Introduction: Why AI Dissertation Topics Are Worth Your Attention Right Now

If you are searching for artificial intelligence dissertation topics, you are in the right place — and you have chosen the right moment. AI is no longer a niche subfield of computer science. It is reshaping healthcare, law, education, recruitment, climate policy, and creative industries simultaneously.

For UK university students in 2025, this creates a genuinely exciting problem: the field is moving so fast that genuinely novel research gaps open up almost every semester. A study published in Frontiers in Education (2024) confirmed that while generative AI has revolutionised natural language processing and creative domains, significant gaps remain in how institutions, policymakers, and professionals respond to these changes. That gap is where your dissertation lives.

This guide gives you 30+ carefully constructed artificial intelligence dissertation topics across undergraduate, masters, and PhD levels. Every topic includes a research aim, key questions, methodology, difficulty rating, and an honest explanation of why it works academically. No vague titles. No recycled ideas. Every topic here is specific, feasible, and designed to help you reach distinction.

How to Use This Guide

• Undergraduate students should look for topics rated Easy or Moderate with clear secondary data routes.
• Masters students should target Moderate or Advanced topics where a mixed-methods or qualitative primary study is realistic within 12–15,000 words.
• PhD candidates should use these as seed ideas that they then narrow further with a systematic literature review.

Quick note on methodology language throughout this guide:

• Quantitative = numerical data, statistical analysis, surveys with Likert scales
• Qualitative = interviews, thematic analysis, document analysis, case studies
• Mixed = both quantitative and qualitative in the same study

AI in Healthcare & the NHS

These topics sit at one of the most active research frontiers in UK academia. The NHS has committed to expanding AI across diagnostics, mental health referral, and administrative workflows, creating a rich landscape for empirical and critical research alike.

Topic 1: Clinician Trust in AI-Assisted Diagnostic Tools Across NHS Secondary Care Settings

Research Aim:
To examine the factors that predict whether NHS clinicians trust and act upon AI-generated diagnostic recommendations, with a focus on explainability and professional accountability.

Key Research Questions:

1. What organisational and psychological factors predict clinician trust in AI diagnostic outputs within NHS secondary care?
2. Does the use of explainable AI (XAI) techniques increase clinician confidence in AI-generated diagnoses compared to black-box models?
3. How do clinicians reconcile professional accountability with AI-assisted decision-making in high-stakes diagnostic contexts?

Suggested Methodology:
Qualitative — semi-structured interviews with 15–25 NHS clinicians (doctors, radiologists, pathologists) across two or three NHS trusts. Thematic analysis using Braun and Clarke (2006) framework. Secondary analysis of existing NHS AI deployment reports to contextualise findings.

Why This Topic Works:
Research consistently confirms that black-box AI outputs cannot easily be explained to clinicians or patients, raising serious questions about accountability and clinical validity. The gap between AI diagnostic accuracy and real-world clinical adoption is a confirmed, growing research priority. This is squarely within UK ethics guidelines (NHS REC approval pathway is well-established), and interview access is feasible through university NHS partnerships.

Difficulty Level: Moderate
Best Suitable For: Masters / PhD

Topic 2: Evaluating the Accuracy and Ethical Acceptability of AI-Based Mental Health Triage Chatbots in UK IAPT Services

Research Aim:
To critically evaluate whether AI chatbots used in NHS Talking Therapies (formerly IAPT) services deliver accurate triage recommendations while meeting ethical standards around consent, bias, and data privacy.

Key Research Questions:

1. How accurately do current AI triage chatbots in NHS Talking Therapies services classify patients by urgency compared to human assessors?
2. What ethical concerns do service users report regarding AI-driven mental health triage, particularly around data consent and perceived dehumanisation?
3. How do NHS Talking Therapies providers currently govern AI chatbot deployment, and what gaps exist in their data protection compliance under the UK Data Protection Act 2018?

Suggested Methodology:
Mixed methods — secondary data analysis of published clinical outcome data (e.g. the published Nature Medicine study involving 129,400 patients across 28 NHS Talking Therapies services) combined with semi-structured interviews with service users and clinical managers. Document analysis of NHS trust AI governance policies.

Why This Topic Works:
A real-world study published in Nature Medicine involving 129,400 patients across 28 NHS Talking Therapies services in England found that AI-assisted clinical assessment produced significant changes in who sought support, including a 179% increase in non-binary individuals. This is a live, data-rich area with clear ethical and governance questions that no single UK dissertation has yet fully synthesised. The Data Protection Act 2018 angle gives it a UK-specific legal hook.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD

Topic 3: Predicting 30-Day Hospital Readmission Using Machine Learning Models on NHS Electronic Health Record Data: A Comparative Algorithm Study

Research Aim:
To compare the predictive accuracy of logistic regression, random forest, and gradient boosting models in predicting 30-day unplanned hospital readmission using a publicly available or secondary NHS dataset.

Key Research Questions:

1. Which supervised machine learning algorithm achieves the highest accuracy, sensitivity, and specificity for predicting 30-day readmission using structured EHR features?
2. What patient-level variables (age, diagnosis codes, comorbidities, prior admissions) contribute most to prediction accuracy across models?
3. How do model performance metrics compare when trained on balanced versus imbalanced class datasets?

Suggested Methodology:
Quantitative — computational study using a publicly available NHS dataset (NHS Digital Hospital Episode Statistics or MIMIC-IV for secondary use). Python or R for modelling. Evaluation using ROC-AUC, F1-score, and confusion matrices. No primary data collection required.

Why This Topic Works:
This is technically rigorous and fully achievable without primary NHS access. Publicly available NHS HES data and MIMIC datasets make it realistic for a master’s student with basic Python skills. It ticks the methodology box for distinction-level computer science or data science dissertations, and machine learning in healthcare prediction is a confirmed high-priority research area in UK academic circles.

Difficulty Level: Moderate–Advanced
Best Suitable For: Masters (Computer Science / Data Science / Health Informatics)

Topic 4: Examining Racial and Gender Bias in AI-Powered Skin Lesion Classification Models Used in UK Dermatology Services

Research Aim:
To investigate whether commercially deployed AI dermatology tools produce statistically different diagnostic accuracy rates across patient subgroups defined by skin tone and gender, and to evaluate the adequacy of current NHS procurement guidelines in addressing this.

