The Challenges of Using AI to Improve Writing

A growing number of users of generative AI are noticing a paradox — it can make your writing sound more professional, but also more generic.

Valerie Sumner, founder and principal of VRS Meetings & Events, Inc., is an avid user of AI platforms, including Spark AI and ChatGPT, which she uses for a lot of event-related analysis and tasks. But Sumner told reporters that while she experienced AI’s ability to improve productivity, she also encountered AI’s limitations. Convening.

For example, when Sumner used AI to help her draft an event marketing email for a client, she created a template with language tailored to the client’s needs and asked ChatGPT to generate 15 different versions. It immediately produced results, including suggested fixes and alternative language. But something was missing, Sumner says: her voice. The grammar and sentence structure created by generative AI may be perfect, but content generated by generative AI sounds a lot like… AI. You can tell it’s AI-produced copy, Sumner says, “after you use it enough — it doesn’t sound like me.”

Valerie Sumner

Sumner’s observation that AI-generated writing tends toward homogeneity is consistent with ongoing research into the trade-offs we might make as we offload more writing tasks to ChatGPT. On the one hand, according to research by Oliver Hauser, an economist at the University of Exeter, and Anil Doshi, a professor of strategy and entrepreneurship at the University College London School of Management, AI is not only outperforming humans in reducing the time it takes to create content, but it’s also objectively improving the creativity and writing skills of most people. In their study, they found that writers who had access to five AI-generated ideas wrote stories that were, on average, 8% more novel and 9% more useful than writers without AI. Using generative AI to “professionalize” writing, Hauser and Doshi wrote in a paper published in July. Scientific progress.

But there’s a catch: When the researchers looked at the AI-generated stories as a group, they found that they lacked diversity compared to the group that didn’t use AI. Their results suggest that — as Sumner observed — as more writers use AI to help them craft their stories, their work “will become less unique overall and more similar to one another,” they write.

In a recent Instagram post, content creator and software engineer Abi Bouhmaida listed dozens of the most common words and phrases in ChatGPT, including “innovation,” “integration,” “thought-provoking,” “exemplary,” “consistent,” “enhancement,” and “transformation.” “It’s like you asked a million people to answer your prompt, and ChatGPT gave you an ‘average’ of all their responses. Its output will be generic and vague,” Bouhmaida wrote in the post. “ChatGPT doesn’t give you the best word for every situation, it just gives you the most common words.”

Hauser and Dorsey note in their paper that generative AI is a rapidly evolving technology whose full potential has yet to be tapped. Advances are so rapid that the generative AI model they used in their study, ChatGPT-4, “may soon become obsolete.” They add that their study is just “a first step in understanding the relationship between generative AI and human creativity.”

Meanwhile, Sumner developed a working relationship with the generative AI and her own creativity—whenever the AI ​​faded her voice, she put it back in.