AI models are always hungry for vast amounts of information, so it can learn how to give us human-like responses. For example, OpenAI’s GPT-3 trained on 570 gigabytes of text, or roughly 225 billion words. But children and babies, however, do not require an enormous pool of information to learn how language works. They listen and try things like imitating the sounds of people around them – a different way of learning that’s more efficient and consumes much less energy. Is it possible for AI to learn more like young children? UC Berkeley Linguistics Professor Gašper Beguš thinks so.