A research team at Florida State University in the U.S. said in a paper titled “Traces of AI in Unscripted English Conversations” on the 1st that the use of certain words in science and education has increased significantly in recent years. This indicates that words used a lot by AI are also increasing significantly in people’s conversations.
The research team analyzed 1,326 episodes of podcasts in which people freely talk without scripts. The purpose of the study is to analyze only what people say without AI. The study was divided into before and after 2022 when ChatGPT was released, and 11 million words were collected and compared at the same rate.
As a result, the number of uses of the word “surpass” has increased by 140.79 percent since 2022, while “boast” has also increased by 140.14 percent. In addition, “strategically,” “align,” and “significant” have increased by 36.59 percent, respectively, by 17.35 percent. All of these words were frequently used by AI.
On the other hand, the increase in the number of octopus ‘delve’ (explore) words, which are well known to be used excessively by AI, was smaller than expected. In addition, words such as ‘realm’ (area) and ‘crucial’ tended to decrease in use. This suggests that the influence of AI may appear differently in writing and speaking.
The research team concluded from these results that “human vocabulary choices are gradually converging with AI vocabulary patterns.” It raised the possibility that AI began to have a direct impact on the human language system itself beyond simple tools.
The research team analyzed that the AI language pattern has the effect of “taking in” the human language. Repeated exposure to AI-generated languages can be internalized in people’s language habits, the researchers said.
This has not only a relatively minor problem of language selection, but also a wider social implication. If AI uses biased language related to political and social beliefs, this ‘spinning effect’ can also affect values and beliefs across society.
Today’s language becomes the AI training data of tomorrow, the AI output of the next day, and eventually becomes part of the language that humans encounter, said Professor Tom Uzek, a computational linguist who conducted the study. “This circular structure foreshadows a new era in which AI and human language interact and affect each other.”
JENNIFER KIM
US ASIA JOURNAL
