“`html
In the span of five days after the launch of ChatGPT in 2022, the AI chatbot attracted over a million users. Currently, more than half of all individuals under the age of 30 and almost half of those aged 30-49 report having utilized it, as per Pew Research Center.
Recent investigations from Florida State University’s Department of Modern Languages and Linguistics, Department of Computer Science, and Department of Mathematics are uncovering significant proof that AI jargon frequently employed by chat-oriented large language models is no longer confined to screens — it’s often manifesting in verbal communication.
Their study, titled “Model Misalignment and Language Transformation: Traces of AI-Associated Language in Unscripted Spoken English,” represents the inaugural peer-reviewed investigation into whether the informal human language framework is evolving following the widespread embrace of chat-based large language models such as ChatGPT. It gained acceptance at the eighth Conference on AI, Ethics, and Society, organized by the Association for the Advancement of AI and the Association for Computing Machinery, scheduled for October. The findings will be published in EIES Proceedings as part of the conference.
“This study addresses a key concern in discussions surrounding AI and language: are these linguistic alterations occurring because we engage with a tool and echo its suggestions or is language evolving due to AI’s impact on the human language framework?” stated assistant professor of computational linguistics and principal investigator Tom Juzek. “Our analysis of lexical developments before and after the 2022 emergence of ChatGPT indicates a convergence between human lexical choices and patterns associated with LLM-related AI language.”
While quick surges in the utilization of certain terms — for instance, Omicron — can happen, these spikes often result from real-world occurrences. Recent broad increases in the usage of terms like “delve” and “intricate” in specific domains, particularly education and scholarly writing, can be attributed to the extensive introduction of LLMs equipped with chat functionalities, such as ChatGPT, which tends to overuse those phrases.
“The alterations we observe in spoken communication are quite striking, especially compared to historical patterns,” Juzek remarked. “What is especially notable is the extensive nature of change: numerous words are experiencing significant growth in usage over a relatively brief timeframe. Given that these are primarily terms commonly overutilized by AI, it is reasonable to hypothesize a connection.”
Terms such as “surpass,” “boast,” “meticulous,” “strategically,” and “garner” have also experienced substantial increases in usage following the launch of ChatGPT. Although these terms are typically employed in a formal or academic context, making them less prevalent in spontaneous spoken communication, researchers uncovered that nearly three-quarters of these target words exhibited heightened usage, with some seeing their frequency more than double.
The research team, which includes Bryce Anderson, a senior in computer science, and Riley Galpin, a junior double-majoring in computer science and pure mathematics, evaluated 22.1 million words from spontaneous and unscripted spoken discourse, incorporating conversational podcasts on science and technology. After the advent of ChatGPT, researchers noted a quantifiable rise in the prevalence of AI-associated jargon but not their synonyms. For instance, the AI term “underscore” demonstrated a significant rise in usage whereas its counterpart “accentuate” did not. FSU scholars remarked that this indicates we’re not merely utilizing AI; the language embraced by AI is gradually infiltrating everyday conversation.
“Language serves as the most potent medium of communication available to humanity, and comprehending how AI will influence this medium is fundamentally crucial and timely,” remarked Anderson, the paper’s lead author. “My drive to pursue this investigation originates from observing AI extend the boundaries of what’s achievable across major sectors, coupled with the understanding that this influence transcends mere tool usage — it can shape societal dimensions, including our linguistic practices.”
This study builds upon the team’s earlier research into AI’s effects on written scientific English, which revealed profound structural transformations brought about by AI in scientific writing. Galpin presented findings on written language at the 38th International Florida AI Research Society Conference in May, and Anderson will accompany Juzek in showcasing the team’s new findings related to spoken language at the AIES Conference in October.
“Our work underscores numerous vital ethical considerations,” Galpin stated. “As LLMs have the capacity to sway human language, broader questions arise regarding how model biases and misalignment, or differences in LLM behaviors, might start to affect human conduct.”
This study was conducted as part of FSU’s Undergraduate Research Opportunity Program, which offers high-achieving undergraduates a chance to delve into their research interests alongside faculty members.
“I am eager for future inquiries in this domain,” Juzek expressed. “While this presents a significant advancement in understanding AI’s impact on language change, a major open question for upcoming work is whether AI is amplifying traditional patterns of language evolution or directly instigating them.”
To discover more about research undertaken in the Department of Modern Languages and Linguistics, visit mll.fsu.edu.
The post On-screen and now IRL: FSU researchers find substantiation indicating ChatGPT alters our speech appeared first on Florida State University News.
“`