I Used AI to Fake a Scientific Paper. Here's What Happened (spectrum.ieee.org)

An experiment conducted by an author aimed to reveal the growing threat of AI-generated content in academic publishing, particularly through a trial submission of a fake scientific paper to the IEEE Access journal. The results showed that low-quality, AI-generated articles—coined as "AI slop"—are becoming prevalent, with projections suggesting that by 2025, 57% of published academic papers may exhibit detectable LLM influence. This situation raises significant concerns about the integrity of the scientific record and underscores the potential erosion of trust in the academic process as AI automates and simplifies the production of research. The author used free tools like ChatGPT and Gemini to generate a plausible paper on optical solitons, managing to create content that included fabricated references. Although the initial submission was flagged for ghost citations, the ease with which a non-expert could produce a convincing paper signifies a concerning trend. IEEE's research-integrity manager noted that the existing peer-review process is becoming increasingly strained, highlighting the need for more robust mechanisms to combat the influx of AI-generated submissions. This challenge underscores broader issues in the publish-or-perish culture, which is intensifying pressures on academics to produce more, potentially compromising the quality and integrity of scientific publication.