Kids outlearn AI—and we still don’t know why
People have been talking to each other for at least 100,000 years, as best we can tell. And in all that time, there has been only one thing in the world that could learn a human language to perfect fluency: a human child. Now there are two. Four short years after the release of ChatGPT,…
People have been conversing with each other for at least 100,000 years, but only human children can learn their native language to perfect fluency. Now, artificial intelligence has caught up with this ability, with models like Claude, DeepSeek, and GPT from OpenAI being able to converse naturally with users. However, creating AI that can use human language is far more demanding than it might seem.
An artificial intelligence model can process far more words than a person would experience in mastering their first language, significantly more than children might hear by their first birthday. This discrepancy between children and AI is known as the data efficiency gap. Researchers are trying to figure out how children can outperform the most advanced language models, which has implications for both AI research and cognitive science.
Kids who grow up in linguistically rich environments may have heard up to 300 million words by age 20. This vast difference in the amount of information processed by AI models and human children highlights the need for more data-efficient AI. Studying human language acquisition could help create AI models that use less data, which could be useful for training AI on video or creating chatbots for minority language communities.
Written by urgent.news from MIT Technology Review's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
