 |
| ChatGPT group conversation. |
The cultural feedback loop between humans and machines has officially become a two-way street—and it’s changing how we talk. New research reveals that the preferred vocabulary of AI chatbots like ChatGPT is seeping into everyday human speech, marking a measurable shift in our spoken language just years after these tools entered the mainstream.
For years, linguists warned that the sterile, algorithmic style of AI-generated text might flatten written communication. Now, evidence shows that spoken words—in YouTube videos, podcasts, and casual conversations—are also being transformed.
The Data: Tracking AI’s Linguistic Fingerprint
A team led by Hiromu Yakura, a postdoctoral researcher at the Max Planck Institute, set out to measure this phenomenon. They sifted through hundreds of thousands of hours of spontaneous spoken content and compared word usage patterns before and after ChatGPT’s launch in late 2022.
The method was straightforward but revealing: identify words that ChatGPT “prefers” when polishing or generating text—terms it selects over their synonyms—and track their frequency in human speech over time.
“We detect a measurable and abrupt increase in the use of words preferentially generated by ChatGPT, such as delve, comprehend, boast, swift, and meticulous after ChatGPT's launch,” Yakura states in the study, now published on arXiv.
https://arxiv.org/abs/2409.01754
These aren’t obscure terms; they are slightly formal, descriptive alternatives to more common words. Where a person might have once said “look into,” “understand,” “show off,” “quick,” or “careful,” the AI-influenced lexicon now leans toward its more polished cousins.
A Closed Cultural Loop Emerges
The implications go beyond vocabulary quirks. The findings suggest a profound shift: machines, originally trained on vast datasets of human language, are now developing their own “cultural traits” and feeding them back into human culture.
“This marks the beginning of a closed cultural feedback loop in which cultural traits circulate bidirectionally between humans and machines,” the authors note. It’s a cycle with unpredictable consequences for the evolution of language and expression.
As noted in a related Scientific American article, this isn’t just about style—it’s about scale and influence.
https://www.scientificamerican.com/article/chatgpt-is-changing-the-words-we-use-in-conversation/
“LLMs often amplify dominant patterns or ideas in a way that distorts their original proportions,” the research highlights. These models are shaped by the data they’re fed—primarily dominant languages and worldviews—which they then reinforce and spread.
Why It Matters: Diversity at Risk
The concern, researchers argue, is the potential erosion of linguistic and cultural diversity. When billions of people interact with models trained on a narrow slice of global language, regional dialects, indigenous knowledge, and nuanced expressions risk being sidelined.
“The richness of human expression could eventually be lost or smoothed out by the ubiquitous use of bots like ChatGPT or Gemini,” says Yakura. “They are predominantly trained in certain languages and certain preferred styles of expression they have been taught to consider of higher quality.”
In other words, the very tools meant to assist communication may inadvertently homogenize it, privileging a “higher quality” style that is, in fact, just one flavor of expression.
Looking Ahead: The Future of Human-Machine Culture
This research opens the door to vital new questions. How do we preserve linguistic diversity in the age of generative AI? Can—and should—these models be trained more inclusively? What are the risks of scalable manipulation when AI begins to shape not just what we read, but how we speak and think?
The moral, as Yakura puts it, is that “it will be hard to fight scale.” The pervasive influence of AI on language appears to be underway, moving silently from our screens into our sentences.
The challenge now is to ensure that in embracing the efficiency of AI, we don’t lose the vibrant, messy, and beautifully diverse tapestry of human speech. The words we choose next—whether “swift” or “quick,” “meticulous” or “careful”—may just depend on who, or what, we’ve been listening to.
 |
| Preferred ChatGPT wording proliferates in spoken language. |