Aerospace & AeronauticsDoctoral Program

Proposed dissertation · Aviation Practice

From ground school
to pilot judgment

Ahmed Allem PhD start year 2020

Can pilots carry what they learn with AI assistance into unfamiliar planning tasks? A proposed course plan and dissertation built around Allem’s work in flight education, pilot software, and aviation access.

Research design manuscript · 200 pages

Traceable AI instruction for pilot judgment

Prepared for Ahmed Allem’s review. The draft develops an independent tutor benchmark, a randomized study of delayed learning, and an assessment in a separate simulated planning environment.

The companion provides 72 fictional cases, scoring examples, and a Python script. The draft documents proposed studies and identifies the evidence needed for a final dissertation.

Plan of study

Coursework that builds the dissertation

Each module produces something the dissertation can use. The proposed sequence uses the current catalog. It covers all six shared-core modules and four Aviation Practice modules. The Registrar holds the official academic record. Eight modules carry a listed subtotal of 50 ECTS. Confirm AE-809, AE-810, dissertation registration, and cohort requirements with the program.

Calibration · first-year plan

Foundations and measurement

Elevation · qualifying and proposal

Candidacy and study design

Use the degree checklist for formal milestones and the Mission Studio rubric for feedback on the artifacts.

Portfolio

What each project contributes

Allem’s public portfolio supplies five aviation case studies. The proposed dissertation concentrates on two platforms and uses the others to put design decisions in context.

Primary study setting

Aviation Infinity

AI-supported aviation theory instruction across 2D, desktop 3D, and optional headsets.

Proposed artifact. Versioned lessons, a common assessment, tutor failure taxonomy, and an instructor handoff rubric.

Transfer study setting

Avioyx

Flight planning and pilot workflow software provides a setting for unfamiliar planning tasks in simulation.

Proposed artifact. A fixed scenario bank, blinded scores for recognition of planted issues, and a workload measure.

Supporting case

AvioSharing

Flight-sharing interfaces raise questions about which responsibilities passengers and pilots understand.

Proposed artifact. An interface comprehension protocol and a scenario-based review memo.

Supporting case

Want To Be a Pilot

A mentorship case for studying what learners bring to an instructor and what they do next.

Proposed artifact. An anonymized or synthetic mentoring handoff and a rubric for the clarity of the next learning action.

Supporting case

New Pilot Shop

A case for how a learner interprets suitability, compatibility, and evidence when selecting equipment.

Proposed artifact. A small usability task and explanation rubric drawn from simulated procurement decisions.

Related work

Early tools, other industries, and mentorship

The academy’s public history describes electronic flight bag, E6B, and navigation tools. A versioned calculator or archived output, if available, could serve as an AE-801 reproducibility exercise.

View the wider portfolio

LegalAgento, MedAgento, PowerAgento, TravelAgento, and CodeAgento provide examples of expert workflows. Bocads, Surfyx, ClickAi, Babonbo, BorderBot, and Comensalaqui add product and marketplace comparisons. Cal Hacks and HackIfy.io provide mentorship and prototyping context. Use public case descriptions to compare design decisions and assess aviation outcomes in the studies below.

Dissertation proposal

A study of learning that transfers

Working title. Traceable AI Instruction for Pilot Judgment: Delayed Learning and Transfer to Simulated Flight Planning.

Research question. Does a source-grounded AI tutor improve how student pilots recognize and explain unfamiliar planning discrepancies two weeks after instruction, compared with the same course and standard feedback?

Proposed contribution. An instructor-rated tutor benchmark, a controlled estimate of delayed learning, and a separate check that the learning carries over to a different simulated planning environment. The study would report where the effect is small, uncertain, or limited to particular tasks.

  1. Establish instructional reliability

    Design. Build a panel-reviewed bank of synthetic theory prompts with expert reference answers. Freeze the syllabus sources, model configuration, and scoring rubric. Compare the Aviation Infinity tutor with source-backed static feedback, including answerable and ambiguous prompts. Two independent instructors score blinded outputs for accuracy, source support, and appropriate referral when information is missing. Resolve disagreements with a third reviewer.

    Decision. Preregister the acceptable rate of consequential errors and its uncertainty before scoring. Proceed to participant studies only after faculty accept the benchmark results. Any material tutor update requires a versioned regression check before use in the trial.

  2. Run a controlled delayed-learning study

    Design. Recruit consenting adult student pilots through participating training providers after approval. Randomize to guided AI feedback or standard instructor-designed feedback within the same 2D Aviation Infinity lesson. Match lesson objectives, practice time, source materials, and instructor availability. Participant interfaces may reveal assignment, so blind outcome scorers and use study-arm codes for analysis.

    Primary outcome. At day 14 (proposed window ±3 days), score novel theory scenarios on whether learners identify the issue, explain the relevant assumption, and justify their decision using provided sources. Analyze the difference between assigned groups with a confidence interval, accounting for baseline score and the randomization strata. Preregister one minimum educationally meaningful difference, a pilot-informed sample size, attrition assumptions, and a missing-data sensitivity analysis.

