Newborn embodied Turing tests: Benchmarks for unified models of animal development

Building unified models of animal intelligence remains a central challenge in science. A key obstacle is the lack of benchmarks for testing whether models develop like animals. To directly compare the development of animals and models, both must be raised in the same environments and tested on the same tasks; otherwise, differences between animals and models could stem from differences in learning mechanisms, training data, or some combination of the two factors. Here, we review matched-experience Newborn Embodied Turing Tests (NETTs): a new class of developmental benchmarks for comparing learning across newborn animals and embodied computational models. Brain models are inserted in body models then raised in virtual environments that match the rearing conditions of animals. Models are evaluated based on their ability to simulate the developmental trajectories of animals. NETTs provide a principled foundation for building unified, scalable, and mechanistic models of animal intelligence.