A More Rational Theory of Memory?

Our experience in the world is very much like a continuous recognition experiment: We have a continuous stream of experiences, some new and some old. There is one big difference with a typical recognition experiment: there is not an experimenter controlling the order of these experiences. Lael Schooler and colleagues have shown that the statistical structure of these real world experiences predict performance in controlled memory experiments. Mining two large data bases of human experience (Twitter messages and Reddit comments) and using a prescription for optimal retrieval (the Reciprocal Square-Root Law), we developed an algorithm that can make a-priori predictions for speed of recognition in any continuous recognition experiment. We discuss its performance in experiments that either have controlled experimental statistics or real-world statistics or a mixture of the two.