Editable and Traceable Language Models for Accountable Human-AI Interaction University of East Anglia in United Kingdom

This PhD project at the University of East Anglia focuses on developing editable and traceable language models for accountable human-AI interaction. The research addresses critical questions about how deep language models acquire knowledge, store memory, exhibit bias, and fail, such as through hallucination or misaligned content generation. You will build, train, and evaluate transformer-based and retrieval-augmented generative models from the ground up, leveraging high-performance computing resources and specialized datasets, including parent-children interaction language.

Evaluation will be conducted along dimensions of responsible AI, including safety (harmful outputs, unintended behaviors, jailbreaks), security (robustness to adversarial inputs and data poisoning), and accountability (tracing outputs back to training data or internal representations). The project culminates in deploying these models on embodied AI systems or social robots, such as Furhat Robots, to conduct real-time, face-to-face human-AI interaction experiments. These experiments will help identify where, why, and how the models succeed or fail in sensitive domains like education and healthcare.

The School of Computing Sciences offers a vibrant research environment, collaborating with multinational companies (Apple, BT, National Trust, Aviva), research institutes in the Norwich Research Park, and other universities and industries in the UK and overseas. The school is also a member of the Turing University Network, advancing world-class research and skills development. The successful candidate will contribute to laboratory support activities for undergraduate and postgraduate courses in Artificial Intelligence, Data Science, and Computing Sciences.

Funding is available to UK applicants through a Faculty of Science funded studentship, covering 'home' tuition fees, an annual stipend for three years, and a research training support budget. Applicants should have a strong academic background in Computer Science, Artificial Intelligence, Machine Learning, or related fields, with experience in deep learning, language models, and high-performance computing considered advantageous. No explicit language test or GPA requirements are mentioned.

The application deadline is June 18, 2026. Interested candidates should apply online via the FindAPhD project link, submitting a CV, academic transcripts, and a cover letter outlining their suitability for the project. For further information, contact the School of Computing Sciences. This opportunity is ideal for those seeking to advance research in trustworthy and reliable AI for deployment in sensitive domains.