A new South African study by the University of the Witwatersrand has revealed striking similarities between how children and Artificial Intelligence (AI) systems learn to communicate, and how culture plays a central role in the development of language.
In the study, scientists explored how language evolves over generations, not only among humans but increasingly in artificial intelligence systems too.
“Culture is key, as well as an understanding of ‘iterated learning’, which posits that language evolves over generations (in humans and computers) to become more structured,” noted Wits University on Tuesday, 26 May 2026.
Led by Dr Devon Jarvis, the researchers examined a concept known as “iterated learning”, the idea that language gradually becomes more structured as it is passed down and reshaped over time. According to the researchers, this process happens in both human communication and modern AI models.
“We built a computer brain with similar characteristics to a child’s, and compared it to behaviours we see in children’s brains. We then fed it data with similar properties found in human language and watched how the generations (versions) of the computer brain learn,” said Jarvis, who is also the lecturer in the School of Computer Science and Applied Mathematics (CSAM), and Fellow in the Wits Machine Intelligence and Neural Discovery (MIND) Institute.
In My Culture… language evolves
For many Africans, the findings may sound familiar.
In My Culture, language is rarely separated from identity, family or community. From isiZulu idioms and Sesotho praise poetry to Afrikaans sayings passed down around the braai, communication has long evolved through storytelling, repetition and shared understanding between generations.
The Wits research suggests this deeply human process may also explain why today’s AI tools are becoming more sophisticated.
Researchers built a computer model designed to learn in ways similar to a young child’s brain. The system was then exposed to language-like information over multiple generations of learning. Over time, the model naturally began to identify patterns and structure communication more efficiently.
The study was recently published in Proceedings of the National Academy of Sciences under the title “Compositionality and Systematicity Emerge from Iterated Learning in Deep Linear Networks“.
Why childhood learning matters in the age of AI
Jarvis explains that children learn through stages of understanding. A child may first believe all birds can fly before discovering that penguins are different. These mistakes, corrections, and repeated interactions gradually help children build more complex knowledge about the world.
Researchers found that AI systems behave in a similar way.
According to the study, language becomes easier to learn when it develops a clearer structure over time. Information that is confusing or inconsistent is more likely to disappear, while simpler and more reusable patterns survive across generations.
“While this progressive acquisition of knowledge has its benefits, the work focused on the implications for generations of learners. A child learns some language from their parents, and they will eventually pass it on to their own children. Due to the complexity of language, this transmission introduces mistakes,” Jarvis said.
Language in South Africa’s multilingual community
As AI becomes increasingly woven into classrooms, workplaces, and everyday life, the research raises questions about language and culture in multilingual communities.
South Africa remains one of the world’s most multilingual societies, with 12 official languages and rich oral traditions that continue shaping modern communication online and offline.
The Wits study suggests the future of AI may depend not only on technology itself, but on the cultural and social interactions as people learn from one another.
Read the study by Wits University here.
In My Culture… knowledge grows when it is shared by all. Visit our social media pages and share your thoughts.