AI-Generated Academic Articles Can Be Distinguished From Human Texts
A study published this week in the journal Cell Reports Physical Science has found that AI-generated academic articles can be distinguished from human texts using a simple set of standardized processes. The study, led by Heather Desaire, a professor at the University of Kansas, aimed to create an accessible method that even low-tutored high school students could use to detect AI writing.
Non-Techies Can Help to Recognize AI
Desaire explained the goal of their project: “We did our best to create an accessible method so that even low-tutored high school students could use an AI detector for different types of writing can build.” She continued in an article published in TechExplore, “There is a need to address AI writing skills, and people don’t need to have a computer science degree to contribute to this field.”
The study highlighted the need to differentiate human writing from AI, as ChatGPT has enabled access to artificial intelligence (AI) for the masses, revolutionizing the way people work, learn, and write. To meet this need, the researchers developed a method to distinguish text generated by ChatGPT from (human) academic scientists.
Simple Characteristics Can Distinguish Humans From AI
This method relied on standardized “classification methods”, using relatively simple characteristics to distinguish humans from AI. For example, science writers tend to write long paragraphs and use more “ambiguous” language, such as words like “but”, “however” and “although” to nuance a point they are making. AI, on the other hand, does not doubt itself.
With a set of 20 such characteristics, the researchers built a model that the author refers to as human or AI. This model was more than 99 percent accurate. “This strategy could be further adapted and developed by others with basic skills in supervised classification,” they argue, “enabling access to many highly accurate and targeted models for detecting AI use in academic writing and beyond.”