Events

Learning Curves: Machine and Human Learning

Speaker: Daniel Rothschild

Location: 5 Washington Place, Room 101

Date: Friday, April 19, 2024

Recent advances in machine learning have revitalised the empiricist program by providing direct demonstrations of learning through experience. Vast neural networks that start out with random configurations develop impressive skills in domains like perception, strategic play, and conversation by gradually learning from experience. Despite these successes, there are reasons to think that AI systems today learn in a different, and, in some respects, slower way than humans do. In this talk, I explore this potential weakness in machine learning, bringing out different senses in which learning by AI today may be less powerful than human learning.