Belgian researchers model next-generation AI children’s language development

The Belgian researchers approach to AI is radically different to that of the current generation of large AI language models such as ChatGPT that rely entirely on text statistics. 

Namur University’s Katrien Beuls said in a press release that "Children learn their mother tongue by communicating with the people around them”.

“As they play and experiment with language, they try to interpret the intentions of their interlocutors. By doing so they gradually learn to understand and use language constructions. The process by which language is acquired through interaction and meaningful context, is at the heart of human language acquisition.”

Models such as ChatGPT learn by observing huge amounts of text to see which words often occur close to each other, to enable them to generate texts that are often indistinguishable from those written by humans.

VUB’s Paul Van Eecke “This creates models that are extremely powerful when it comes to many forms of text generation, from summarising or translating texts to answering questions. However, at the same time they exhibit a number of inherent limitations".

The Belgian researchers proposed an alternative in which artificial agents learn by participating in meaningful interactions to develop language that is linked directly to the environment in which they operate and their sensory perceptions.

This will create models that are less prone to bias because their understanding of language is based on direct interaction with the world and uses data and energy more efficiently. These models will also understand language and context in a more human-like way.

“Integrating communicative and situated interactions into AI models is a crucial step in the development of the next generation of language models. This research provides a promising path towards language technology that is closer to how humans understand and use language.”