Autobiography Based Prediction in a Situated AGI Agent

Abstract

The ability to predict the unfolding of future events is an important feature of any situated AGI system. The most widely used approach is to create a model of the world, initialize it with the desired start state and use it to simulate possible future scenarios. In this paper we propose an alternative approach where there is no explicit model building involved. The agent memorizes its personal autobiography in an unprocessed narrative form. When a prediction is needed, the agent aligns story-lines from the autobiography with the current story, extends them into the future, then interprets them in the terms of the current events. We describe the implementation of this approach in the Xapagy cognitive architecture and present some experiments illustrating its operation.