Key Research Questions:

1. Do published validation studies for AI dermatology tools demonstrate equivalent sensitivity and specificity across the Fitzpatrick skin tone scale?
2. What proportion of training datasets used in CE-marked AI dermatology products sold to NHS trusts include adequate representation of darker skin tones?
3. How do current NHS AI procurement frameworks (including the NHS AI Lab’s guidance) address algorithmic bias in dermatology tools?

Suggested Methodology:
Mixed — systematic review of published AI dermatology algorithm validation papers (2018–2024) with a data extraction framework, combined with document analysis of NHS procurement guidelines and CE-marking standards.

Why This Topic Works:
Algorithmic bias in healthcare AI is one of the most urgent topics in medical ethics and health policy. AI systems embedded in high-stakes decision-making in healthcare can yield devastating results when trained on non-representative data. Skin lesion classification is a particularly well-evidenced case of this problem with a directly relevant NHS policy angle. This topic requires no primary data collection and is highly suitable for students in nursing, public health, or biomedical science combining with an AI research methodology.

Difficulty Level: Moderate
Best Suitable For: Undergraduate (Health Sciences) / Masters

AI Ethics, Regulation & Governance

UK AI regulation is at a pivotal point. The government’s 2023 AI Regulation White Paper proposed a “pro-innovation” approach through sector-specific regulators rather than blanket legislation — a position that generates genuine controversy and academic debate.

Topic 5: A Critical Analysis of the UK AI Regulation White Paper’s Proportionality Principles Against EU AI Act Standards

Research Aim:
To critically compare the UK government’s sector-led, principles-based AI governance approach with the EU AI Act’s risk-classification framework, evaluating which better protects citizens while enabling innovation.

Key Research Questions:

1. How do the UK AI White Paper’s five cross-sectoral principles compare structurally to the EU AI Act’s prohibited and high-risk AI categories?
2. What specific governance gaps does the UK’s “no blanket regulation” approach create for high-risk AI applications such as hiring, credit scoring, and predictive policing?
3. How have Russell Group universities in the UK responded to the absence of binding AI regulation, particularly in their generative AI policies?

Suggested Methodology:
Qualitative — comparative legal and policy analysis using document analysis methodology. Primary sources: UK AI White Paper (2023), EU AI Act (2024), UK ICO guidance, Russell Group university AI policies. Secondary analysis of academic commentary from Oxford Internet Institute, Ada Lovelace Institute, and AI Now Institute.

Why This Topic Works:
The UK government’s approach to AI regulation emphasises proportionality and sector-specific oversight, creating genuine uncertainty about accountability for algorithmic decisions affecting UK citizens. Published research on Russell Group AI policies confirms that universities themselves have varied widely in their responses. This is a topic with rich primary documentary sources, strong academic literature support, and direct public policy relevance that supervisors will immediately recognise.

Difficulty Level: Moderate
Best Suitable For: Masters (Law / Politics / Public Policy / Computer Science with Ethics)

Topic 6: Algorithmic Bias in UK Automated Hiring Systems: An Empirical Investigation of Employer Practices and Equality Act 2010 Compliance

Research Aim:
To examine how UK employers using automated hiring systems (AHS) currently manage risks of algorithmic discrimination and whether their practices align with obligations under the Equality Act 2010.

Key Research Questions:

1. What bias identification and mitigation practices do UK employers using AHS currently employ, and how do these vary by sector and company size?
2. To what extent do UK employers’ AHS procurement and governance practices align with Equality Act 2010 requirements regarding protected characteristics?
3. How do HR professionals in UK organisations perceive the fairness and accountability of AHS outputs?

Suggested Methodology:
Qualitative — semi-structured interviews with HR managers and procurement officers at 10–15 UK organisations across financial services, retail, and public sector. Document analysis of publicly available AHS vendor documentation and equality impact assessments.

Why This Topic Works:
Published research in the Journal of Law and Society (2025) confirmed that substantial gaps exist in understanding how AI hiring systems are used in practice, particularly regarding real-world risks of discrimination. This fills a confirmed empirical gap. The Equality Act 2010 provides a clear UK-specific legal framework, and HR interview access is feasible through university placement or alumni networks.

Difficulty Level: Moderate
Best Suitable For: Masters (Business / HRM / Law)

Topic 7: Public Awareness and Understanding of Algorithmic Decision-Making Among UK Adults: A Survey-Based Study

Research Aim:
To measure the extent to which UK adults understand when and how algorithmic systems make decisions that affect their lives, and to identify demographic predictors of algorithmic awareness.

Key Research Questions:

1. What proportion of UK adults can correctly identify algorithmic decision-making in scenarios involving credit scoring, benefits assessment, and social media curation?
2. Do age, educational attainment, and digital literacy predict algorithmic awareness independently after controlling for other sociodemographic variables?
3. Does algorithmic awareness correlate with attitudes toward AI regulation and consumer demand for transparency?

Suggested Methodology:
Quantitative — online survey (target n = 300–500) using validated digital literacy scales (e.g. van Deursen and van Dijk framework) combined with purpose-designed algorithmic awareness vignettes. Analysis using logistic regression and SPSS or R.

Why This Topic Works:
Transparency and explainability are among the five principles in the UK’s own AI governance framework, yet almost no empirical UK-based research measures actual public understanding of these issues at a population level. This is a clear, feasible primary quantitative study. The survey can be distributed via Prolific Academic, making it achievable within a standard dissertation timeline.

Difficulty Level: Easy–Moderate
Best Suitable For: Undergraduate / Masters (Psychology / Sociology / Political Science / Computer Science)

AI in UK Higher Education

This is arguably the hottest area for UK-based dissertation research right now. A 2025 HEPI/Kortext survey found that 92% of full-time undergraduate students in the UK now use AI in some form, up from 66% in 2024, and 88% have used generative AI for assessments — yet only 36% have received support from their institution to develop AI skills.

Topic 8: Student Motivations for Using Generative AI in University Assessments: A Qualitative Exploration Across UK Russell Group and Post-92 Institutions

Research Aim:
To explore the motivations, perceived benefits, and ethical reasoning of UK undergraduate students who use generative AI tools for academic assessments, comparing experiences across different institutional types.

