This paper addresses the relations between information retrieval (IR) and AI. It examines document retrieval, summarising its essential features and illustrating the state of its art by presenting one probabilistic model in detail, with some test results showing its value. The paper then analyses this model and related successful approaches, concentrating on and justifying their use of weak, redundant representation and reasoning. It goes on to other information management tasks and considers how the concepts and methods developed for retrieval may be applied to these, concluding by arguing that such ways of dealing with information may also have wider relevance to AI.