Why Children Learn Language Faster Than AI

The article traces decades of AI language research, from Chomsky's theory of innate grammar to the rise of statistical models like BERT and GPT-2. The shift to data-hungry neural networks solved fluency but at the cost of efficiency. This efficiency gap is not just academic; it has practical implications as data supply dwindles.

Researchers hope that studying child development can inspire new AI architectures that learn from less data, potentially enabling applications like video-based learning or supporting minority languages. The stakes include settling debates at the intersection of linguistics and cognitive science—questions about whether language is innate or learned, and how biology shapes our language abilities.

The commentary notes that while LLMs are powerful statistical learners, they are not brains. The challenge is to close the efficiency gap without sacrificing capability, a goal that remains unrealized but promising.