Tool promises to make lazy academics' AI-written papers sound more human

There's a certain audacity to building a tool whose entire job is covering the tracks of another tool. That's essentially what MorphMind has done with Academic Humanizer, a Claude skill designed to strip the telltale verbosity and inflated phrasing out of AI-assisted academic drafts. University of Minnesota associate professor and MorphMind cofounder Jie Ding frames the problem as one of voice: AI drafts drift from an author's natural style and sacrifice the precision that scholarship demands. So the fix, apparently, is more AI — layered on top to smooth out what the first round of AI already produced.

The tool can be pointed at a researcher's earlier published work, letting it mimic that person's phrasing so the final output sounds less like a bot and more like them. Ding insists this isn't about generating new content or dodging peer review — it's positioned strictly as an editing aid for clarity, not a content generator. The GitHub readme backs that up, warning that using Academic Humanizer doesn't erase a researcher's obligation to disclose AI involvement, whether AI helped with parts of a paper or wrote the whole thing.

But there's an obvious gap here. If the tool only polishes prose and doesn't touch findings, data, or citations, it also has no mechanism for catching what's underneath. Weak arguments, shaky evidence, or sloppy reasoning don't get flagged — they just get dressed up to sound more convincingly like a human wrote them. And that's a strange feature to ship into a research ecosystem already struggling with output quality.

The backdrop makes this look worse than a niche curiosity. Researchers at the University of Surrey flagged a wave of formulaic, superficially-analyzed papers last year that bore all the hallmarks of LLM authorship. More recently, detection outfit GPTZero found 100 hallucinated references scattered across 51 papers that had actually been accepted at NeurIPS, one of the field's marquee AI conferences. Meanwhile MIT research suggests students who lean on AI to write essays show reduced brain activity and retain less of what they supposedly learned.

So the timing of a tool built specifically to make AI-generated academic writing harder to spot is, at minimum, awkward. It doesn't create the underlying problem of AI slop flooding journals and conference proceedings. It just makes the symptom quieter while the disease keeps spreading underneath.