How AI Hides Itself: Academic Humanizer for Academic Documents is Born

A tool designed to erase the stylistic traces of AI from scientific papers is dividing the academic world. It is called Academic Humanizer and was launched on GitHub by Jie Ding, an associate professor at the University of Minnesota and co-founder of the startup MorphMind.

The functionality does not involve a dedicated app: it is essentially a skill for Claude, a set of instructions that anyone can upload into their preferred AI model to proofread drafts already generated by another artificial intelligence. The declared target is 'papers and grant proposals,' meaning scientific articles and funding requests.

The set of rules instructs the model to eliminate so-called linguistic ticks of AI, such as the excessive use of long dashes, which are now recognized markers of artificially generated text. Ding specifies that the goal is not to flatten the text, but to align the writing with the author's voice, also aiming the tool at the user's previous works to replicate their style. The system is also designed to strengthen unsupported scientific claims with additional evidence, a detail that goes beyond simple stylistic rewriting.

Criticisms and Ding's Response

After initial reports on the risk of misleading use, Ding modified the GitHub page and replaced the phrase 'removes typical traces of AI' with the more neutral 'improves clarity and voice.' In Nature, he stated, 'I would distinguish the tool from the behavior,' while emphasizing that the ethical problem concerns the failure to disclose the use of AI and not the existence of the tool itself. He also added an explicit note reminding that Academic Humanizer does not exempt the author from the obligation of disclosure regarding the AI assistance received.

The reaction from the scientific community remains divided between those who see the tool as a valid aid for non-native English speakers and those who fear it will become a means to mask content entirely generated by a model. This would exacerbate the already known problem of the production of poor and standardized academic texts, increasingly difficult to distinguish from genuinely original content.