Chat assistants ≠ collaboration
Most “AI copilots” are Q&A beside the work. A collaboration thesis must change how two agents (human + AI) share state, decide, and hand off.
Research · PhD topic
AI & Human Collaboration
How people and AI systems work together in operational settings — with clear oversight, handoff, and shared awareness.
A defendable doctorate sits between generic AI copilots (chat beside the work) and full autonomy claims (remove the human). The research object is joint work: shared situational awareness, authority, handoff, and measurable collaboration quality in real operational settings.
With an aviation industry and IT infrastructure background, the natural exam settings are airline/airport ops centres and infrastructure NOC / SRE rooms — both already practise human-in-the-loop culture (CRM, checklists, change boards, incident command).
HITL
Human-in-the-loop is the claim, not a slogan
2 domains
Aviation ops + IT infrastructure as exam settings
1 track
Pick one disciplinary home below
3–4 yrs
Doctoral plan with yearly checkpoints
Working title
Human–AI Collaboration in Safety-Critical Operations: Shared Awareness, Authority, and Handoff
Swap the subtitle when you lock a track below. Keep “collaboration” and “operations” in the title: they are the claim, not decoration.
Field settings
Pick one primary setting; optionally use the other as a transfer check.
| Setting | What it is | Why it matters | PhD leverage |
|---|---|---|---|
| Airline / airport ops centre | Dispatch, disruption recovery, turnaround coordination | Safety-critical, time-pressed, multi-actor | Strong fit if aviation is your primary industry home |
| IT infrastructure NOC / SRE | Incident response, change windows, runbook automation | High change rate, tool sprawl, on-call pressure | Strong fit if infra ops is your primary craft |
| Cross-domain bridge | Same collaboration patterns in both settings | Comparability of handoff, trust, and oversight | Best if you can access both field sites or proxies |
Research gap notes
Aviation and infra already have HITL culture
Checklists, CRM, on-call, change boards. The gap is how generative AI joins those rituals without dissolving accountability.
Your background is the advantage
Aviation industry + IT infrastructure lets you ground claims in real ops language — escalation, MEL/NOTAM-style constraints, change freezes, post-incident review — instead of toy tasks.
Choose one PhD track
One thesis cannot cover all four. Lock a track and a primary field setting before building prototypes.
Human–AI Collaboration in Safety-Critical Operations: Shared Awareness and Joint Decision Support
Claim
AI assists operators only when it improves shared situational awareness and joint decision quality — not when it merely answers questions beside the console.
Research questions
- How should an AI partner present uncertainty, alternatives, and evidence so operators keep a correct mental model under time pressure?
- Which handoff protocols (AI → human, human → AI) reduce missed cues in aviation ops or IT incident rooms?
- When does collaboration improve recovery time or decision quality versus a human-only or automation-only baseline?
Method
Field or high-fidelity simulation studies in one primary setting (ops centre or NOC). Scenario library + think-aloud + task metrics. Prototype a collaborative assistant with explicit uncertainty and handoff UI. Compare against status quo tools.
Contribution
A collaboration model and interaction patterns for human–AI joint ops work, grounded in one safety-critical domain and tested with measurable outcomes.
Avoid
A chatbot demo with satisfaction surveys only. Examiners want decision quality, error modes, and handoff failures.
Fit for aviation + IT infra
Best default for aviation + IT infra: you already know console work and escalation culture.
Human-in-the-Loop Control for AI-Assisted Infrastructure Operations
Claim
Infrastructure automation fails socially before it fails technically: authority, approval, and rollback must stay inspectable when AI proposes or executes change.
Research questions
- What approval and veto patterns keep humans accountable when AI drafts runbooks, changes, or remediations?
- How do operators calibrate trust after false positives and missed alerts over repeated incidents?
- Which telemetry and explanations make an automated action auditable after an outage?
Method
Instrument a lab NOC (or partner environment) with AI-assisted change/remediation proposals. Controlled incident drills. Log approval latency, override rate, and post-incident auditability. Interview on-call staff.
Contribution
A HITL control pattern for infra AI: authority levels, audit trail, and trust calibration metrics usable by SRE/NOC practice.
Avoid
Another auto-remediation agent paper that removes the human without measuring accountability.
Fit for aviation + IT infra
Strong if your IT infra depth is deeper than aviation ops access.
Trust, Authority, and Override in Human–AI Operational Partnerships
Claim
Collaboration breaks when authority is ambiguous. The thesis object is the authority contract — who may act, who must confirm, and how override is recorded.
Research questions
- How do operators assign authority to AI suggestions under different risk levels (advisory vs gated vs autonomous)?
