The Evolution of Language: The LLM Revolution and Future of Writing
How will LLMs affect the future of writing and communication?
As an aspiring lawyer (or perhaps a better description is “wanna-be-lawyer”), it is hard to resist the temptation to use ChatGPT or another Large Language Model (LLM) for my work. While I enjoy writing, it is undoubtedly a laborious and difficult process. Using tools like ChatGPT to finish my work quickly and painlessly is deliciously tempting. Ignoring this urge requires an active effort to stay focused and avoid using these tools.
Despite my best efforts, I sometimes succumb to the temptation and use AI to help me write. Why? Am I just lazy and unwilling to exert the effort required to compose a new article? Probably yes, but that’s not the whole picture. Unfortunately, ChatGPT frequently is better at explaining a concept than I am. Even if I only intend to use ChatGPT as a brainstorming tool, it generates clear and concise written content that is easy to include in my final product. If I am writing about an unfamiliar topic, the content created by an LLM is typically cleaner and more comprehensive than anything I could muster.
This week, I had a project to research and write about the 2010 US Stock Market Flash Crash. I was roughly familiar with what transpired and the general causes, but writing an accurate analysis required that I conduct additional research. After about two hours, I had a better grasp of the timeline and a few of the specifics, so I started cobbling together some vague language on “Algorithmic High-Frequency Trading” and “E-minis.” However, due to my tenuous understanding of the material, I struggled to write anything I deemed worthy of publishing. So, what did I do? I begrudgingly turned to ChatGPT.
In a matter of minutes, ChatGPT created a comprehensive overview of the Flash Crash and provided details on specific areas I wanted to highlight for my project. Even with the time it took to edit the materials to better align with my project (and to make it sound less like AI-generated content), I finished in about half an hour. All told, using ChatGPT was a much faster (and easier) process than arduously conducting research and created a significantly better written deliverable than anything I had put together.
As I derided myself for letting the machines win one over on me, this led me to the central question of this article: How will LLMs affect the future of writing and communication? Specifically, as ChatGPT and other LLMs create more online (and print) content, will we all start to write more like AI?
Before I can answer this, I first want to touch on the field of linguistics. Without delving too far into theory, language is a set of agreed-upon rules and norms for communicating information. Since these rules are in place to help guide effective and clear communication, they are constantly updated to respond to changes in how people communicate. For example, in July 2023, a Canadian judge ruled for the first time that a “Thumbs Up” emoji could represent agreement to a legally binding contract.
Beyond the generic rules governing the English language, there are also linguistic discrepancies in the same language across different groups and regions. A frequently cited example of this phenomenon is African American Vernacular English, a variation of modern English with unique grammar and vocabulary used by a large segment of American and Canadian society. One of the most beautiful parts of the English language is that it is akin to a living organism by constantly adapting to its environment.
Since language is constantly evolving, I believe that modern English may begin to imitate the style of LLMs for the following reasons: (1) Exposure and Familiarity; (2) Normalization of AI-Generated Language; (3) Ease and Efficiency; (4) Standardization and Consistency; and (5) Cross-Cultural Communication. Before anyone starts to wonder, yes, I used ChatGPT to help me brainstorm some of the ideas here. Here’s a little more on each of these:
- Exposure and Familiarity: People will be increasingly exposed to AI writing styles as LLM-generated content becomes more prevalent across various platforms (Internet forums, news sites, speeches, curriculums, books, the list could go on). Familiarity with such language patterns might lead individuals to adopt similar communication styles.
- Normalization of AI-Generated Language: As AI-generated content becomes increasingly commonplace, readers might not even realize they are consuming materials generated by LLMs. Consequently, this normalization could lead to the acceptance and integration of AI stylistic language patterns into everyday communication.
- Ease and Efficiency: AI language models generate text that is not necessarily accurate but is cogent, coherent, and grammatically correct, which may appeal to users. The efficiency and clarity of these outputs may incentivize writers to emulate these clear writing patterns.
- Standardization and Consistency: The style of LLMs provides a consistent and standardized way to write and communicate. By standardizing a writing style, certain language conventions may become more widespread and accepted across all forms of communication.
- Cross-Cultural Communication: One of AI’s greatest applications is facilitating communication across language barriers, enabling people speaking different languages to interact. Therefore, certain linguistic features might blend and influence communication, especially if the users are not native speakers or did not learn the language in school.
With all these brilliant reasons spinning through your heads, let’s return to my original question: How will LLMs affect the future of writing and communication, and will we all start to sound more like AI? Unsurprisingly, the key to answering this is to focus on the readers’ age. Overall, I believe that children (and, consequently, future generations) are more likely to communicate in ways that mirror LLMs than adults, who have a stronger command of language.
