Kids outlearn AI—and we still don’t know why
Interesting AI research context but no immediate implications for Australian government AI governance, strategy, or practice.
Key points
- Children acquire language from far less data than LLMs require, and researchers do not yet understand why.
- The article traces how Chomskyan linguistics shaped early rule-based AI and contrasts that with modern statistical LLMs.
- Limited direct relevance to APS AI governance or policy work - this is cognitive science and AI history.
View original source
Copied.
"Kids outlearn AI—and we still don’t know why"
Source: MIT Technology Review – AI
Published: 24 August 2026
URL: https://www.technologyreview.com/2026/08/24/1141740/kids-machines-language-learning/
MIT Technology Review profiles ongoing scientific debate about how children acquire language so efficiently compared with large language models. Researchers note that GPT-2 trained on 30 million words produces incoherent output, while children achieve fluent language from far sparser exposure. The piece traces the influence of Chomsky's generative grammar on early symbolic AI, the subsequent AI winter, and the eventual vindication of statistical learning through transformer-based LLMs. The science remains unresolved; the article is exploratory rather than findings-focused.
Retrieved from SIMS, 16 September 2026.