- What override UX and policy reduce both automation bias and alert fatigue?
- How should organisations train and govern collaborative AI so responsibility stays clear after failure?
Method
Mixed methods: policy analysis + scenario experiments varying authority levels + incident retrospectives (real or simulated). Produce an authority taxonomy and evaluation protocol.
Contribution
An authority model for collaborative AI in ops, with measurable override behaviour and governance recommendations.
Avoid
Generic “trust in AI” surveys detached from a concrete operational workflow.
Fit for aviation + IT infra
Good if you like policy, governance, and safety culture more than building schedulers.
Evaluating Human–AI Collaboration in Complex Operational Work
Claim
Accuracy of the model is not the success metric. Collaboration quality needs joint metrics: shared awareness, handoff success, recovery time, and residual risk.
Research questions
- Which metric suite predicts real ops outcomes better than model accuracy alone?
- How transferable are collaboration metrics between aviation ops and IT incident response?
- What experimental design lets a lab study claim external validity without claiming a full airline trial?
Method
Build a reusable evaluation harness (scenarios, coding schemes, metrics). Run studies in one domain; validate transfer claims with a second lightweight domain or expert review panel.
Contribution
An evaluation framework and open scenario pack for human–AI collaboration in operational settings — usable by other labs and by NEXORA-iQ class work.
Avoid
Leaderboard-style LLM benchmarks that ignore the human partner.
Fit for aviation + IT infra
Strong if you want a methods-heavy thesis that supports later applied studies.
Doctoral plan
Four-year shape with near-term lab checkpoints. Adjust dates to your university calendar; keep the sequence.
Frame and literature
- Confirm disciplinary home and one primary field setting (aviation ops or IT NOC)
- Literature map: HITL, CSCW, aviation human factors, SRE/incident literature
- Gap statement and working title locked
- Ethics / access plan for participants or simulation proxies
- Lab two-page proposal published; outline for full paper
Prototype and pilots
- Build the collaboration prototype (awareness, uncertainty, handoff, audit)
- Scenario library for drills (disruption recovery or major incident)
- Pilot study (n small) — refine instruments
- Write methods chapter draft; present at lab hours / community seminar
Main study
- Main experiment or field study
- Analyse decision quality, handoff failures, trust/override behaviour
- Compare against human-only and/or automation-heavy baselines
- Draft results + discussion; submit lab working paper
Thesis and venue
- Thesis write-up and revisions
- Venue paper (HCI, CSCW, human factors, or systems venue as fit)
- Open materials for NEXORA-iQ Learning Hub / lab library where allowed
- Viva / defence preparation
Near-term milestones
| When | Milestone | Track |
|---|---|---|
| Oct 2026 | Topic locked: AI & Human Collaboration as primary PhD | Decision |
| Nov 2026 | Choose one track + primary field setting | Research |
| Dec 2026 | Literature map + full outline checkpoint | Research |
| Mar 2027 | Prototype v1 + pilot protocol ready | Build |
| Sep 2027 | Pilot complete; instruments fixed | Study |
| Jun 2028 | Main study data collected | Study |
| Dec 2028 | Lab working paper + thesis draft core chapters | Write |
| 2029–2030 | Thesis submission and venue paper | Submit |
Proposal summary
This proposal frames human–AI collaboration as joint operational work in aviation and IT infrastructure settings. The candidate locks one track (recommended: operational collaboration), builds a prototype that supports shared awareness and explicit handoff, and evaluates collaboration quality beyond model accuracy.
Problem
Organisations deploy AI assistants into ops rooms without a clear authority contract or handoff protocol. Operators either over-trust suggestions (automation bias) or ignore them (disuse). Neither outcome is collaboration.
Aim
Design and evaluate interaction and authority patterns that improve joint decision quality and recovery under time pressure, while keeping human accountability inspectable.
Objectives
- Map collaboration failures in one primary ops setting.
- Prototype shared awareness, uncertainty display, and handoff/override.
- Measure collaboration outcomes against human-only and automation-heavy baselines.
- Publish an open evaluation pack for the lab series.
Core reading anchors
- Endsley — situational awareness. Foundational SA model; use carefully, extend to human–AI shared awareness.
- Parasuraman & Riley — automation use/misuse/disuse. Classic for trust calibration and automation bias.
- CSCW / joint activity (Klein, Feltovich, et al.). Common ground and coordination costs — useful vocabulary for “collaboration”.
- Aviation CRM and human factors literature. Crew resource management patterns for multi-agent ops under stress.
- SRE / incident response studies. Coordination under uncertainty in IT operations; pair with AI assistance carefully.