Key Research Questions:

1. What primary motivations drive undergraduate students at UK Russell Group and post-92 universities to use generative AI in assessed work?
2. How do students ethically rationalise their use of AI tools in relation to their institution’s academic integrity policies?
3. Do perceptions of institutional AI guidance quality mediate students’ willingness to use AI transparently?

Suggested Methodology:
Qualitative — focus groups (3–4 groups of 5–8 participants each) at two or three UK universities selected to represent different institutional types. Thematic analysis. Participants recruited through module mailing lists or student union channels.

Why This Topic Works:
HEPI’s 2025 student survey confirms that 88% of students have used generative AI in assessments, but there is almost no qualitative UK evidence exploring why students make this choice, how they reason about it ethically, and whether institutional policies influence their behaviour. This is a confirmed literature gap, ethically manageable (no sensitive populations), and practically very feasible for students at UK universities.

Difficulty Level: Easy
Best Suitable For: Undergraduate / Masters (Education / Psychology / Sociology)

Topic 9: Divergence in UK University Generative AI Policies: A Comparative Analysis of Russell Group Institutional Responses

Research Aim:
To systematically compare the generative AI policies of Russell Group universities, identifying areas of consensus, divergence, and key governance gaps.

Key Research Questions:

1. To what extent do Russell Group universities’ generative AI policies align with the principles outlined in the Russell Group’s own 2023 AI position statement?
2. What specific variations exist in how UK universities define and penalise unauthorised AI use in assessed work?
3. How have UK university AI policies evolved between 2023 and 2025, and what triggers institutional policy updates?

Suggested Methodology:
Qualitative — systematic document analysis of publicly available AI policies from all 24 Russell Group universities, supplemented by analysis of academic senate minutes and official university blog announcements where available. Content analysis using NVivo or MAXQDA.

Why This Topic Works:
Published research confirms that the literature on generative AI policies in UK universities is still very limited, with only 16 relevant articles identified in one 2024 systematic search, all published in 2023 or 2024. This is a research space that is actively underpopulated. Document analysis is one of the most achievable methodologies for a dissertation student and requires no primary recruitment.

Difficulty Level: Easy–Moderate
Best Suitable For: Undergraduate / Masters (Education / Higher Education Management / Policy)

Topic 10: The Generative AI Literacy Divide: Do UK Students from Lower Socioeconomic Backgrounds Use AI Tools Less Effectively Than Their Peers?

Research Aim:
To investigate whether socioeconomic background predicts differential access to and effective use of generative AI tools among UK undergraduate students, with implications for the digital equity agenda.

Key Research Questions:

1. Is there a statistically significant association between students’ self-reported socioeconomic background (using POLAR4 quintile or parental education as proxies) and frequency or quality of generative AI tool use?
2. What types of generative AI barriers (cost, confidence, awareness, institutional support) are more commonly reported by students from lower quintile backgrounds?
3. Does the perceived usefulness of AI tools differ significantly by socioeconomic group after controlling for subject studied and prior digital experience?

Suggested Methodology:
Quantitative — structured survey (n = 200–400) distributed at one or two UK universities. Regression analysis to test socioeconomic background as a predictor of AI use patterns. Survey instrument to include validated digital self-efficacy scales.

Why This Topic Works:
Research published in Frontiers in Education confirmed that access to powerful LLMs is unevenly distributed due to subscription costs, and that the integration of AI for adaptive learning could widen rather than close existing disparities. The HEPI 2025 survey confirms the problem is live and growing but does not disaggregate by socioeconomic background — a clear gap this study fills.

Difficulty Level: Moderate
Best Suitable For: Masters (Education / Sociology)

AI & Natural Language Processing (NLP)

Topic 11: Detecting AI-Generated Academic Text in UK University Submissions: An Evaluation of Current Detection Tools and Their Accuracy Across Disciplines

Research Aim:
To empirically evaluate the accuracy, false positive rate, and disciplinary consistency of commercially available AI text detection tools when applied to authentic student submissions and AI-generated text.

Key Research Questions:

1. How accurately do Turnitin AI Detection, GPTZero, and Copyleaks classify authentic student writing versus AI-generated text across STEM and humanities disciplines?
2. What is the false positive rate for AI detection tools when applied to high-achieving student writing, and what characteristics of authentic writing trigger false positives?
3. Are current AI detection tools more likely to misclassify writing from non-native English speakers as AI-generated?

Suggested Methodology:
Quantitative — experimental design. A controlled corpus of authentic student writing samples (collected with ethical consent) and matched AI-generated samples (generated using ChatGPT-4o and Claude 3) is processed through three detection tools. Accuracy, precision, recall, and F1-score calculated and compared across tools and text types.

Why This Topic Works:
Research published in the Journal of University Teaching and Learning Practice (2023–2024) has highlighted how simple bypass techniques can circumvent AI text detectors, particularly for non-native English speakers — a serious academic integrity concern. This is technically achievable, ethically straightforward with proper consent, and fills a directly measurable empirical gap.

Difficulty Level: Moderate
Best Suitable For: Undergraduate (Computer Science) / Masters

Topic 12: Fine-Tuning a BERT-Based Sentiment Analysis Model on UK Political Debate Transcripts: Accuracy, Bias, and Deployment Considerations

Research Aim:
To develop and evaluate a fine-tuned BERT model for sentiment classification of UK Parliamentary Hansard debates, assessing accuracy against human annotation and identifying potential political bias in model outputs.

Key Research Questions:

1. What classification accuracy does a fine-tuned BERT model achieve on UK Parliamentary debate sentiment compared to a baseline naïve Bayes model?
2. Does fine-tuning on domain-specific political text reduce or amplify sentiment classification bias compared to a general-purpose pre-trained model?
3. What deployment considerations (computational cost, transparency, reproducibility) affect the practical usability of BERT-based NLP in political science research?

Suggested Methodology:
Quantitative/Technical — Python (HuggingFace Transformers library), Hansard parliamentary transcripts (publicly available under Open Parliament Licence). Human annotation of a gold-standard validation set by two independent coders (inter-rater reliability measured using Cohen’s kappa).

