Study: Is AI changing human language?

Within a short space of time, the use of large language models has changed the way people use language, warn researchers. The consequences could be far-reaching.

listen Print view
ChatGPT app on a smartphone

(Image: Tada Images/Shutterstock.com)

2 min. read

A study by the Max Planck Institute for Human Development provides the first evidence that AI models such as ChatGPT are changing the way people speak. The study of almost 280,000 YouTube videos shows that Just a few months after the release of ChatGPT in November 2022, academics began to use typical “AI words” significantly more often in their lectures.

The international team of researchers led by Hiromu Yakura systematically analyzed English-language videos from over 20,000 YouTube channels from academic institutions. They focused on words that had been identified in previous studies as characteristic of ChatGPT-edited texts, such as “delve”, “realm”, “meticulous” or “adept”.

The result is clear, the researchers write: after the publication of ChatGPT, the use of these terms in spoken language increased significantly. For example, the use of the word “delve” increased by 48 percent, “realm” by 35 percent and “adept” by 51 percent – in each case within the first 18 months of ChatGPT's market launch.

From the researchers' standpoint, the discovery is also remarkable because humans have so far mainly adapted machines to human behavior. Now the opposite is happening. This is reminiscent of phenomena from chess or Go, where humans have adopted strategies from AI systems.

Videos by heise

A further analysis of 50 randomly selected videos in which the word “delve” was used revealed that the speakers showed no signs of reading in 58 percent of cases. This indicates that the AI language was adopted spontaneously.

However, the discoveries raise major questions. For example, it is to be feared that the LLMs are accompanied by a reduction in language diversity because the vocabulary is becoming standardized. This is particularly to be feared if AI models are increasingly used in education. It should also be questioned socially whether language models could perhaps even be used for mass manipulation. However, further research is needed in this area.

The use of AI as a useful helper in everyday life could initiate a profound cultural change. The consequences for the development of human language and communication are not yet foreseeable.

(mki)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.