Visual XOR tasks are easy for monkeys

Abstract

Monkeys exhibit great difficulty learning and performing visual classification tasks in which class assignment depends on conjunctions of features — in particular, the exclusive-or (XOR) of two distinct features. This suggests that they might not readily or naturally utilize holistic configural information when identifying images, instead relying upon single diagnostic cues. We hypothesized that familiarity and/or expertise could transform a “hard” XOR task into an easy one, comparable in difficulty to a one-dimensional classification task (in which each object may be identified by a single feature). To address this, we first overtrained monkeys on a one-dimensional classification task. New hybrid stimuli were then constructed by dividing each familiar stimulus into two pieces and re-pairing these pieces within each class. When shown these hybrid stimuli without differential reinforcement, monkeys classified them according to their component parts, reliably and with no impairment to reaction time. They were then trained to classify the hybrids into the opposite categories as their component pieces, while maintaining performance on the original stimuli. They were able to learn this expanded XOR task much more easily and quickly than a simpler XOR classification task involving unfamiliar stimuli. We also investigated performance on the hybrid classification task after prolonged passive exposure to the original stimuli. It appears that monkeys solve classification tasks by finding a diagnostic feature for each object. When this strategy is insufficient, they have great difficulty learning the task. However, familiarity seems on some level to engender complete representations of the stimuli, even though only a small part of each stimulus may be behaviorally relevant. This allows for performance on an XOR task comparable to performance on one-dimensional tasks, possibly by dividing the XOR task into two separate one-dimensional tasks.