Why This Topic Works:
Hansard data is freely available and academically legitimate. BERT fine-tuning is well within the capability of a masters-level computer science student with Python skills. The political bias question adds an ethics dimension that makes this suitable for distinction-level work beyond pure technical execution.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Computer Science / Data Science / Political Science with NLP skills)

Topic 13: Evaluating the Reliability of Large Language Models as Research Synthesis Tools: A Comparative Study of Hallucination Rates Across Academic Disciplines

Research Aim:
To measure the rate and nature of factual hallucinations produced by GPT-4o and Claude 3 when used to synthesise literature across law, medicine, and history, and to evaluate whether output reliability varies by domain.

Key Research Questions:

1. At what rate do GPT-4o and Claude 3 produce factually incorrect claims (hallucinations) when asked to summarise peer-reviewed literature across three disciplines?
2. Does hallucination rate vary significantly by academic discipline, and if so, what domain characteristics predict higher hallucination frequency?
3. What verification strategies best reduce researcher reliance on hallucinated LLM outputs in academic research workflows?

Suggested Methodology:
Mixed — quantitative hallucination measurement (standardised prompts, outputs verified against source texts by two independent expert reviewers; inter-rater reliability measured) combined with qualitative document analysis of hallucination patterns and types.

Why This Topic Works:
This is a fast-growing and genuinely understudied area. As LLMs are increasingly used in academic research workflows, the question of their reliability is urgent for UK research integrity frameworks. No existing systematic study directly compares hallucination rates across disciplines using the same methodology. Data collection requires no participants — just prompt engineering, output recording, and verification against source materials.

Difficulty Level: Moderate
Best Suitable For: Masters (any discipline with strong research methods foundation)

AI & Machine Learning in Business and Finance

Topic 14: Predicting SME Loan Default Using Gradient Boosting Models: An Analysis of UK Open Banking Data

Research Aim:
To develop and validate a gradient boosting model for predicting SME loan default using transactional features derived from open banking data, and to compare its performance against traditional logistic regression credit scoring models.

Key Research Questions:

1. Does a gradient boosting classifier (XGBoost) outperform logistic regression in predicting 12-month SME loan default using open banking transaction features?
2. Which transaction-derived features contribute most to model performance, and how do these compare to traditional creditworthiness indicators?
3. What fairness and explainability considerations arise from deploying machine learning in SME credit scoring, particularly for underrepresented business owner demographics?

Suggested Methodology:
Quantitative — technical machine learning study using publicly available or synthetic open banking datasets (e.g. UK Finance industry datasets or academic equivalents). Python (XGBoost, SHAP for explainability). SHAP values used to interpret feature importance.

Why This Topic Works:
Open banking in the UK is a live policy and industry trend. UK Finance data confirms that SME credit access is a national policy priority. This topic is technically rigorous, has clear secondary data availability, and the fairness angle means it goes beyond pure technical execution to address regulatory relevance — exactly what distinction-level marking criteria reward.

Difficulty Level: Advanced
Best Suitable For: Masters (Finance / Data Science / Business Analytics)

Topic 15: How Do UK FinTech Firms Implement AI Risk Management Frameworks Under FCA Guidance? A Multi-Case Study

Research Aim:
To examine how UK-regulated FinTech firms operationalise AI risk management practices in response to Financial Conduct Authority guidance, identifying gaps between regulatory expectations and industry practice.

Key Research Questions:

1. How do mid-sized UK FinTech firms currently translate FCA AI and algorithmic accountability guidance into internal governance processes?
2. What barriers prevent smaller FCA-authorised firms from implementing comprehensive AI risk frameworks?
3. How do FinTech risk managers assess and document model risk for AI systems used in consumer-facing financial decisions?

Suggested Methodology:
Qualitative — multiple case study design (3–5 FinTech firms). Semi-structured interviews with compliance officers, Chief Risk Officers, and data scientists. Document analysis of publicly available internal model risk frameworks and FCA supervisory statements.

Why This Topic Works:
The FCA has published clear expectations around algorithmic accountability, but empirical research on how smaller UK FinTechs actually implement these is almost absent from the literature. Interview access is feasible through LinkedIn and FinTech UK networks. This is a genuinely industry-relevant topic likely to attract strong supervisor support.

Difficulty Level: Moderate
Best Suitable For: Masters (Finance / Risk Management / FinTech)

Topic 16: AI-Driven Dynamic Pricing in UK Retail: Consumer Perceptions of Fairness and Algorithmic Transparency

Research Aim:
To investigate UK consumer perceptions of algorithmic dynamic pricing — as used by major online retailers — focusing on perceived fairness, trust, and demand for transparency.

Key Research Questions:

1. Do UK consumers perceive AI-driven dynamic pricing as fair, and does perceived fairness differ by pricing context (travel vs. groceries vs. entertainment)?
2. Does disclosure of algorithmic pricing mechanisms increase or decrease consumer trust in the pricing process?
3. Are there demographic differences in consumer tolerance of dynamic pricing, and do prior experiences of price discrimination predict attitudes?

Suggested Methodology:
Mixed — online survey (n = 250–400) using validated consumer trust and fairness scales, combined with three experimental vignettes presenting pricing scenarios with and without algorithm disclosure. Analysis using ANOVA and regression.

Why This Topic Works:
Dynamic pricing has become a mainstream practice and a live public controversy in the UK, particularly following debates around concert ticket pricing (e.g. Oasis 2024 tour dynamic pricing controversy). This gives the topic clear contemporary relevance and easy participant engagement. Survey design is straightforward and data analysis is achievable at masters level.

Difficulty Level: Easy–Moderate
Best Suitable For: Undergraduate / Masters (Marketing / Business / Consumer Psychology)

AI, Society & Ethics

Topic 17: Deepfake Detection and Democratic Harm: An Evaluation of AI Detection Tool Accuracy for UK Electoral Disinformation Scenarios

Research Aim:
To evaluate the accuracy and limitations of current AI deepfake detection tools when applied to politically themed video disinformation content, in the context of UK general elections.

Key Research Questions:

1. How accurately do three leading deepfake detection tools (e.g. Microsoft Video Authenticator, Deepware Scanner) identify synthetically manipulated political video content?
2. Do detection tools perform differently on deepfakes of UK political figures compared to US political figures, and if so, why?
3. What policy interventions do stakeholders (journalists, electoral officials, digital platform researchers) consider most feasible for reducing deepfake-enabled electoral harm in the UK?

