The tables are turned: ChatGPT teaching people to speak? 360,000 videos and 770,000 podcasts have confirmed it.

You think you're in control of AI, but in fact, AI is taming you! The latest research warns that ChatGPT is changing the way English is expressed, subtly implanting its own preferences. It's time to re - examine whether we're expressing ourselves or just being AI's "repeaters".

Is the first step of AI taming humans teaching people to speak English anew?

Recently, more and more media have noticed that people are speaking more and more like ChatGPT!

This is not imitation but "invasion".

Imperceptibly, humans are being "assimilated" by AI:

AI has not only invaded your writing but also quietly occupied your mind.

Finally, it pours out from your mouth!

Watch out when speaking English! ChatGPT is brainwashing you

The latest arXiv pre - print gives a startling conclusion: ChatGPT may be quietly changing our word - using habits.

Whether you find it convenient or feel uneasy, ChatGPT has quietly become a regular guest in writing, searching, and even daily conversations, gradually reshaping our language and way of thinking.

Meanwhile, there has been a heated discussion on the Internet:

Can we spot AI - written content at a glance through details like "dashes"?

The answer may be changing.

However, this new study gives an unexpected answer: "AI traces" are becoming increasingly difficult to identify - because human language styles are becoming more and more like ChatGPT.

Researchers found that since the launch of ChatGPT, in just 18 months, the use of so - called "GPT words" in people's daily communication has soared.

This is the latest research from the Max Planck Institute for Human Development (MPIB) in Germany.

Paper link: https://arxiv.org/abs/2409.01754

Previous studies have pointed out that ChatGPT can affect human written expression. This time, researchers want to know if the rapid popularization of AI is also quietly changing our "way of speaking".

For this study, millions of pages of emails, articles, academic papers, and news reports were uploaded to ChatGPT for polishing.

Then, researchers analyzed the words that AI prefers, such as "delve", "realm", "meticulous", etc. They call these GPT words.

Subsequently, in more than 360,000 YouTube videos and 770,000 podcast clips, they tracked the frequency changes of these words and found a startling upward curve!

Figure 2: Quantifying ChatGPT's word preferences and their impact on human communication

(A) Temporal trend of the usage of the word "delve" in academic YouTube reports, showing a significant increase in its usage after the release of ChatGPT (p = 0.010 compared with the synthetic control group)

(B) Measurement method for ChatGPT's word preferences: By comparing the word - frequency distributions of human - written texts and ChatGPT - modified versions, the logarithmic ratio (i.e., GPT score) calculated can quantify the degree to which a specific word is preferred by ChatGPT

(C) ChatGPT's characteristic words (GPT word set) measured by GPT scores. Different versions of ChatGPT show consistent strong preference for "delve"

(D) Constructing a synthetic control group with a similar pre - release pattern through the convex combination of donor words for inferring causal effects

This latest study tells us an interesting fact: these "AI accents" may be increasingly difficult to detect - because human speaking styles are gradually becoming like ChatGPT.

Even after excluding the influence of synonyms and scripted content, the study found that the usage rate of GPT words has indeed increased significantly in spoken English.

This study is thought - provoking, but there are also some notable limitations.

First, the data analyzed by the researchers come from specific GPT models: GPT - 4, GPT - 3.5 - turbo, GPT - 4 - turbo, and GPT - 4o. This limits the study to these specific versions of ChatGPT.

In the next few months and years, OpenAI will undoubtedly launch new models, and these new versions may exhibit new forms of language use and word preferences.

Therefore, this study may soon become outdated.

In addition, it is still unclear whether ChatGPT actually has an impact on more casual spoken expressions, especially considering that the researchers obtained a large amount of data from academic sources.

Moreover, the use of language and words evolves over time due to various factors. Although ChatGPT may contribute to the changes in the words we use to some extent, it is important to note that many other sources in society and culture also contribute to language changes.

The rise of "pseudo - humans"

Co - author Levin Brinkmann explained to the media: "The language patterns stored in AI technology are feeding back into human thinking. Humans are naturally inclined to imitate each other, mainly those who are knowledgeable or respected. We tend to copy their language behaviors."

Levin Brinkmann: A researcher at the Max Planck Institute for Human Development, focusing on the deepening integration process between machines and human society. He uses complex system theory, machine learning, and experimental social science methods to explore the evolution laws of human - machine hybrid systems. He holds a doctorate in computational social science and a master's degree in physics from the University of Göttingen

In other words, a cultural feedback loop is forming between humans and AI: we train AI with written texts, AI sends back statistically reconstructed texts to us, and after we capture these patterns, we start to unconsciously imitate them.

The study points out that more and more people regard AI as a cultural authority, which enables "machines initially trained on human data to quantifiably reshape human culture after forming their own cultural characteristics".

Brinkmann emphasized:

"Delve" is just the tip of the iceberg. Other high - frequency GPT words include:

"underscore" (emphasize), "comprehend" (understand), "bolster" (support), "boast" (possess), "swift" (quick), "inquiry" (investigation), "groundbreak" (pioneering)...

Figure 3(a): Linear regression analysis of the occurrence frequency of high - frequency GPT words and the synthetic control group. The dots represent monthly aggregated frequencies, and the turning points marked by the dashed lines on the trend line. Words like "comprehend", "boast", and "swift" show significant growth similar to "delve"

Figure 3(b): The correlation between the GPT score of a word (x - axis) and the amplitude of its usage trend change (y - axis). The bar chart represents the 95% high - density interval of the posterior distribution, and the shaded area in the inset is the 95% confidence interval of the Gaussian process regression. Focusing on the top 20 GPT words (the rightmost part of the inset), a group of words with an annual growth rate of about 25% - 50% can be observed

Although this study mainly focuses on the word level, many researchers have noticed that the influence of AI is gradually penetrating into tone -

For example, speech becomes longer, more organized, and emotional expression is weakened at the same time.

A study at Cornell University found that using smart replies in chats makes people more cooperative and enhances intimacy because users tend to choose more positive and emotional language.

However, once people suspect that the other party may be using AI, they will feel that the other party is insincere and become more tough in attitude. The key is not whether the other party actually uses AI, but that feeling of "suspicion".

Paper link: https://www.nature.com/articles/s41598-023-30938-9

As Malte Jung, an associate professor of information science at Cornell University and a co - author of the study, said, we judge others based on language features, and what really affects our impression is the nature of the language itself.

Malte Jung's research aims to explore the interpersonal interaction dynamics in groups and teams and how these dynamics are affected by machines

Mor Naaman, a professor of information science at Cornell Tech, pointed out that the contradictory phenomenon of "AI improving communication while causing suspicion" reflects a deeper trust crisis.

Mor Naaman is the associate dean for provostial affairs at Cornell Tech, the Jacobs Institute of Technology and Cornell, and the Ann Bowers College of Computing and Information Science at Cornell University. He is also the Don and Mibs Follett Professor of Information Science. He focuses on cross - disciplinary topics such as technology and media. He obtained a doctorate in computer science from the InfoLab at Stanford University. In addition, he used to play professional basketball

He proposed three signals of the loss of humanity in AI - mediated communication