Paradigms of Artificial General Intelligence And Their Associated Risk
This project seeks to identify forms that AGI could take and the types of risks they may pose.
Much of this work focuses on the evaluation of AI systems, specifically their capabilities, generality, and safety.
A thorough understanding of these properties of AI systems is essential for ensuring that AI is beneficial to humanity.
Robust Evaluation of Cognitive Capabilities and Generality in Artificial Intelligence
I am a post-doc on the RECOG-AI Project at the Leverhulme Centre for the Future of Intelligence. RECOG-AI aims to improve on cognitive evaluation for AI systems, taking inspiration from Comparative Psychology and Psychometrics.
Automating Abstraction For Potential Based Reward Shaping
This was my PhD project. Here, I looked at creating methods for agents to learn their own abstractions for Reinforcement Learning. This automation of the abstraction process helped agents to glean more useful information about their experiences and improve their learning speed.