Suggested Methodology:
Mixed — quantitative testing of detection tools against a purpose-built or publicly available deepfake video corpus (e.g. FaceForensics++ or DGM4 dataset), combined with qualitative expert interviews (5–8 journalists, electoral officials, or digital platform policy researchers).

Why This Topic Works:
Deepfake videos have increased by 550% online since 2019. Research published in Frontiers in Artificial Intelligence (2025) confirms this represents a systemic and accelerating threat to democratic information ecosystems. The UK context (Electoral Commission, Online Safety Act 2023) provides a specific regulatory anchor. The combination of technical testing and policy interviews gives this genuine mixed-methods credibility.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Computer Science / Politics / Media Studies)

Topic 18: Examining the Relationship Between Social Media Algorithmic Curation and Political Polarisation Among UK Young Adults (18–25)

Research Aim:
To investigate whether self-reported exposure to algorithmically curated political content on TikTok and Instagram is associated with higher levels of political polarisation among UK young adults.

Key Research Questions:

1. Is there a statistically significant correlation between self-reported algorithmic news exposure on TikTok and Instagram and affective political polarisation among UK 18–25 year olds?
2. Does the directionality of political content exposure (predominantly left-leaning vs. right-leaning feed) predict attitudes toward political outgroups?
3. Do news media literacy skills moderate the relationship between algorithmic exposure and polarisation?

Suggested Methodology:
Quantitative — cross-sectional online survey (n = 300+) using validated affective polarisation scales (Druckman and Levendusky adaptations for UK context), a social media use inventory, and the Digital News Media Literacy scale. Hierarchical regression analysis.

Why This Topic Works:
AI-driven social media recommendation algorithms are a confirmed major research priority in communications, political science, and digital sociology. The UK 18–25 demographic is well-studied enough to benchmark against, while being genuinely underrepresented in algorithmic polarisation research specifically. This is a feasible quantitative dissertation with well-validated instruments already available.

Difficulty Level: Moderate
Best Suitable For: Masters (Political Science / Communications / Sociology / Psychology)

Topic 19: Do UK Consumers Accept AI Systems in Intimate Service Roles? A Study of AI Companion Apps Among Elderly Users

Research Aim:
To examine the attitudes and experiences of UK adults aged 65+ toward AI companion applications, exploring acceptance, emotional attachment, and ethical concerns.

Key Research Questions:

1. What factors predict AI companion app acceptance among UK adults aged 65 and over, as measured by the Technology Acceptance Model (TAM)?
2. Do elderly users report emotional attachment to AI companions, and how do they make sense of this attachment ethically and socially?
3. What safeguarding concerns do family caregivers and care home managers raise regarding AI companion use among older adults with cognitive decline?

Suggested Methodology:
Qualitative — semi-structured interviews with 12–18 participants across three groups: elderly app users (6–8 participants), family caregivers (3–5), and care home managers (3–5). Interpretive Phenomenological Analysis (IPA).

Why This Topic Works:
The UK has a well-documented loneliness and social isolation crisis among its elderly population (Office for National Statistics, 2023). AI companions are a fast-growing commercial category (Replika, Character.ai etc.) but systematic empirical research in the UK care context is almost absent. IPA is well suited to the emotionally complex territory, and it is a methodology examiners appreciate for its rigour and depth.

Difficulty Level: Moderate
Best Suitable For: Masters (Psychology / Social Work / Gerontology)

AI & Climate / Sustainability

Topic 20: Mapping the Carbon Footprint of Training Large Language Models: A Systematic Review of Transparency in AI Environmental Reporting (2019–2024)

Research Aim:
To systematically review published transparency in carbon emissions reporting for large language model training runs, identifying patterns of disclosure, omission, and methodological inconsistency.

Key Research Questions:

1. What proportion of published LLM research papers between 2019 and 2024 include quantitative carbon emissions data for model training?
2. How do reported carbon footprint figures vary by compute provider, model size, and geographical training location, and are these variations adequately explained in the source literature?
3. What methodological standards (if any) do leading AI developers (OpenAI, Google DeepMind, Meta AI) use for environmental reporting, and how do these compare to GHG Protocol standards?

Suggested Methodology:
Qualitative / Systematic Review — PRISMA-guided systematic review of LLM research papers published on arXiv, ACL Anthology, and NeurIPS/ICML proceedings (2019–2024). Data extraction framework based on Strubell et al. (2019) energy reporting criteria.

Why This Topic Works:
AI energy consumption and carbon transparency is a rapidly emerging area of academic and policy interest, and systematic reviews are highly valued at distinction level because they require structured, defensible methodology. No existing published systematic review specifically maps transparency standards across LLM training reporting. This is a clear, achievable gap with defined search parameters.

Difficulty Level: Moderate
Best Suitable For: Masters (Computer Science / Environmental Studies / AI Ethics)

Topic 21: AI-Optimised Demand Response in UK Smart Grid Management: Evaluating Reinforcement Learning Approaches Against Baseline Scheduling Methods

Research Aim:
To compare the energy efficiency gains of reinforcement learning-based demand response optimisation against rule-based baseline methods using simulated UK residential grid load data.

Key Research Questions:

1. Does a deep Q-learning demand response algorithm achieve statistically greater peak load reduction than a rule-based scheduling baseline on simulated UK residential grid data?
2. How does algorithm performance vary across seasonal demand profiles and increasing proportions of EV charging loads?
3. What practical deployment barriers (latency, explainability, hardware constraints) would affect real-world adoption of RL-based demand response in UK smart homes?

Suggested Methodology:
Quantitative/Technical — simulation study using publicly available UK National Grid smart meter and demand data. Python (OpenAI Gym or custom RL environment). Results evaluated against National Grid Electricity System Operator baseline benchmarks.

Why This Topic Works:
The UK’s net zero commitments and National Grid investment in smart grid infrastructure make this a genuinely policy-relevant topic. Simulation-based RL studies do not require real hardware access, making them feasible for students with intermediate Python skills. This is a strong engineering or computer science dissertation topic with clear external validity.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Computer Science / Electrical Engineering / Energy Systems)

Computer Vision & Deep Learning

Topic 22: Evaluating Transfer Learning Approaches for Low-Resource Medical Image Classification in UK NHS Radiology Contexts

Research Aim:
To compare the diagnostic accuracy of three transfer learning architectures (ResNet-50, EfficientNet-B4, Vision Transformer) for classifying chest X-ray pathologies using a limited training dataset, simulating low-resource NHS deployment scenarios.

