Infants outperform AI chatbots in language learning efficiency

A recent study highlights that infants demonstrate significantly greater language learning capabilities compared to current AI chatbots, with much higher efficiency. According to an analysis published by The MIT Technology Review, researchers emphasize that while AI has made notable progress in generating human-like sentences, it still lags behind human infants in both learning speed and adaptability. Michael C. Frank, a cognitive scientist at Stanford University, noted that even with recent advancements, AI systems require vast amounts of data—effectively 'burning through a forest' of information—to achieve what infants naturally accomplish in a year. Infants can begin forming coherent sentences after exposure to around 10 million to 30 million words, whereas training a model like GPT-2 with the same amount of data results in a system that produces nonsensical output. The study underscores that this efficiency gap presents a major challenge for the AI industry, particularly for companies relying on scaling language models to reach human-level intelligence. Many experts argue that this approach is inherently flawed and unlikely to achieve the desired outcome.