AI is making massive strides in its ability to produce human-like writing, and it is becoming almost impossible to detect.

AI writing is no longer in the elementary-school era, and LLMs are steadily advancing their ability to become more humanlike than previously. Even new AI detection tools may not be able to identify AI from human efforts, as new research indicates.

As of 2024, the sophistication of AI-generated books and articles is rising rapidly. Of major concern is that AI-generated research poses ethical concerns in academia, where originality is paramount, making it challenging to produce truly undetectable work. But it is also a concern for anyone who is a professional writer or aspires to be one, as well as MFA programs in colleges.

In two separate studies, participants who were not trained poetry readers failed to recognize AI-generated poems at a rate higher than chance (46.6%, N= 16,340). Notably, when asked to identify the author of a poem, participants were more likely to respond that a person rather than an AI wrote it.

AI-generated poems were mistaken for human-written poetry because they scored higher on rhythm and elegance. Due to the similarity between human and AI poetry, participants may have confused the simplicity of AI poetry with human complexity.

Why Poetry May Be Safe From AI

On the other hand, AI-written poetry might not be a threat to poetry writing in most cases. Some have stated that no matter how good LLMs are at writing different types of content, they will never be able to produce poetry of any quality because poetry is all about meaning and creativity and AI-generated prose is boring and lacks imagination. Poetry's subjective nature, defying strict criteria and often inverting meaning, makes it difficult to understand and analyze, especially for non-experts. AI follows rules and its learning and neither fit into human poetry writing; creativity is key here.

What did one study propose? Is it that poems created by AI were more popular? It could be that the simplicity of AI poetry is a contributing factor to its higher overall rating across all measures. The study found that AI-generated poems were easier to understand and relate to than human-authored poetry. When asked to explain their reactions, participants used the phrase “makes little sense” more frequently when explaining poems written by humans. Metaphors and complexity, again, are easier to process than human text and may make reading poetry simpler for readers who wish for simplicity.

So, while poetry may seem to be a challenge for AI, the fact that it beat Shakespeare in writing poetry should give us pause. Anyone can use LLMs to imitate any writer, including us, which means they could pass off work written by an LLM as their own. If we wanted Hemingwayesque writing, we could do it. Is this writing as we would like it to remain, or is it the inauthenticity that we find in painters who copy great works of art? We’ve seen them sitting, sometimes, in museums where they are copying a painting.

Where Are LLMs Going Regarding Writing?

Researchers investigating LLMs and their creative writing capabilities beyond poetry in 2024 are primarily focused on determining the limits of these programs. They are parsing out crafting intricate plots, developing compelling characters, and capturing nuanced human emotions in fiction, including short stories and longer works.

Evidence suggests that LLMs frequently produce formulaic and predictable narratives, devoid of the unique ideas and varied viewpoints present in works of creative writing by humans. Their training data is to blame for this, since it encourages the use of cliches and tropes. But in the world of AI, nothing is stagnant and learning marches on.

A significant challenge for LLMs is the construction of complex plotlines with interesting conflicts and unexpected turns, which frequently leads to predictable and linear narrative structures. Computer engineers are looking at ways to improve the collaborative creative process by guiding LLMs to generate more imaginative and interesting stories through prompts and user feedback.

In addition, research is looking into new ways to train LLMs on more complicated and varied literary data, such as world-building, character-arc analysis, and theme exploration components. All the elements are there, and the training to produce them is being created.

Prompts are the wave of the future, and programs exist to help writers develop better prompts for their original work through algorithms. A quick look at YouTube videos will reveal how to get hundreds of prompts from someone or how to write better prompts. Currently, programs can provide precise titles for creative work, outlines for articles or stories, and immediate rewrites for something provided seconds ago. The programs never tire, but the tokens do pile up.

What is a token? They are the equivalent of bits of text that writers pay for (in paid programs), which can be thousands of words per month. Yes, tokens are the wave of the here-and-now and the future. If you use up your tokens, you can buy more to keep writing with the AI programs, most of which may keep your content on the cloud rather than on your computer.

For anyone strapped for funds, some free programs will provide you with monthly tokens (often around 3K). Free programs can summarize or write content based on research articles you upload. This can be invaluable to anyone who needs research summaries quickly and can be an important educational assist to students. As Sal Khan (developer of The Khan Academy) believes, AI may save education.

Writing will not be solely the province of AI or computer engineers, but they can provide new tools to make writing more provocative, informative, and smoother. Effort in individual input in our writing remains a given, even when AI is providing content for us. Everyone knows about programs that catch AI-produced products, but that old adage, don’t throw the baby out with the dishwater, still applies. Use AI intelligently, and it can make all of us better writers.