Cog Lunch: Ben Lipkin "Solving composite tasks with modular systems"
Description
Speaker: Ben Lipkin
Title: Solving composite tasks with modular systems
Abstract: Humans routinely solve composite problems, such as scheduling a doctor's appointment around personal and professional constraints, which requires complex coordination of language, social inferences, situation modeling, and formal reasoning. While strong evidence from neuroscience supports an architecture for the mind and brain as being composed of a collection of specialized modules, a detailed account of how these modules are orchestrated to work together, and the format of the interface between them, is an active area of debate. In parallel, pockets of research from AI have been concerned with building such modular structures into systems, frequently via neuro-symbolic approaches, and often struggling with similar issues about the format of the interface between components. In this talk, I will briefly discuss a combination of existing research projects as well as an early-stage project proposal within this broader space. I hope to promote questions and discussion.