AI is still dumber than a baby – at least when it comes to learning a language

If a human learned a language at the same rate as ChatGPT, it would take them 92,000 years. Which is not impressive. Not impressive at all.

Even the smartest machines cannot beat toddlers at language learning, according to a new paper by scientists from the Max Planck Institute for Psycholinguistics in the Netherlands.

“While machines can crunch massive datasets at lightning speed, when it comes to acquiring natural language, children leave artificial intelligence in the dust,” the institute said.

This is mostly because children use all their senses, such as seeing, hearing, smelling, listening, and touching, to make sense of the world and build their language skills.

Meanwhile, machines – at least for now – primarily learn from written text and do so passively rather than through an active process driven by growing social, cognitive, and motor skills.

This means that AI misses out on the rich mix of sights, sounds, and other sensory signals that help humans learn language. Also, unlike machines, children do not just wait for the language to come to them – they actively explore their surroundings, constantly creating new opportunities to learn.

All this enables children to master language so quickly, according to the lead author of the paper, Professor Caroline Rowland. “AI systems process data ... but children really live it,” she said.

Understanding how children learn a language is imperative for AI developers if they want to build better models.

“AI researchers could learn a lot from babies,” Rowland said.

“If we want machines to learn language as well as humans, perhaps we need to rethink how we design them – from the ground up.”

Rowland published the paper in Trends in Cognitive Science. Co-authors include researchers at the ESRC LuCiD Centre in the UK.