Key Research Questions:

1. Which transfer learning architecture achieves the highest accuracy on chest X-ray classification tasks when fine-tuned with a limited dataset (n < 500 labelled images)?
2. How does classification accuracy degrade as training set size decreases from 1,000 to 100 labelled samples, and does this degradation differ significantly across architectures?
3. What data augmentation strategies most effectively mitigate overfitting in low-resource medical image classification scenarios?

Suggested Methodology:
Quantitative/Technical — experiments using the NIH Chest X-ray Dataset (publicly available, 112,120 images). Python (PyTorch or TensorFlow). AUC-ROC curves, sensitivity, and specificity measured for each architecture across multiple training set size conditions.

Why This Topic Works:
NHS radiology departments face real constraints in labelled training data. Research confirming that AI diagnostic tools can achieve strong performance in low-resource settings would have direct practical significance. The NIH Chest X-ray dataset is freely available, well-documented, and widely used for reproducibility benchmarking.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Computer Science / Medical Informatics)

Topic 23: Bias in Facial Recognition Technology Used by UK Police: A Critical Analysis of Accuracy Disparities Across Demographic Groups

Research Aim:
To critically evaluate published accuracy data for facial recognition technology (FRT) deployed by UK police forces, focusing on performance disparities across racial, gender, and age groups, and the adequacy of current regulatory responses.

Key Research Questions:

1. What accuracy disparities across demographic groups are reported in the published evaluations of FRT systems deployed by UK police forces (e.g. South Wales Police, Metropolitan Police)?
2. How do the UK Surveillance Camera Commissioner’s guidance and the Biometrics and Forensics Ethics Group’s recommendations address demographic accuracy disparity in police FRT?
3. What legal challenges have been brought against UK police FRT use, and what do these cases reveal about regulatory gaps?

Suggested Methodology:
Qualitative — critical document analysis of published FRT evaluation reports, judicial review case documents, parliamentary select committee evidence, and regulatory guidance (2018–2024). Framework analysis using a human rights and algorithmic accountability lens.

Why This Topic Works:
UK police facial recognition deployment is a live and contested policy area. The Metropolitan Police and South Wales Police’s FRT programmes have been the subject of legal challenge and parliamentary scrutiny. This topic requires no primary data collection, involves highly accessible documentary sources, and directly addresses AI ethics, race, and law — a combination that supervisors in law, computer science, criminology, and sociology will all recognise as urgent.

Difficulty Level: Moderate
Best Suitable For: Undergraduate / Masters (Law / Criminology / Computer Science)

AI in Education Technology (EdTech)

Topic 24: Student Perceptions of AI-Powered Adaptive Learning Platforms in UK Further Education Colleges: A Mixed-Methods Study

Research Aim:
To examine how students in UK Further Education (FE) colleges experience AI-powered adaptive learning platforms, exploring perceived academic benefit, engagement, and concerns about data privacy.

Key Research Questions:

1. Do students who use AI adaptive learning platforms in UK FE colleges report higher perceived learning gains than peers in conventional taught settings?
2. What concerns do FE students express about the data collected by adaptive learning platforms, and how do these concerns affect engagement?
3. How do FE lecturers perceive the role of adaptive learning AI in their professional practice — as augmentation or replacement?

Suggested Methodology:
Mixed — student survey (n = 150–250) at one or two FE colleges using validated self-regulated learning and technology acceptance scales, combined with semi-structured interviews with 8–10 lecturers. Integration at interpretation stage.

Why This Topic Works:
Most UK research on AI in education focuses on higher education universities. FE colleges are significantly under-researched, despite serving a large and socioeconomically diverse student population. This creates a genuine gap. FE college access is achievable through local contacts or cold outreach, and the topic directly addresses JISC’s published priorities for AI in UK education.

Difficulty Level: Moderate
Best Suitable For: Masters (Education / Educational Technology)

Topic 25: Designing an AI-Assisted Feedback Tool for UK Secondary School Essay Writing: A Design Science Research Study

Research Aim:
To design, prototype, and evaluate an AI-assisted formative feedback tool for secondary school essay writing in the UK, grounded in pedagogical principles and tested through teacher and student feedback.

Key Research Questions:

1. What design features do UK secondary school English teachers identify as essential for an AI feedback tool to be pedagogically useful and trustworthy?
2. Does prototype AI feedback on student essays align with teacher assessment criteria and GCSE mark scheme priorities?
3. What ethical and safeguarding concerns do teachers raise about AI feedback tools processing student writing?

Suggested Methodology:
Mixed — Design Science Research Methodology (DSRM). Phase 1: teacher focus groups (needs analysis). Phase 2: prototype development using OpenAI API. Phase 3: iterative evaluation sessions with teachers and students. Qualitative analysis of evaluation feedback.

Why This Topic Works:
Ofsted and UK government policy both emphasise the quality of formative feedback in secondary schools. AI-assisted writing feedback is commercially active (e.g. tools like Grammarly, Turnitin Feedback Studio) but systematic design research grounded in UK curriculum standards is rare. Design Science Research is a well-recognised and examiner-approved methodology for computing education dissertations.

Difficulty Level: Advanced
Best Suitable For: Masters (Computer Science / Educational Technology)

Emerging Topics — PhD and Advanced Masters

These topics are specifically designed for PhD candidates or Masters students with strong methodological backgrounds looking to contribute directly to emerging research conversations.

Topic 26: Explainability Requirements in the EU AI Act and Their Implications for UK AI Developers Post-Brexit: An Empirical Study of Developer Perceptions

Research Aim:
To investigate how UK-based AI developers perceive and are preparing for EU AI Act explainability requirements, and to assess whether the UK regulatory divergence creates a competitive disadvantage or advantage for UK AI companies.

Key Research Questions:

1. How do UK AI developers currently interpret and operationalise “explainability” requirements, and how do these practices compare to EU AI Act Annex IV technical documentation requirements?
2. Do UK developers perceive Brexit-driven regulatory divergence on AI explainability as a burden, opportunity, or neutral factor for UK market competitiveness?
3. What strategies are UK AI firms adopting to simultaneously comply with UK ICO guidance and EU AI Act requirements for products sold into EU markets?

