links autonomy, meaningful choice, knowledge, and causal influence, then carries those distinctions into organizational design and a practical audit. The experiments support conditional attribution patterns in banking scenarios and a simulated drone task, but actual accountability still requires examining authority, preventability, system design, and evidence beyond the studies.
Learning outcomes
- AI involvement does not move responsibility in one direction; the result depends on where meaningful control is located.
- In AI–human team scenarios, the human received more responsibility attribution because the AI appeared constrained and the human appeared discretionary.
- In the simulated drone task, severe choice restriction improved accuracy and speed while weakening perceived responsibility, causality, and knowledge.
- The studies support a control-and-understanding framework, not a universal rule or a legal allocation of liability.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
Does delegating a decision to AI transfer responsibility?
No. Delegation can redistribute tasks and influence, but responsibility still depends on who could understand, choose, and change the outcome, as well as who designed the surrounding conditions.
Why can AI make humans seem more responsible?
In the AIHR loan-processing scenarios, people attributed more responsibility to the human teammate because the AI appeared constrained and the human appeared to retain discretion; this is an attributional finding, not proof of actual fault.
Can human oversight remain meaningful with fewer options?
Yes, if the remaining action can affect the outcome and the overseer can understand and use it; fewer options can improve speed and accuracy, but a purely confirmatory choice weakens causal control.
How should organizations assess responsibility in AI workflows?
They should map decision rights, real and timely choices, epistemic access, causal influence, escalation, and the contributions of the developer, deployer, interface, and organization; the audit is evidence-informed, not a legal test.