Thematic Deep Dive

Literature Review &
Theoretical Framework

An interdisciplinary review identifying how socio-economic status, AI literacy, and structural barriers create compounded exclusion.

1. Socio-economic

Systemic inequalities in gender, race, and finance filter out talent before they apply.

2. AI Literacy

Individual barriers where AI literacy and self-efficacy act as psychological gatekeepers.

3. Structural

Analysis of institutional policy, financial stratification, and market mechanisms.

Introduction

The present review aims to identify potential barriers to PhD/Doctoral studies on AI in Bristol and the United Kingdom. To understand why the doctoral pipeline remains exclusive despite high demand, this report examines barriers across three interconnected dimensions: Socio-economic Barriers, Individual Barriers (AI Literacy), and Structural Barriers.

Defining 'AI' for Research: Empirical definitions of AI remain scarce. We conceptualise AI as computer systems designed to perform tasks typically requiring human intelligence—learning, problem-solving, and reasoning. AI is generally used to process existing data (processing AI) or to create new content (generative AI).

1. Socio-economic Barriers to Doctoral Studies In AI

1.1 Reflecting On the Literature

The literature search was conducted with support from the University of Bristol subject librarian. The review revealed a substantial gap in studies directly examining socio-economic barriers to AI studies in higher education, particularly at the doctoral level. Most identified studies focused on skills gaps in relation to labour-market demands and proposed strategies to mitigate their impact, heavily dominated by studies conducted in the United States.

1.2 Socio-economic Barriers to AI Adoption

While doctoral students generally recognised the efficiency and usefulness of AI tools, ethical concerns emerged as major barriers. Participants expressed ambivalence towards AI’s transformative potential, reporting negative emotional responses associated with a perceived loss of agency and autonomy.

1.2.1 Gender And Its Intersections

Approximately 73.4% of AI PhD supervisors are male, reinforcing a "male discipline" perception. Women often report lower familiarity with AI tools, largely due to limited access to training and institutional support.

Male Supervisors 73.4%

Source: MIOIR (2025) analysis of UK STEMM PhD theses.

1.2.2 Authorship

Women accounted for only 10.26% of authors in leading AI-related systems conferences worldwide. Male-authored publications also tend to receive more citations, partly due to men’s greater participation in larger, male-dominated collaboration networks.

10.3%
Female Authors

1.2.3 Location and Gender

Geography plays a major role. A study found half of AI professionals concentrated in just four countries (USA, France, Germany, UK). Geographic disparities intersect with gender.

Female Co-authorship (%)
Netherlands30%
Japan / Singapore10-16%
Concentration Hub

50% of global AI professionals are located in just 4 countries, creating a geographic filter for opportunities.

1.3 Race and Ethnicity

In the UK, 74% of postgraduate research students were white, compared with 5% Black and 9% Asian students. Economic barriers were identified as a major factor affecting completion.

1.3.1 The Socio-economic Inequalities Loop In AI

Barriers to AI doctoral education reflect broader inequalities within the AI ecosystem. This creates a feedback loop where existing power imbalances in AI research influence future educational opportunities.

Figure 1: The Inequality Loop

AI PhD Studies
Socio-economic Barriers
Gender
Geography
Race/Ethnicity
Underrepresentation & Unequal Progression
AI Research & Authorship
AI Labour Markets & Governance

1.4 Conclusion and Recommendations

There is an urgent need to expand EDI-focused research in AI doctoral education. Educators must actively challenge exclusionary academic cultures. Future research should examine the psychological and behavioural dimensions of AI adoption, including its impact on academic identity.

2. AI Literacy

AI literacy is defined by Long and Magerko (2020) as: “a set of competencies that enables individuals to critically evaluate AI technologies; communicate and collaborate effectively with AI; and use AI as a tool online, at home, and in the workplace”.

2.1 AI Literacy Development

AI literacy does not develop automatically. Institutional and psychological filters govern its trajectory.