Suggested Methodology:
Qualitative — expert interviews with 15–20 senior AI developers, legal counsel, and compliance leads at UK AI companies. Purposive sampling via UK AI industry directories (TechNation, UKRI, UK FinTech Alliance). Framework analysis.

Why This Topic Works:
Brexit’s implications for UK AI regulation are a genuinely underexplored empirical area. The EU AI Act came into force in August 2024, and UK developers exporting to EU markets face dual compliance pressure. There is almost no empirical evidence of how UK developers are responding. This fills a policy-urgent, academically novel gap suitable for PhD or high-distinction Masters work.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Law / Computer Science / Business / Policy)

Topic 27: Longitudinal Analysis of Predictive Policing Algorithm Outcomes in UK Police Force Deployments: Accuracy, Drift, and Community Trust (2019–2024)

Research Aim:
To analyse published and publicly available outcome data from UK police predictive analytics deployments, examining accuracy over time (model drift), and comparing algorithm-flagged areas with community-reported trust measures.

Key Research Questions:

1. Do published accuracy claims for predictive policing tools deployed by UK police forces remain consistent over time, or is model drift documented in force review data?
2. Is there a statistically significant association between the intensity of predictive policing deployment and community trust measures (as reported in HMICFRS inspection reports)?
3. How transparent are UK police forces in publishing algorithm accuracy reviews, and does this transparency vary by force size and region?

Suggested Methodology:
Mixed — longitudinal document analysis of HMICFRS inspection reports (2019–2024), Home Office data releases, and individual force transparency reports. Where available, quantitative correlation analysis between deployment density and community trust scores from force-level Crime Survey data.

Why This Topic Works:
Predictive policing is one of the most ethically contested AI applications in UK law enforcement. Model drift — where algorithm accuracy degrades as social conditions change — is a confirmed technical problem with real human rights implications. HMICFRS inspection reports and Home Office transparency publications provide a rich secondary data source that is publicly available and academically legitimate.

Difficulty Level: Advanced
Best Suitable For: PhD (Criminology / Computer Science / Sociology)

Topic 28: How Do UK Journalists Use AI Tools for Investigative Reporting? An Ethnographic Study of Newsroom Practices

Research Aim:
To explore how journalists at UK national and regional news organisations are integrating AI tools into investigative reporting workflows, examining professional identity, editorial standards, and accuracy concerns.

Key Research Questions:

1. What AI tools are UK investigative journalists currently using in reporting workflows, and for what specific tasks?
2. How do journalists negotiate tensions between AI-assisted efficiency and core journalistic values of accuracy, source protection, and editorial independence?
3. Does the use of AI in reporting vary significantly between national broadsheet newsrooms and regional/local news outlets, and if so, why?

Suggested Methodology:
Qualitative — ethnographic observation (2–4 weeks) at one or two UK newsrooms, supplemented by semi-structured interviews with journalists, editors, and fact-checkers (15–20 participants total). Reflexive thematic analysis.

Why This Topic Works:
AI in journalism is an active research priority for organisations including the Reuters Institute for the Study of Journalism (Oxford) and the BBC Research & Development unit. UK newsroom access is feasible through journalism school connections. Ethnographic research is valued at distinction level for the depth and contextual richness it offers.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Media Studies / Journalism / Communication)

Topic 29: Federated Learning for NHS Patient Data: Privacy Guarantees, Model Performance Trade-Offs, and Governance Readiness

Research Aim:
To evaluate federated learning (FL) as a privacy-preserving machine learning approach for NHS multi-site clinical data, assessing the performance trade-offs of federation against centralised training and the governance maturity of UK health data institutions.

Key Research Questions:

1. What is the performance loss (measured by AUC-ROC) of federated versus centralised training for a binary clinical outcome prediction task across simulated NHS trust partitions?
2. How does differential privacy noise injection in federated learning affect model utility across varying levels of data heterogeneity?
3. How do UK NHS data governance frameworks (NHS DSPT, Data Access Agreement requirements) accommodate or obstruct federated learning deployments?

Suggested Methodology:
Mixed — technical simulation study using publicly available clinical datasets (e.g. MIMIC-IV partitioned to simulate multi-trust federation) using PySyft or Flower (FL frameworks), combined with document analysis of NHS data governance frameworks and expert interviews with NHS informatics leads.

Why This Topic Works:
Federated learning is one of the most actively developing solutions to the NHS’s long-standing challenge of enabling AI research without compromising patient privacy. This is a technically complex but well-defined research question with publicly available tools and datasets. The governance analysis component makes it genuinely interdisciplinary and policy-relevant.

Difficulty Level: Advanced
Best Suitable For: PhD / Masters (Computer Science / Health Informatics)

Topic 30: Generative AI and the Future of UK Graduate Employment: Do Employers Perceive AI Literacy as a Core Graduate Competency?

Research Aim:
To investigate UK employer perceptions of AI literacy as a graduate competency, examining how expectations differ by sector and whether AI literacy is explicitly integrated into UK graduate recruitment criteria.

Key Research Questions:

1. How do UK employers across finance, healthcare, legal, and technology sectors currently define “AI literacy” as a graduate competency, and how consistently is this definition applied?
2. To what extent is AI literacy explicitly included in graduate job specifications and assessment criteria at UK organisations?
3. Do employers perceive a significant AI literacy gap between graduate expectations and what UK universities are currently preparing graduates to demonstrate?

Suggested Methodology:
Mixed — quantitative content analysis of 200 UK graduate job advertisements (coding for explicit AI literacy requirements) combined with semi-structured interviews with 12–15 graduate recruitment managers across four sectors.

Why This Topic Works:
HEPI’s 2025 student survey confirmed that students overwhelmingly believe AI skills are essential for employment, yet only 36% have received institutional support. Employer-side evidence to match this student perception gap is almost entirely absent from published UK research. This makes it a timely, practical, and clearly scoped dissertation with a straightforward mixed-methods design.

