A new “humanizing” academic tool aims to personalize the tone of research papers written with an artificial intelligence program, in part by erasing apparent signs of AI use. Some researchers praise the tool, but others express concern.

The tool, launched on June 20, is designed for “articles and grant proposals,” according to its developer, Jie Ding, a machine learning researcher at the University of Minnesota in Minneapolis. It can be used to “match the author’s own voice,” according to the tool’s GitHub site, which lists under “basic principles” the need to “strip AI says without casualizing.”

The new humanizer is “not sophisticated,” says Max Spero, CEO of a New York City company that makes an AI sensing platform called Pangram. In its initial tests of “humanized” text, Pangram caught most, but not all, of the AI-generated language. Spero says improved versions of Pangram are being designed specifically to detect the use of humanizers.

Scientists are increasingly turning to artificial intelligence systems to help them write papers, apply for grants, and even conduct peer reviews. AI assistance can be a huge help, especially for people writing in a language other than their first language, and many editors allow some degree of AI use in article preparation.¹ if such use is declared.

AI linked to explosion of low-quality biomedical research articles

Some favorable reviews about the humanizer come from scientists. “I’m using the tool a lot,” mainly to help me write emails and code documentation, says Francisco María Calisto, a health informatics researcher at the University of Lisbon. “It’s the best thing I’ve ever used.”

Other researchers say humanizers are not good for science. “I don’t like it,” says Miguel Ángel Blázquez Rodríguez, a plant biologist at the Polytechnic University of Valencia in Spain. “It’s misleading.”

Rodriguez and others worry that humanizers will tempt more scientists to use AI without disclosing it. “I’m afraid the use case is detrimental to science,” says Cassidy Sugimoto, an information scientist at Carnegie Mellon University in Pittsburgh, Pennsylvania. “I’m worried.”

When asked about these concerns, Ding responded, “I would separate the tool from the behavior,” adding that failing to disclose AI assistance when needed is misconduct, regardless of how a piece of text was produced. “The ethical issue is non-disclosure and the intent behind it, not the existence of editing aid.”

After being asked about the potential for deception, Ding made updates to the GitHub site, such as changing a description of the tool’s function from “removes common AI prompts” to “sharpens clarity and voice.” He also added an ethics and disclosure note that he says clarifies “that the tool is an editing aid and does not remove the author’s obligation to disclose AI assistance.”

AI scribes

Humanizers are gaining more and more popularity among researchers using AI, Calisto says, because all the different AIs “sound the same.” The tone of AI-generated text is sometimes inappropriate for academic writing, Ding says. For example, the tools can exaggerate the strength of a scientific claim, he says. And most AI results have a similar style, which readers tire of, he adds.

‘Obviously ChatGPT’: How critics accused me of scientific fraud

Ding adapted a humanizer previously developed for academic content. Their version tells an AI to remove your ‘tics’ from a text and add more evidence for unsupported scientific claims. Anyone can copy Ding’s guidelines into their AI and tell it to apply the rules, which include avoiding phrases like “not just X, but Y” and eliminating hyphens. Both constructs have been associated with AI, according to the tool’s website.