Children learning to read and write are the group most vulnerable to writing and communicating more like LLMs. Child development and education researchers often describe children’s brains as “sponges” when they learn language, meaning they pick up anything and everything they hear and use it to understand the world. As children learning to read interact with an increasing amount of material (both in pure volume and percent of overall written work) created by AI, they are likely to pick up some of LLMs stylistic ticks. (If you’re interested in how children learn language, here, here, and here are some interesting links!)
Beyond just the increased interactions that students will have with AI-generated materials, there is also a push to integrate AI tools into the classroom. One of the leading initiatives proposes deploying AI “chatbot tutors” to provide personalized, one-on-one academic support to students. While this idea is still nascent, proponents see potential benefits such as democratizing access to tutoring, freeing up teachers, and allowing self-paced learning.
While there are likely some benefits to incorporating AI tutors into the classroom, the rise of these tools may lead students to inadvertently learn and imitate the stylistic norms of the LLMs tutoring them. The more students interact with the AI tutors who give them the answers to math, English, grammar, history, and science questions, the more likely they are to start following the writing style and structure of their AI tutors.
Comparatively, adults or older students with a better command of language are less likely to replicate AI. I anticipate the most significant change in adults’ writing styles will be a rise in the amount of written material generated by LLMs that we pass off as our own. Consequently, while the words we actually write may not imitate ChatGPT, the overall content produced will begin to resemble the writing style of LLMs because we will include more AI-generated material.
Okay, so let’s say my hypothesis is true: although adults are less likely to imitate the writing style of LLMs, children learning to read and write will be more vulnerable to imitating AI. What are the societal and individual implications of this phenomenon?
Most importantly, adopting LLMs’ writing styles may erase linguistic diversity. English has diverse voices, writing styles, and vernaculars unique to specific people and places. African American Vernacular English, which I touched on earlier in this article, is a single example of the rich diversity of English dialects and variations. Linguists have identified over 150 different vernaculars in the English language (and over 30 in the United States alone) that offer immense cultural value to those who use them. The danger of imitating LLMs is the potential loss of these idiosyncratic communication methods.
Secondarily, it will be increasingly challenging to identify AI-generated materials, including content written by “bots.” Right now, it is usually pretty clear when you come across LLM-generated writing. The style is robotic and formulaic, and something just feels off about the writing. Fortunately, developers created AI text detection tools, such as GPTzero. These tools are, for the moment, pretty good at detecting AI-generated writing. Additionally, these tools may be the only way for teachers to check if students write their essays or copy them from AI tools. However, if people start to imitate the writing style of LLMs, then these tools will slowly become less accurate and may fall into obsolescence. As a result, teachers will be less able to identify when a student has copied material from an LLM. Identifying bots on websites (such as Reddit) will also be more challenging if more people write like an LLM instead of using their unique voice.
So what can be done? Is the lure of LLMs and AI too strong to resist? Have we doomed future generations to communicate like robots? Probably. However, hopefully, there is something that we can do. Circling back to the beginning of my article, I think the best solution is in front of us: take the hard route when it comes to writing. Resist the temptation to use LLMs. Rely on our minds, not AI tools, to create content. While these tools make the process easier, working without them makes the quality of our writing higher and the content we create more unique. Mastering the craft of writing is difficult, but the best way to preserve it for future generations is to write in your voice and use LLMs sparingly in the writing process.
As an aside, while I was talking to a friend about this idea, he brought up an interesting point: LLMs speak in the style of the data used to train them, and since most of the data is from 1980 to 2020, it will likely be a while before they change their writing style. The volume of written materials from this period is enormous and is the baseline for all LLMs, meaning that until there is new written material, the baseline writing style will likely remain static for the foreseeable future. I know this is not directly relevant, but I thought it was a great addition!
TLDR. Here is a ChatGPT-generated summary of this article:
The article discusses the author’s struggle as a writer in resisting the temptation to use AI language models like ChatGPT for content generation. The author explores the question of how large language models (LLMs) will impact the future of writing and communication, arguing that the increasing use of LLMs may lead to a convergence in writing styles towards AI-generated content. The article identifies reasons for this potential shift, including exposure, normalization, ease, standardization, and cross-cultural communication. The author suggests that children learning to read and write may be more vulnerable to imitating LLMs, leading to the erasure of linguistic diversity. The article concludes with a call to preserve individual voices in writing and use LLMs sparingly.
