Linguists say ChatGPT is now influencing how humans write and speak

Researchers are finding that not only is AI learning from human-generated content, but it’s also influencing how humans write and speak, homogenizing our style and vocabulary:

• Words that abound in AI-generated text like “delve, “meticulous,” and “adept” have taken over the human lexicon in both written and spoken communication.
• For example, linguists from the Max Planck Institute in Germany recently found that academic YouTubers began using words favored by AI chatbots up to 51% more frequently after the release of ChatGPT.

While technology has always shaped the way we write and speak (thx, phones), this is the first time in history that machines are directly dictating communication style. So, let’s delve into consider what AI becoming a linguistic force heralds for our oral and written expression.

It’s not just ChatGPT addicts

AI tools are integrated into software ranging from word processors to messaging apps, successfully enticing millions of users to seek assistance from a machine when composing text. But even people who don’t use AI can end up parroting it, famed linguist and AI critic Emily Bender told the Atlantic. She believes that the profusion of synthetic text online and its use in email communications make it impossible to dodge the tech’s influence.

In the case of the academic YouTubers, the Max Planck Institute study suggests that reading ChatGPT-generated term papers may be leaving a mark on their speech patterns. The researchers found that in most instances, AI’s favorite word, “delve,” showed up in many videos that appeared to be unscripted.

Are we becoming rambling bores?

Linguists say that the influence of AI-generated texts is making language cookie-cutter, flattening the quirks and eccentric irregularities that make words pop off the page. Experts point out that this could make human writing less effective, since text peppered with clichés and buzzwords is less likely to keep readers interested.

Meanwhile, some scholars are concerned that AI is homogenizing regional and international variations of English, making it less likely that someone will casually describe their affairs as “hunky-dory”:

• A recent Cornell study found that AI may be influencing the writing style of Indian English speakers to sound more American.
• UC Berkeley researchers found that ChatGPT preferred Standard American English and was more likely to produce exaggerated caricature-like outputs when prompted by speakers of other dialects.

Besides getting us addicted to a stock set of words and turns of phrase, AI is also making us wordier, because chatbots tend to prefer long-winded, flowery prose. People produced more verbose Facebook Marketplace listings after being exposed to ChatGPT-written ones, a study by Swinburne University of Technology in Australia, which has yet to be peer-reviewed, showed. Listings written post-ChatGPT exposure were 87 characters on average, up from the 33 characters participants cranked out before they saw the AI-generated version.

And it impacts how we view others’ communication

Research indicates that AI-assisted communication can also influence how people perceive each other. A recent Cornell University study found:

• When “smart replies” were used in work chats, trust often improved, as AI tends to use more emotionally positive language.
• But people who suspected their colleagues were sending AI-generated messages were more likely to perceive the sender as less collaborative and more demanding, raising the possibility that AI use might undermine interpersonal relationships.

We can adapt

Some experts claim that identifying the problem sets humanity on the track to combat it, and there’s already a backlash brewing to address what’s seen as AI dimming the sparkle of human language.

Many academics are consciously avoiding AI-coded terms in their work, while one professor told the Atlantic that some of his students are refusing to use AI, taking pride in their ability to naturally formulate crisp sentences—the expressive equivalent of folks who do all their grocery shopping at their local organic farm.

Even AI users have some control over how their synthetic writing sounds. Ritesh Chugh, an IT professor at Central Queensland University in Australia, suggests adjusting settings within AI tools to prompt them to avoid repetition, use precise language, and personalize tone.

If you don’t want to sound like a robot… try your best to avoid relying on these words and phrases preferred by ChatGPT.—SK

This report was originally published by Morning Brew.