Literature Review
Our theoretical framework, grounded in a pragmatic paradigm (Islam, 2022), identified three core dimensions of exclusion guiding the entire project design.
Socio-economic
Gender · Race · Finance
AI Literacy
Self-Efficacy Gaps
Structural
Policy & Funding
Data Scraping
Live PhD listings were collected programmatically to build a structured market analysis dataset of the AI postgraduate landscape in the UK.
Listings were scraped from FindAPhD.com to capture a representative snapshot of AI PhD opportunities across UK institutions. Data points included funding status, eligibility criteria, stipend information, grade requirements, and supervisor details — enabling systematic market analysis of structural access barriers.
Training Intervention
The "Getting Started in AI" course, run by the Jean Golding Institute (JGI), served as the core intervention. Participants were self-selecting undergraduates who attended the programme.
Eligibility Criteria
Session 1
Cohorts A & B- AI concepts & foundations
- Pre-training survey completed
- Post-training survey window opens
Session 2
Cohorts A & B- Applied AI skills
- Interview invitations issued
- Post-training survey reminder
All confirmed attendees were contacted by email about optional surveys. Time was allocated within sessions for completion. The research team supervised both cohorts and confirmed no significant differences in delivery between them.
Quantitative Surveys
A longitudinal pre- and post-test design tracked quantitative shifts in student attitudes and AI literacy before and after training.
Hypotheses Tested
AI-LITS (Short Version)
AI Literacy Scale (Hornberger et al., 2025) — measured pre/post training; question order randomised in post-survey to reduce recognition bias.
DAISY SES Indicators
Ethnic background, gender, caregivers' education, and current occupation — based on DAISY (2022) and Census 2021 categories.
Likert Scales
Five-point agreement scales measuring attitudes towards using and studying AI, and motivation for attending the training.
Semi-structured Interviews
Survey participants were invited to attend structured online interviews (up to 60 mins) via Microsoft Teams following Session 2.
Research Questions
Each interviewer was accompanied by an observer taking field notes. Interviews were automatically transcribed in Microsoft Teams, then reviewed and corrected by hand before analysis.
View Qualitative FindingsData Analysis Strategy
The final phase synthesised both data streams to triangulate findings across quantitative and qualitative sources.
Thematic Analysis
Deductive approach (Clarke & Braun, 2017) applied to interview transcripts. Shared codebook reviewed asynchronously and in dedicated meetings.
Content Analysis
Abductive approach (Erlingsson & Brysiewicz, 2017) applied to free-text survey responses; frequency analysis across the dataset.
Descriptive Stats
Mean, mode, and range compared pre- and post-training survey scores across AI literacy items and agreement scales.
Note: The JGI training focused on AI literacy rather than all barriers identified in the literature. This partial misalignment meant some codes (e.g., gender inequality, confidentiality concerns) were not addressed in post-training interviews, and admission policy questions could not be explored with an undergraduate sample.
Ethical Compliance
Approved by the University of Bristol Faculty of Science and Engineering Ethics Committee (Ref: 29619). The average pre-training survey took approximately 25 minutes; interviews lasted up to one hour. All data will be anonymised and securely held in accordance with the committee's requirements.