Difficulty Level: Moderate
Best Suitable For: Undergraduate / Masters (Business / Education / HRM / Careers Studies)

Topic 31: Comparing Accuracy and Fairness of Credit Scoring Models With and Without AI Enhancement Among Thin-File Applicants in the UK

Research Aim:
To compare the accuracy and demographic fairness of traditional credit scorecard models against machine learning-enhanced alternatives for predicting default risk among UK thin-file applicants (those with limited credit history).

Key Research Questions:

1. Do gradient boosting models using alternative data sources (open banking, rental payment history) outperform traditional scorecards in predicting 12-month default for UK thin-file applicants?
2. Do machine learning models display greater or lesser demographic disparity in false positive rates across ethnicity groups compared to traditional scorecards?
3. What explainability methods best support UK Consumer Duty compliance for AI credit decisions affecting thin-file applicants?

Suggested Methodology:
Quantitative — technical study using synthetic or partner-provided credit data (or publicly available Home Credit Dataset). Python (XGBoost, SHAP). Fairness measured using Equality of Opportunity and Demographic Parity metrics.

Why This Topic Works:
FCA Consumer Duty (effective 2023) requires firms to demonstrate fair outcomes for customers, including those with limited credit histories. Thin-file applicants are a confirmed vulnerable and underserved group in UK credit markets. This topic has direct regulatory significance, uses well-established technical methodology, and has a fairness angle that pushes it beyond routine ML benchmarking.

Difficulty Level: Advanced
Best Suitable For: Masters (Finance / Data Science / FinTech)

Topic 32: AI-Assisted Clinical Decision Support in UK General Practice: GP Perceptions of Usefulness, Trust, and Professional Autonomy

Research Aim:
To explore how UK GPs perceive AI-driven clinical decision support systems (CDSS) in their day-to-day practice, focusing on trust, perceived usefulness, and concerns about professional autonomy and medicolegal accountability.

Key Research Questions:

1. What factors predict GP trust in AI-based CDSS recommendations, and do these differ from trust predictors for non-AI clinical decision tools?
2. Do GPs perceive AI CDSS as augmenting or threatening their professional autonomy and clinical judgement?
3. What medicolegal concerns do GPs raise about liability for patient harm arising from AI CDSS recommendations they acted upon?

Suggested Methodology:
Qualitative — semi-structured interviews with 15–20 UK GPs recruited through GP federation partnerships or RCGP research network. Thematic analysis informed by Professional Autonomy Theory and Technology Acceptance literature.

Why This Topic Works:
NHS England’s 10-Year Health Plan explicitly commits to expanding AI clinical decision support. Yet GP-specific perceptions of these tools — particularly medicolegal accountability concerns — are significantly understudied. This is a study a primary care research group would immediately recognise as filling a practical policy gap. Ethics approval through university NHS REC pathways is well-trodden for GP interview studies.

Difficulty Level: Moderate
Best Suitable For: Masters / PhD (Medicine / Public Health / Health Services Research)

Three Bonus Topics — Niche but Highly Distinctive

These three topics are deliberately less common. If you want your dissertation to stand out for genuine originality, these are the ones supervisors are unlikely to have seen before.

Topic 33: AI Content Moderation and the Online Safety Act 2023: Can Automated Systems Meet the Act’s Proportionality Requirements?

Research Aim:
To critically evaluate whether current AI content moderation systems — as deployed by major platforms — are technically and ethically capable of meeting the proportionality and freedom of expression balancing requirements of the UK Online Safety Act 2023.

Key Research Questions:

1. How do major platform AI content moderation systems (Meta, YouTube, TikTok) handle the proportionality tension between harmful content removal and freedom of expression under UK law?
2. Does the Online Safety Act’s Ofcom-regulated “safety by design” framework provide sufficient technical specificity to guide AI moderation system design?
3. What independent audit mechanisms exist or are proposed for AI content moderation under the Online Safety Act, and are these adequate?

Suggested Methodology:
Qualitative — critical policy and legal analysis using document analysis of Online Safety Act legislation, Ofcom codes of practice (2024–2025), platform transparency reports, and parliamentary select committee evidence. Expert interviews with 5–8 digital rights lawyers, platform policy researchers, or civil society advocates optional.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Law / Communications / Policy)

Topic 34: Using Transformer Models to Classify UK Employment Tribunal Decisions: A Legal NLP Feasibility Study

Research Aim:
To evaluate the feasibility and accuracy of using a fine-tuned transformer model to classify UK Employment Tribunal decisions by case outcome, protected characteristic, and remedy awarded.

Key Research Questions:

1. What classification accuracy does a fine-tuned Legal-BERT model achieve on Employment Tribunal decisions compared to a TF-IDF baseline?
2. How well does the model handle class imbalance in legal outcomes (e.g. rare remedies)?
3. What are the practical and ethical implications of using NLP classification tools to support legal research and access to justice in UK employment law?

Suggested Methodology:
Quantitative/Technical — fine-tuning Legal-BERT (or Legal-RoBERTa) on publicly available UK Employment Tribunal decisions (BAILII database, open access). Python (HuggingFace). Evaluation using macro-F1 and per-class precision/recall.

Difficulty Level: Advanced
Best Suitable For: Masters / PhD (Computer Science / Law with Data Science skills)

Topic 35: AI-Generated Music and UK Copyright Law: Do Current Intellectual Property Frameworks Adequately Protect Human Composers?

Research Aim:
To critically analyse whether the UK’s current intellectual property framework — particularly the Copyright, Designs and Patents Act 1988 — adequately protects human music composers from the commercial use of AI-generated music trained on their work.

Key Research Questions:

1. Under current UK copyright law, can AI-generated music that reproduces stylistic elements of a named composer’s work constitute copyright infringement?
2. How have recent UK IPO consultation responses and case law developments addressed the copyright ownership gap for AI-generated creative works?
3. How do UK music industry stakeholders (composers, publishers, streaming platform legal teams) perceive the adequacy of current IP protections against generative AI music tools?

Suggested Methodology:
Qualitative — critical legal analysis of UK case law, CDPA 1988, UK IPO AI and IP consultation responses (2022–2024), and European comparative developments (EU AI Act Art. 53 transparency provisions). Supplemented by 8–12 expert interviews with music industry IP practitioners, composers, and music tech researchers.

Difficulty Level: Moderate–Advanced
Best Suitable For: Masters / PhD (Law / Music Business / Creative Industries)