Pengi: an implementation of a theory of activity

TLDR

AI traditionally models everyday activity as plan-following, yet real-world complexity demands continuous moment-to-moment improvisation beyond preplanned actions. By analyzing routine activity dynamics, the authors identified regular patterns in how simple systems interact with their surroundings. Using these dynamic theories, they built Pengi, a program that performs complex, seemingly planful tasks without explicit world models.

Abstract

AI has generally interpreted the organized nature of everyday activity in terms of plan-following. Nobody could doubt that people often make and follow plans. But the complexity, uncertainty, and immediacy of the real world require a central role for moment-to-moment improvisation. But before and beneath any planning ahead, one continually decides what to do now. Investigation of the dynamics of everyday routine activity reveals important regularities in the interaction of very simple machinery with its environment. We have used our dynamic theories to design a program, called Pengi, that engages in complex, apparently planful activity without requiring explicit models of the world.