AI and the World's Hard Problems · Life sciences · arXiv · 2026-10-20
Simulating Infant Senses with Age-Tuned Networks
Measurements of infant senses over the first two years were taken separately for each sense, so the authors combine them into age-tuned networks that simulate sensory perception. Each sense is modeled as a fixed, age-indexed input filter with a small prediction network trained behind it. In vision, the networks extract far less than the input allows at early ages, with a critical period and perceptual narrowing emerging without being built in, while hearing tracks the input; the same filters power the free Baby Senses Lab website, which is a research and teaching tool rather than a diagnostic test.