Distributed Algorithms as a Gateway To Deductive Learning (Invited Talk)

With the book Thinking Fast and Slow, Daniel Kahneman popularized the idea that the human brain can think in two different modes. The fast mode is instinctive and automatic, while the slow mode is deliberative and logical. As of 2023, one can argue that machine learning understands how to think fast. Deep neural networks are remarkably successful in rapidly classifying and regressing data. Thinking slow on the other hand is still a mystery. Large language models may provide an illusion of being able to think slow. However, prompts that need multiple deductive steps are generally beyond the capabilities of large language models. Distributed algorithms have the potential to help understanding deductive reasoning. Distributed algorithms usually consist of several little steps, iteratively applied, each step being easily learnable. As such distributed computing may provide an interesting bridge towards understanding deduction, extrapolation, reasoning, and everything else needed to think slow. In the talk, we will discuss some exciting case studies from graph generation to origami folding.