Evaluating the effects of large language models on attitudes towards migrants in the UK

About the Project

Project Overview:

False narratives that fuel social and public opinion generated by large language models (Papageorgiou et al., 2024) move through social spaces faster than they can be mitigated (Kandel, 2020). As a result of social media and personalised news algorithms, people are strikingly isolated from dissenting ideas, as these narratives are spread and repeated by trusted family and friends through social platforms (Ahmed et al., 2024; Bastick, 2021; Wardle & Derakhshan, 2017). Individuals are easily persuaded by LLMs (Bastick, 2021), even when their narratives are false, as they can include seemingly credible citations (Hackenburg et al., 2025). Further, “liking” and “sharing” of these narratives implies endorsement of their content (Muhammed T & Mathew, 2022; Wardle & Derakhshan, 2017), catalysing social stigma, and potentially resulting in extreme responses, such as violence (Ștefăniță & Buf, 2021), and political extremism (Marwick & Lewis, 2017). Large language models (LLMs) and image-generating AI are trained primarily on publicly-available posts and platforms (Pan et al., 2023). Thus, the social impact of false information extends unchecked, furthering polarization and increasing the risks of violence against vulnerable groups.

By examining the mechanisms that drive attitude change and social stigma so rampant in social media and LLM interaction, we intend to provide a meaningful framework upon which to build interventions that reduce these technologies’ abilities to further harm marginalized groups. Thus, through a multi-pronged approach where cognitive, social, and policy experts work collaboratively, the foundations for effective, evidence-based, accessible interventions may be generated. The current project has three primary aims:

(1) Evaluate properties of influence from closed-loop LLMs compared to other, static media sources (blog posts and “trusted” news sources)

(2) Determine components of LLMs which contribute most to attitude/belief/behaviour change

(3) Create closed-loop (private) LLM system and evidence-based toolkit to address influence of false information for general population use (uptake by policymakers/lawmakers, etc.)

As part of this project, you will engage in skills such as: experimental design, implementation evaluation, quantitative methods (e.g., latent variable modelling), stakeholder engagement, and multidisciplinary team coordination.

University of Reading:

The University of Reading, located west of London, England, is ranked at 194 globally, according to the QS World University Rankings 2026. 98% of research at the University is of international standing (REF 2021, combining the University’s world leading, internationally excellent and internationally recognised submissions). The University’s main Whiteknights Campus is set in 120 hectares of beautiful, award-winning parkland, less than a 30-minute train ride to London Paddington and is approximately 30 miles from London Heathrow airport.

During your PhD at the University of Reading, you will expand your research knowledge and skills, receiving specialist supervision. We also provide dedicated training in important transferable skills that can help support your career aspirations. If you need to develop your academic English skills before you start your studies, then the University has an excellent International Study and Language Institute which can help with this.

Eligibility:

• Applicants should have a good bachelor’s degree (minimum of a UK Upper Second (2:1) or equivalent)/master’s degree in Psychology or a strongly-related discipline.
• International applicants will also need to meet the University’s English Language requirements. We offer pre-sessional English courses that can help with meeting these requirements.

The University of Reading is committed to a policy of equal opportunities and non-discriminatory treatment for all members of its community.

How to apply:

Submit an application for a PhD via our online application system.