Proceedings of the Annual Meeting of the Cognitive Science Society

Psychologists recognize Raven’s Progressive Matrices as an effective test of general intelligence. While many computa-tional models investigate top-down, deliberative reasoning on the test, there has been less research on bottom-up perceptualprocesses, like Gestalt image completion, that are also critical in human test performance. We investigate how Gestalt vi-sual reasoning on the Raven’s test can be modeled using generative image inpainting techniques from computer vision.We demonstrate that a reasoning agent using an off-the-shelf inpainting model trained on object photographs achieves ascore of 27/36 on the Colored Progressive Matrices, which corresponds to average performance for nine-year-old chil-dren. When our agent uses inpainting models trained on other datasets (faces, places, and textures), performance is lower.Our results illustrate how learning visual regularities in real-world images can translate into successful reasoning aboutartificial test stimuli, and also how different learning inputs translate into different levels of performance.