    Secondary outcomes. Capture immediate performance, workload, accessibility barriers, and important errors. Explore desktop 3D or headsets separately only when a spatial objective and feasibility justify the comparison. Hold instructional support constant for a modality study.

  3. Test transfer in a separate planning environment

    Design. Use the same randomized groups in a held-out Avioyx sandbox with fictional flights and a new scenario bank. Both groups receive identical reference material and complete the assessment without access to the study tutor. Use a common rubric for identifying pre-specified discrepancies, explaining assumptions, and deciding what information to verify. Counterbalance task order and blind independent scorers.

    Interpretation. Treat this as a prespecified secondary transfer endpoint from the main trial. Report scoring agreement, uncertainty, and failure patterns. Use the evidence to bound claims to performance on the simulated tasks and studied population.

Freeze the participant flow, task order, scoring manual, and analysis plan before outcomes are examined. Use the AI-use plan and committee evidence checklist to record changes.

Dissertation literature

What earlier dissertations teach us

Seven selected doctoral precedents shaped the protocol. Stanford defense and author records are labeled by source type. Stanford Libraries identifies SearchWorks as the catalog for dissertations preserved in the Stanford Digital Repository.

Research synthesis. A systematic review of XR in pilot training identified gaps in robust transfer studies. That supports separating feasibility, measured learning, and transfer. The Stanford-affiliated LeRAAT research paper describes AI assistance during simulation. Here the tutor is removed for the final task so the assessment measures what the trainee can carry forward independently.

How to extend the archive search

Search scope. This is a targeted review of public records as of November 16, 2026. Direct catalog enumeration through Stacks and SearchWorks was unavailable during this review. Use the official SearchWorks search guide to extend coverage, including catalog records whose full text is restricted.

  1. Search Stanford’s catalog for Aeronautics and Astronautics and limit to theses or dissertations where the catalog permits. Run topic searches for pilot training simulator transfer, aviation human automation workload, and autonomous systems simulation validation. Check the degree and department on each record.
  2. Search MIT Theses and the Embry-Riddle doctoral collection with the same terms. Check references in the closest records.
  3. Record matches and exclusions by stable link, institution, year, degree, review stage, and full-text access. Deduplicate by author, title, and year. Put articles and master’s theses in a supporting bibliography.

Manuscript structure

From proposal to dissertation

The proposed dissertation uses one research question across three studies and links each chapter to the protocol, source material, or analysis that supports it.

  1. Introduction. Define pilot judgment, traceable feedback, delayed learning, and transfer. State what the studies can establish.
  2. Literature and theory. Connect tutoring, human–automation interaction, training transfer, and simulation validation. Document the search and its limits.
  3. Materials and common methods. Describe the platform versions, study population, conditions, rubrics, ethics review, and founder-conflict controls.
  4. Instructional reliability. Present Study 1, scorer agreement, error categories, and the decision about proceeding.
  5. Delayed learning. Present Study 2, participant flow, primary estimate, confidence interval, and planned sensitivity analyses.
  6. Simulation transfer. Present Study 3 and explain which task features and settings the findings cover.
  7. Synthesis and recommendations. Reconcile the studies, discuss limitations and alternative explanations, and identify what an independent team should replicate.

Research practice

AI is allowed in this dissertation plan

Allem may use AI during literature discovery, drafting and editing, coding, pilot analysis, and reproducibility checks. Record substantial assistance and verify the resulting work before it enters the dissertation.

Writing and sources
Check every substantive claim and quotation against the underlying source. Resolve references against the original record. Allem remains responsible for the argument, interpretation, and final text.
Code and analysis
Keep prompts or a reproducible description of substantial AI assistance, code versions, test output, and independent review. Use synthetic or approved data in tools and environments cleared for those data.
AI as the intervention
Record the tutor model, version, prompts, source material, and configuration separately from AI used to write or analyze the dissertation. Keep assessors independent for the primary outcome.
Disclosure and defense
For each material use, log the tool, version, date, task, affected artifact, data category, and verification. Include a dissertation disclosure and bring the log to committee review. The candidate explains and defends the work under the program’s examination rules.

The disclosure and verification workflow draws on UC Santa Barbara’s graduate guidance. The Research.University handbook governs the evidence record.

Aviation Futures

Published foundations

These three pieces are listed on Allem’s journal author page. They inform the studies proposed here. Each article has its own editorial and evidence record.

Review and deposit

Evidence the committee can inspect

The portfolio documents the design context. The dissertation’s contribution would come from the new evaluation, the evidence it generates, and the limits the committee can examine.

At proposal
Obtain faculty approval for the plan, formal determinations for human-participant work and data access, and a written plan to manage Allem’s role as founder. Assign recruitment, scoring, and analysis checks to people independent of the product team. Document who can see unblinded data.
At defense
Present a preregistration or dated protocol, versioned tasks, deviations, uncertainty, evidence of reproducibility, and a clear account of individual contributions. Report null and adverse findings alongside any improvements.
At deposit
Prepare an integrative manuscript, a public abstract, and the materials cleared for release. Obtain the official deposit instructions through the program and Registrar.