Grant: Canada Research Chairs

The recent strides in artificial intelligence (AI), epitomized by technology like ChatGPT, have garnered widespread enthusiasm from both the general public and policy-makers alike. Indeed, AI harbors the potential to significantly alter the economic and social landscapes of societies in unforeseen ways. The present era underscores a pressing need to cultivate a profound scientific comprehension of the core principles that govern AI-based systems. It is imperative to pinpoint fundamental constraints, such as misalignment with human values and algorithmic unfairness (when such systems are used in domains like the criminal justice system, or the micro credit market, for example), and devise robust strategies to discern and alleviate the repercussions of misuse. The proposed research program aspires to formulate mathematical methodologies to address these intricate technical challenges, and device algorithms where possible, to solve them. To make this vision come to life, the proposed research program will focus on three main concrete issues associated with misalignment and scale of AI systems: (1) Fairness in AI systems; (2) Robustness to changes in the environment of AI systems; and (3) Neural Scaling Laws for understanding how to better allocate resources for building aligned AI systems.