We study how to build and deploy AI responsibly. Our work spans explainability, privacy, security, AI literacy, and the ethics of data collection. We bring together researchers from machine learning, theory, cybersecurity, human-centered computing, and beyond. We're interested not only in technical solutions but how those solutions play out in real world situations, with particular attention to the socio-technical context of AI construction and adoption.
Our research spans across multidisciplinary research
Effective governance of AI requires frameworks that can keep pace with rapid technological change, that are grounded in empirical understanding of how AI systems are actually built and used, and that take seriously the perspectives of diverse stakeholders. Our research sits at the intersection of technical AI expertise and policy, drawing on human-computer interaction, cybersecurity, and sociotechnical systems research to inform how AI should be regulated, governed, and deployed in the public interest.
Our work engages with questions of AI literacy and public understanding, how local, state, and federal governments provision and deploy AI, the governance of AI in creative and co-creative contexts, and the design of institutional mechanisms for oversight and accountability. We collaborate with policymakers, civil society organisations, and industry partners, and we are committed to research that is legible and useful beyond the academy, contributing to standards bodies, policy consultations, and public discourse about the future of AI.
This research aims to strengthen the governance of AI by producing evidence-based insights that can inform regulation, institutional design, and public policy. By bridging technical understanding and policy expertise, our work helps governments, organisations, and communities make better-informed decisions about where and how AI should be deployed, what safeguards are necessary, and how accountability should be structured — contributing to an AI ecosystem that is not only capable, but genuinely responsive to democratic values and public needs.
Associate Professor Kanchana Thilakarathna, Professor Joseph Davis, Dr Mary Lou Maher, Dr Katy Gero, Dr Clément Canonne
AI systems are increasingly being used to make or inform consequential decisions, in healthcare, finance, education, and beyond. Yet many of the most powerful models remain opaque, offering little insight into how or why they reach a given output. Our research addresses this by developing methods that make AI systems more interpretable, transparent, and worthy of human trust. We work across machine learning, human-computer interaction, and interactive systems to understand both how explainability can be built into models and how explanations are actually received and used by people.
Our work includes developing explainable machine learning methods that surface the reasoning behind predictions, and studying how users interpret transparency cues in multi-agent AI interfaces, AI-powered news systems, and conversational agents. We are attentive to the gap between technical explainability and meaningful human understanding: we recognise that a system can be technically transparent without being genuinely interpretable to the people who use it, rely on it, or are affected by its decisions.
This research aims to improve the trustworthiness of AI systems by making their behaviour more understandable to the humans who interact with and depend on them. By combining technical work on explainable models with empirical research into how people actually make sense of AI outputs, our work supports better human oversight, reduces the risk of misplaced trust or unwarranted scepticism, and helps ensure that AI systems deployed in high-stakes settings can be appropriately scrutinised and held accountable.
Professor Irena Koprinska, Professor Eduardo Velloso, Dr Sasha Rubin
The development of AI models depends on enormous quantities of data, whether that be text, images, code, or traces of human activities. Much of this is produced by human labour that is poorly compensated, rarely acknowledged, and often extracted without meaningful consent. Our research examines the ethical dimensions of how data is sourced, curated, and used in AI development, with particular attention to the people whose work underpins these systems. We bring together perspectives from human-computer interaction, information systems, and the social study of technology to surface the values and power relations embedded in AI data pipelines.
Our work investigates questions such as how creative writers understand and respond to their work being used as AI training data, how incentive structures in data labelling and annotation shape the quality and fairness of AI outputs, and how recommendation systems can be designed to reflect genuine human diversity rather than narrow optimisation targets. We are interested in both the structural conditions that produce problematic data practices and in the interventions that might address them.
This research aims to make AI development more equitable and accountable by surfacing and addressing the ethical dimensions of data collection and human labour in the AI pipeline. By documenting harms, identifying systemic incentive gaps, and engaging directly with affected communities, including creative workers, data annotators, and platform users, our work informs better practices for AI developers, clearer expectations for platforms, and stronger policy frameworks for protecting the people whose contributions make modern AI possible.
Dr Katy Gero, Professor Joseph Davis
As AI systems grow more capable and pervasive, they increasingly depend on vast quantities of personal data. This raises urgent questions about how that data is collected, processed, and protected. Our research develops rigorous mathematical and systems-level approaches to privacy in AI, with a focus on differential privacy, federated learning, and encrypted data analysis. We work at the intersection of machine learning theory and practical deployment, building tools that allow powerful models to be trained and applied without compromising individual privacy.
Our work spans the design of differentially private algorithms for statistical learning, techniques for machine unlearning that allow individuals' data to be effectively removed from trained models, and privacy-preserving systems for sensitive applications such as video moderation and encrypted network traffic classification. We are interested in both the theoretical foundations of privacy and in engineering systems that bring those guarantees into real-world use.
This research aims to make AI systems safer and more trustworthy by ensuring that powerful models can be built and deployed without exposing sensitive personal information. By developing privacy-preserving techniques with provable guarantees, our work reduces the risk of data misuse in high-stakes applications such as healthcare, communications, and content moderation — helping individuals, organisations, and governments build confidence in AI systems that handle sensitive data responsibly.
Dr Clément Canonne, Associate Professor Kanchana Thilakarathna
