Title
Challenges for a modern vision of AI
Abstract
While historical visions of AI have typically been cold, analytical, and robotic, AI in 2026 does not fit this mold. Instead, many of the pressing questions today have more to do with social intelligence, creativity, and generally capturing aspects that might be considered more human than machine. In this talk, I will walk through our recent work exploring these questions for today’s large language models (LLMs). For instance, probing the extent to which LLMs can be creative or unpredictable, and characterizing the ongoing challenges for these models in simulating other systems or processes.
Bio
Dr. Peter West is an Assistant Professor at the Department of Computer Science, where he researches natural language processing and artificial intelligence, with a focus on the capabilities and limitations of large language models and other generative AI systems. His work explores how model behaviour differs from human intuition, how language models generate creative and unpredictable responses, and how algorithms can unlock model capabilities without additional training. He takes a natural sciences approach to AI, investigating what these models can do, where they fail and how their behaviour differs from human intelligence.
