Language in a loop: AI-generated text and its influence on our language

How many times have you used the word “delve” in the past five years? And how many times have you seen this word lately? As a non-native speaker, I was not even aware of this word until March 2023, which is when I started using ChatGPT.

Now I see it several times a day in various contexts in the digital space. It made me think of the substantial changes the use of textual genAI has on the formation of the language we use.

The rise of text-generating artificial intelligence (AI) marks a significant shift in our interaction with written content. Yet, this innovation carries an overlooked consequence: a cyclical pattern where AI's language output is influenced by human-generated content, which then shapes human language practices.?

Feedback loop: The cyclical influence of AI on language

This loop begins with AI algorithms that are trained on vast datasets of human-generated text. These datasets include everything from literary works and news articles to everyday online communication.

An essential facet of this feedback loop is the nature of generative AI itself, which operates largely on the principles of probability. Generative AI systems, when constructing responses or generating text, favor words and phrases that are most commonly used in their training datasets. This is because the algorithms are designed to predict and produce language based on the likelihood of word usage, learned from vast amounts of text data.

This probabilistic approach, while efficient in generating coherent and contextually appropriate text, can inadvertently lead to a prevalence of certain linguistic patterns. As these patterns are more frequently generated, they become increasingly common in AI-produced content, creating a self-reinforcing cycle. The more a particular word or phrase is used in the training data, the more likely it is to be used by the AI in future outputs.

This aspect of AI's design is crucial in understanding the perpetuation of certain language trends and the potential narrowing of linguistic variety. It's a technological echo where the most likely becomes the most common, further influencing the language that feeds back into the system.

It is not just about the overused words and phrases

To illustrate the effect of the feedback loop, consider some of the commonly overused words and phrases in AI-generated content. Their frequent use exemplifies the tendency towards linguistic homogenization.?

For instance, the phrase "embark on a journey" has become a cliché, often used in various contexts ranging from corporate communications to motivational speeches. Similarly, the word "delve" is frequently employed in academic and professional writing, to the point of overuse. These examples highlight how repeated language patterns can diminish the freshness and originality of our expressions.

What this means for our future?

I see now two possible scenarios unfolding in the future, and I want to show you these on the example of two of the best known sci-fi novels.

A/ Negative scenario - George Orwell's "1984"

In "1984," Orwell introduces Newspeak, a language designed by a totalitarian regime to restrict freedom of thought. It’s characterized by a diminishing vocabulary, the elimination of certain words, and simplified grammar.

This fictional language reflects the regime's desire to limit the capacity for rebellious or independent thought. By controlling language, Newspeak aims to control the very way people think and perceive reality.

Relate this to the potential consequences of an uncontrolled AI-language feedback loop. If AI prompted by us to do so, continues to promote a narrower range of language, could we unintentionally edge towards a Newspeak-like scenario where linguistic simplification occur?

B/ Positive scenario - Neal Stephenson's "The Diamond Age"

Contrast Orwell’s dystopia with the world of "The Diamond Age," where technology serves as a liberating force. The novel features the "Young Lady's Illustrated Primer," a smart, interactive book that adapts to the reader’s needs.

The Primer not only teaches Nell, the protagonist, how to read and think critically but also exposes her to a rich variety of ideas and cultures.

Future unfolding

I would like to see a future where AI, like the Primer, is used to expand and enrich our linguistic abilities. In this scenario, AI tools could introduce users to a diverse range of languages and dialects, thus enhancing our creativity and cultural appreciation.

Our current trajectory with AI-generated text and its influence on language will likely determine which of these futures becomes our reality.

Global impact

The AI language feedback loop, while here discussed in the context of English, has global implications, potentially impacting the diversity in non-English languages, because other? languages, particularly those less commonly spoken, have smaller digital footprints.?

This means that when AI systems are trained, they may have a more limited range of text samples in these languages. Consequently, the linguistic patterns that the AI reinforces, could be even more narrow than in English, impacting the richness and diversity of these languages.

Conclusion

It is ultimately up to us, as individuals and as a society, to decide how we use these powerful tools. Will we let AI-generated text narrow our linguistic richness and thought diversity, or will we use it as a catalyst for enhanced creativity and deeper understanding?

The choice is ours to make. What we need for the positive scenario to unfold, is a pro-active approach. That includes education about how AI and GenAI works and about its strenghts and weaknesses and how to use the tools to the reach the best possible results, together with accessibility for all to try and test for themselves.

With the wealth of resources, including free GenAI tools, tutorials, and guides, we stand on the brink of a promising future.

Sources:?

"1984" by George Orwell: 1984 - Wikipedia

"The Diamond Age" by Neal Stephenson: The Diamond Age - Wikipedia

Disclaimer: This article was brainstormed and drafted with the help of a ChatGPT persona that I have created for this type of collaboration. The final text was revised by ChatGPT for grammar and stylistic mistakes.