Empirical evidence of Large Language Model's influence on human spoken communication (AI column)

Have you ever wondered how our language works and where we get our vocabulary from? So scientists from the Max Planck Institute for Human Development also thought about it and decided to conduct a study of oral speech and how it was influenced by large language patterns. With written speech, everything is clear for a long time, but oral speech required research. Its methodology was as follows:

- Phase one: Researchers uploaded millions of pages of emails, essays, scientific articles and news materials to ChatGPT, asking the AI to polish the texts. They then highlighted the words that ChatGPT used significantly more than the original authors. - Stage two: The team analyzed more 360 000 YouTube videos and 771 000 Podcast episodes recorded before and after the release of ChatGPT. The total volume of the analyzed content was over 740 000 hours of human speech. - Control group: To ensure the reliability of the results, the scientists created “synthetic control groups” from synonyms that ChatGPT used less often, such as “examine” and “explore” instead of “delve”.

As a result of the study, the authors found the emergence of closed feedback loops between humans and AI in terms of culture. Study co-author Levin Brinkmann explains this phenomenon by saying that people tend to emulate those who are considered authoritative sources. And AI can very confidently and structurally present material, making it an expert source in the eyes of people. And this leads to the adoption of the vocabulary, as well as the nature of the conversation – people’s speech becomes longer, more structural and less emotional.

The effect is particularly noticeable in science, education and business. The greatest impact was found in academic YouTube channels, where the use of GPT words has grown in popularity. 25-50Percent a year. The study also showed that the influence of AI contributes to the unification of English on the American model. All this leads to social consequences in the form of loss of cultural diversity.

It is funny that scientists have found a paradox: if others are aware of the frequent use of AI by a person, its eloquence is perceived negatively. Interviewers find them less engaged and empathetic, especially in online communication.

I wonder what other implicit consequences of AI we are already experiencing and what they will lead to in the end.

#AI #Software #PopularScience #GenAI