In an era where artificial intelligence (AI) is revolutionizing industries and everyday life, a surprising revelation has emerged: preschool children, aged 3-5, excel in visual object recognition, outperforming even the most advanced AI models. This finding, discussed in a study led by Assistant Professor Vlad Ayzenberg from Temple University, underscores a key limitation of current AI capabilities, despite its rapid advancement.
AI’s potential is monumental, with global market value projections reaching $4.8 trillion by 2033, according to a UN Trade and Development report. Its applications are vast, spanning from autonomous driving to medical imaging analysis. Yet, as this study published in Science Advances reveals, even the most cutting-edge AI models struggle to match the visual perceptual abilities of young children, particularly when it comes to swiftly and accurately recognizing objects under challenging conditions like brief exposure and visual noise.
Ayzenberg’s research, conducted with collaborators from Emory University, involved testing preschoolers’ object recognition skills with images presented at speeds as fast as 100 milliseconds, amidst visual distractions. Despite the complexity of the task, designed primarily with adults in mind, the children outperformed AI models—raising intriguing questions about the innate efficiency and robustness of the human visual system.
In stark contrast to human efficiency, AI requires extensive data and significant energy resources—training a large language model, for example, can have a carbon footprint multiple times greater than a human’s annual consumption. Ayzenberg suggests that insights into young children’s cognitive processes could inform the development of more efficient, human-like AI systems. His newly established Vision Learning and Development Lab at Temple aims to explore this by integrating behavioral, neuroimaging, and computational techniques to study early childhood cognitive development.
While AI technology impresses with its rapid growth and broad applications, this study highlights significant differences in efficiency and perceptual robustness when compared to humans, even at a young age. The potential for cross-learning between AI advancements and human cognitive insights could lead to more efficient AI models, helping to bridge the gap between current technological capabilities and the innate skills of young children. As AI evolves, understanding and mimicking the efficient learning processes of the human brain remains a pivotal goal for researchers like Ayzenberg.