Human–AI Collaboration in Safety-Critical Operations: Shared Awareness, Authority, and Handoff

1. Problem

Organisations are placing generative AI assistants into operational rooms — airline and airport operations centres, and IT infrastructure NOC / SRE environments — without a clear model of collaboration. Most deployments are chat beside the console: the human remains solely responsible for state, while the model answers questions. Under time pressure this produces two failure modes: automation bias (operators accept weak suggestions) and disuse (operators ignore the assistant after early false positives). Neither is joint work. A doctorate is needed on how humans and AI systems share situational awareness, negotiate authority, and hand off control so decision quality and accountability improve together.

2. Aim and research questions

Aim: Design and evaluate interaction and authority patterns for human–AI collaboration in one primary safety-critical operational setting (aviation ops or IT infrastructure operations), with optional transfer checks to the second domain.

1. How should an AI partner present uncertainty, alternatives, and evidence so operators keep a correct mental model under time pressure?
2. Which handoff and override protocols reduce missed cues while keeping human accountability inspectable?
3. When does collaboration improve recovery time or decision quality versus human-only or automation-heavy baselines?

3. Contribution

A collaboration model and evaluated interaction patterns for joint operational work — not a new foundation model. Expected artefacts: authority taxonomy (advisory / gated / autonomous), prototype supporting shared awareness and audited override, scenario pack, and metric suite beyond model accuracy (shared awareness, handoff success, recovery, residual risk).

4. Method (overview)

Mixed methods. Year 1 locks literature and access/ethics. Year 2 builds a collaboration prototype and runs pilots. Year 3 conducts the main study in high-fidelity simulation or partnered field drills. Year 4 writes the thesis and a venue paper (HCI, CSCW, human factors, or systems). Baselines: human-only workflow and automation-heavy assistant without explicit handoff design.

5. Why this candidate background

Aviation industry practice and IT infrastructure operations both already encode human-in-the-loop culture: crew resource management, checklists, escalation, change freezes, and post-incident review. That background supplies authentic task language and failure modes that toy chatbot evaluations miss. The thesis uses one domain as primary and may treat the other as a transfer check rather than claiming universal coverage.

6. What this thesis is not

It is not a generic “trust in AI” survey, not an airline product, not a fully autonomous ops agent, and not an LLM leaderboard paper. Model quality matters only insofar as it changes joint human–AI work.

7. Near-term plan (lab checkpoints)

8. Indicative references

1. Endsley, M. R. — situational awareness in dynamic systems.
2. Parasuraman, R., & Riley, V. — humans and automation: use, misuse, disuse, abuse.
3. Klein, G., Feltovich, P. J., Bradshaw, J. M., & Woods, D. D. — common ground and coordination in joint activity.
4. Aviation CRM / human factors literature on multi-crew decision making under stress.
5. SRE and incident-response studies of coordination under uncertainty in IT operations.

NEXORA-iQ community lab working paper — educational research track, not a commercial consultancy proposal. Topic brief: /research/ai-human-collaboration/