Visualisation patterns in visual reasoning tasks with different complexity levels: insights from human and machine approach

Abstract

Task complexity plays a crucial role in cognitive development, problem-solving skills, and effective learning. Engaging with complex tasks stimulates higher-order thinking, boosts learner motivation, and prepares students for real-world challenges and advanced studies. This study investigates how different levels of task complexity (low, medium, high) affect visual attention sequences and contrasts human scanpaths with those derived from the Compositional Language and Elementary Visual Reasoning (CLEVR) dataset’s computational model. Using eye-tracking technology, we analyzed the visual attention patterns of Indian students (ages 18–35) working on visual reasoning tasks of varying complexity within a controlled laboratory setting. Our findings show that greater task complexity leads to more dispersed and prolonged fixations, reflecting a shift in attention sequence and a more thorough search strategy. Additionally, we compared human scanpaths with the CLEVR model and found notable differences. Human search behavior is more intricate, featuring frequent revisits and considering various attributes such as color, shape, size, and material. In contrast, the CLEVR model operates with a more linear and streamlined approach. These results highlight the significant differences between human and algorithmic visual reasoning processes. The complexity and frequent revisits observed in human behavior demonstrate the adaptable nature of human cognition. This study underscores the importance of integrating task complexity into educational strategies to enhance cognitive development, engagement, and the alignment of educational tools with human cognitive processes, ultimately improving learning outcomes.