The primary aim of my research is to leverage the respective strengths of knowledge graph reasoners and large language models. While LLMs can efficiently generate highly expressive natural language output, their probabilistic nature precludes consistency as a guaranteed property. Further, questions remain as to LLMs’ ability to succeed at more complex abstract reasoning tasks. In contrast, the logical basis of knowledge graph reasoners allows for consistency checking, transparency, and verifiability, but at the cost of efficiency and scale.
Intuitively, bridging these approaches would allow us to reap their respective benefits, and to mitigate their weaknesses. As such, my work focuses on the development of a novel neuro-symbolic reasoner. I am particularly interested in abstract and numeric reasoning, temporal reasoning, and reasoning about semantic change over time. In the context of safe and trusted artificial intelligence, this project aims to secure explainability and verifiability, without sacrificing the human character of natural language outputs. I was motivated to join the CDT mainly due to its interdisciplinary approach to AI, and by my own belief in the importance of algorithmic transparency as a means of protecting the interests of all present and future stakeholders.
–
Undergraduate Qualification: BA Philosophy, Sarah Lawrence College
Postgraduate Qualification: MA Philosophy, King’s College London