Acting on Information: Representing Actions That Manipulate Information

Abstract

Information manipulation is the creation of new information based on existing information sources. This paper discusses problems that arise when planning for information manipulation, and proposes a novel action representation, called ADLIM, that addresses these problems, including: How to represent information in a way sufficient to express the effects of actions that modify the information. I present a simple, yet expressive, representation of information goals and effects that generalizes earlier work on representing sensing actions; How to concisely represent actions that copy information, or produce new information that is based on existing information sources. I show how this is a generalization of the frame problem, and present a solution based on generalized frame effects; and How to generate a pipeline of information-processing commands that will produce an output containing exactly the desired information. I present a new approach to goal regression.