AI is transforming human cultures and societies at a rapid pace; yet the AI ethics landscape is cross-sectoral, interdisciplinary, and highly complex. We are using computational methods (e.g. natural language processing, graph computing) to map topics, trends, and networks within the various disciplinary academic literatures, amongst policy-makers, and in the communities of practice beyond, to identify knowledge gaps, and deliver embedded research for impact….
Creativity is central to the arts and sciences – and even to the human condition. The question of whether, and if so how, AI can achieve it is crucial for understanding ourselves and our AI futures – a matter we have discussed in TEDx talks. Working alongside creative practitioners, we are exploring the role of AI in scientific discovery (via neural, symbolic and neuro-symbolic modelling), the aesthetics of AI artworks, whether LLMs can understand and make genuinely creative use of language, and other key questions; and we have developed a prototype automated theorem prover….
Our informational environment has been revolutionized since the advent of the Internet, and is increasingly influenced by (classificatory and generative) AI. We are interested in empowering people through education to engage with information critically and constructively. We have investigated the psychology and pedagogy of critical reasoning, and built the Logic Calculator and the Digital Diploma; engaged in knowledge exchange with an AI-powered EdTech start up LearnerShape; and used information design to produce effective communication tools for stakeholders in the information landscape. We are also computationally investigating human questions and answers in online fora to improve prompt engineering….
Simulation is a common technique in the natural and social sciences. It is less common (though not unheard of) in the humanities. We have developed and are exploiting the PolyGraphs simulation framework to conduct philosophical research surrounding the nature of rational opinion formation (in individuals and groups), the effectiveness of various information processing strategies under adverse conditions, and the truth-conduciveness (and other pro-democratic characteristics) of various social network structures. Our Python code is scalable, allowing us to explore large datasets on realistic networks (e.g. on the National Internet Observatory, or through access enabled under the EU’s Digital Services Act); and it is built to enable machine learning, on the deep graph learning library….