# Responsible AI

Chapter 16 of Loop Engineering argues that the architecture is a tool and tools carry obligations: disclosure, recourse, opt-out, data transparency, measured calibration, meaningful human oversight, decision records, and operator accountability.

This page extends Ch 16 in two directions: cross-disciplinary reading paths (psychology, philosophy, cognitive science, STS) and the operator obligations checklist mapped to the architectural layers.

### Cross-disciplinary reading paths

- Psychology - cognitive load (Sweller), trust calibration (Lee & See 2004), automation bias (Parasuraman & Riley 1997; Parasuraman & Manzey 2010), attention (Wickens et al.).

- Philosophy - moral agency (Floridi), consent in AI partnerships, the boundary between operator obligation and architecture obligation.

- Cognitive science - unified cognitive architectures (Anderson's ACT-R, Newell's SOAR), dual-process models (Kahneman).

- STS / sociology of work - algorithmic harm (Eubanks), accountability voids (Mittelstadt), Ironies of Automation (Bainbridge 1983).

### Operator obligations checklist

The Ch 16 obligations mapped to which architectural layer carries each one:

- Disclosure - dialog interface, presentation layer.

- Recourse - fleet layer (non-AI escalation), human-in-the-loop posture.

- Opt-out - dialog interface, consent capture.

- Dataset transparency - memory layer, training-data documentation.

- Calibration support - dialog interface (uncertainty signals), evaluation layer (measured confidence against outcomes), prefrontal (visible evidence and limits).

- Meaningful human oversight - materiality dial, authority to intervene, adequate time and information, appeal and escalation paths.

- Decision record - defensibility bundle, action log, decision context, approvals, and outcomes; protocol traces alone are insufficient.

- Operator accountability - deployment context, fleet governance.