As AI systems become embedded in consequential decisions, the question shifts from 'does it work?' to 'does it behave as intended, remain fair over time, and fail safely?' This theme addresses specification, monitoring, and governance of AI systems.
Theme Overview
AI systems are rarely built from scratch — they are trained, deployed, and updated in messy real-world environments. This theme focuses on the requirements and engineering methods needed to keep AI systems trustworthy: before deployment (specification and fairness requirements), during operation (runtime monitoring), and as they evolve (agentic AI guardrails and governance frameworks).
Related Publications
Research on runtime monitoring approaches for ML systems — ensuring deployed models continue to satisfy their behavioural requirements as data distributions shift. Led by PhD student Hira Naveed.
Empirical study identifying ethical concerns raised by users of LLM-powered healthcare apps — informing a requirements framework for responsible clinical AI. Led by PhD student Yutan Huang.
Systematic approach to eliciting, specifying, and verifying fairness requirements for AI systems — covering protected attributes, fairness metrics, and their translation into testable requirements.
Analysis of safety guardrail mechanisms for autonomous LLM-based systems — covering bounded autonomy, scope specification, and fallback behaviour requirements.
Practitioner-facing analysis of the ethical dimensions of integrating LLMs into requirements engineering workflows — covering bias, accountability, and transparency. Invited article for the International Requirements Engineering Board publication.
People
Related PhD Research
Interested in collaborating or contributing to this research?
Get in Touch →