Opinion: Learning Intuitive Physics Requires More Than Visual Data

Humans expertly navigate the world by building rich internal models founded on an intuitive understanding of physics. Meanwhile, despite training on vast quantities of internet video data, state-of-the-art deep learning models still fall short of human-level performance on intuitive physics benchmarks. This work investigates whether data distribution, rather than volume, is the key to learning these principles. We pretrain a Video Joint Embedding Predictive Architecture (V-JEPA) model on SAYCam, a developmentally realistic, egocentric video dataset partially capturing a child's visual experience. Although this dataset contains 0.01\% of the data volume used to train SOTA models, our model performs comparably on the IntPhys2 benchmark. Our results suggest that merely scaling up the size of visual datasets or using a developmentally realistic dataset is insufficient for current architectures to learn robust representations that support intuitive physics reasoning.