Strands of Memory

Software-intensive systems in difficult environments often have to operate without timely human control. That level of autonomy argues for systems that have highly capable self-and situation-awareness processes, to protect the system, perform its tasks, and also to predict its external environment, so that appropriate plans can be made as far in advance as possible, and modified as conditions change. For this level of adaptability, we have long advocated self-modeling systems, which are systems that have models of their behavior, that they interpret to produce that behavior. That means that changing the models changes the behavior. In this paper, we describe two aspects of the memory in such systems, as reflected in two properties about the management of situation awareness: How to navigate the world and not get surprised (very much). How to recognize situations when they recur (soon enough to do something useful). Our conclusion is that the system must maintain many many “strands” of memory, that is, partially completed stories about what the environment has done, is doing, and is expected to do, and each one of the strands is (usually implicitly) waiting for some kind of extending, supporting, confirming, or refuting observations or other evidence. The details to be determined are about what those observations can be, what levels of abstraction are required, how they are determined, how the stories are managed, and how they relate to each other. This paper is a first step in that direction.