  • 2.1.1 Institutional Access and Technical Gatekeeping Doctoral study in AI globally continues to carry implicit expectations of technical preparedness. Extended sequences of technical training reinforce entry barriers for non-technical learners.
  • 2.1.2 Value Perception as a Motivational Filter Engagement is driven more by perceived utility than technical exposure. Framing AI mainly as a technical field can weaken professional identity for students seeking socially oriented roles.
  • 2.1.3 AI Self-Efficacy as a Psychological Boundary Stereotypes of the "tech geek" act as psychological boundaries. Low confidence is linked to early disengagement, particularly among underrepresented groups and women.

2.2 Intervention Design

AI literacy has a proven positive effect on career intention and interest. Evaluative literature in HE or pre-HE contexts tends to focus either on improving career interest, or on improving participants' AI literacy itself.

  • 2.2.1 Enhancing AI Career Interest Interventions often implicitly target AI literacy competencies. Mapping programme features to associated competencies suggests that participants are afforded opportunities to develop these skills.
  • 2.2.2 Barriers to Attendance The digital divide remains the most significant barrier to AI literacy development. Variability of internet access necessitates hybrid engagement modes, facilitating attendance for disabled students or those with caring responsibilities.

2.3 Conclusion and Recommendations

Given that participants' engagement with AI learning is cyclically enabled by their AI literacy and self-efficacy, and that AI literacy increases career interest, this concept is a useful focus for the design of training.

3. Structural Barriers

3.1 Shaping AI use on campus – global and UK perspective

The fragmentation of AI policy in HE is highlighted in the UK Russell Group study. Institutions employ a wide variety of approaches to determine what is permitted.

Policy Dimension University of Bristol (Co-Pilot) UCL (Designer) University of Oxford (Gatekeeper)
AI Literacy "Foundational." Focuses on long-term skills but warns against "learning loss". "Core Competency." Explicitly treats AI proficiency as a required future workforce skill. "Cautious Engagement." Focuses on limitations and risks (hallucinations/bias).
Staff Guidance "Co-Pilot Model." Staff use AI to sense-check. AI cannot replace core concepts. "Designer Model." Staff redesign assessment to either embrace or exclude AI. "Gatekeeper Model." Tutors define strict boundaries. Emphasis on Socratic defence.
Assessment "Transparency & Trust." Disclose why/when used. Links to EDI goals. "Three-Tier System." 1. Banned, 2. Assisted, 3. Integral. "Prohibited unless specified." Default stance is restriction.
Academic Integrity "Student Agency." Permitted only when it does not diminish capacity to derive meaning. "Acknowledgement." Focus on citation. Declare prompt and output. "Authorship Definition." AI cannot be an author. Submission as own work is plagiarism.

3.2 Conditions Beyond AI Policies

While AI use policies in HE shape academic practices, they do not operate in isolation. Wider structural barriers stemming from the broader STEM disciplines and socio-economic conditions often outweigh institutional guidance.

Finance & Accessibility

Financial costs remain a primary barrier. For a four-year PhD at Bristol, fees for home students exceed £20,000. A study found that 79% of PGRs require supplementary income to cover essential living costs.

Migration & Hidden Costs

Approximately 86.5% of international PGRs experience hidden costs including visa fees, National Health Service (NHS) surcharges, and educational resources, creating a "False Hope" for global applicants.

Spotlight: University of Bristol AI Position

Named the UK’s AI University of the Year in 2024, Bristol's strategy necessitates an encompassing approach to inclusion. Below is a summary of the framework for student AI engagement:

Area Key Focus
Pedagogical Integrating AI for learning support (revision, planning) while maintaining academic thinking and fact-checking.
Governance Consistent yet flexible policies where module leaders define specific guidance. Mandatory disclosure of AI use.
Operational Provision of training resources and citation support. Framing AI as an "assistive companion" rather than a substitute.

3.3 The UK PhD Landscape: Primary Analysis of Barriers to Entry

Analysis of 2,643 live listings from FindAPhD.com (Dec 2025) using NLP.

3.3.1 Financial Stratification

23.1%

Unfunded listings. This poses a significant financial barrier, segregating access by personal financial means.

3.3.2 Global Ambition

79.2%

Listings available worldwide. Contrasts with strict visa costs and funding caps.

3.3.3 Estimating Types of AI Projects

Applied AI53%
Tangential / Non-Core23%
Policy & Ethics8%

References

Looking for our full bibliography?

Download the complete reference list in BibTeX format.

Download .bib File