A cognitive memory designed to record sensory input patterns, visual, auditory, tactile, etc., stores them into memory folders of conventional design. Sensory inputs are recorded only if they are “interesting.” Recordings are placed into folders that have available space. Each folder is connected to a multiplexer that in turn feeds training patterns into an individual, autoassociative neural network. Training is ongoing at all times. Recorded patterns are retrieved in response to prompt inputs. These are sensory input patterns derived from the same sensors that provided signals which were recorded into the folder originally. The autoassociative neural network connects the prompt patterns to patterns recorded in the many folders of the memory system, if recorded patterns can be found that are like the prompt patterns. When there is a “hit,” the contents of the respective folder are discharged onto the memory output line.