A recent scroll through social media provides an immediate, sometimes startling lesson in the modern crisis of trust. For many users, a significant portion of their daily feed is now filled with what has colloquially come to be known as "AI slop"—low-quality, machine-generated content that astute observers can quickly spot as artificial. This digital material often serves to embarrass the unsuspecting individuals who unknowingly share it, while occasionally amusing the posters who distribute it deliberately, either for entertainment or as a calculated tool to sow confusion across public platforms.
This modern landscape stands in stark contrast to the early days of generative artificial intelligence, particularly the era of GPT-3. Back then, users could ask a chatbot virtually any question and watch it instantly, and with absolute confidence, spit back a plausible yet entirely fabricated answer. It was during this period that the word "hallucination" firmly entered the modern lexicon to describe the tendency of AI systems to invent facts out of whole cloth. Six years and several evolutionary iterations of various chatbots later, that fundamental lack of trust remains. Even though advanced models like Claude or Gemini routinely perform genuinely useful work, the foundational rule of the current technological era has become clear: trust, but verify. And verify you must.
Apply that maxim at scale to maintain the structural integrity of the global scientific record, and one begins to grasp the monumental challenge currently facing publishers of peer-reviewed scholarly research. In a feature titled “I Used AI to Fake a Scientific Paper,” published by IEEE, freelance writer and doctor of applied mathematics Benjamin Skuse explores this exact phenomenon. He details the efforts of the human gatekeepers who are trying to beat back a rising tide of AI slop by attempting to get his own entirely synthetic papers past the stringent defenses maintained by the Institute of Electrical and Electronics Engineers.
As features and AI beat editor Eliza Strickland explained, Skuse was initially surprised at just how effortless it was to generate a paper that looked, sounded, and felt like legitimate science. For one of his experiments, he essentially bypassed any rigorous drafting phase. As Strickland described it, he basically hit “run experiment” and “write paper,” and then he didn’t do a damn thing to it. He simply launched it straight into the maw of the publishing engine to see what would happen.
What Skuse discovered through his investigative experiment was that the combination of IEEE’s rigorous human-review processes and automated research-integrity checker programs are currently managing to keep pace in what has rapidly evolved into a high-stakes cat-and-mouse game. In this ongoing contest, everyday researchers and bad actors alike utilize free, readily available AI tools to generate text, while publishers deploy counter-software to catch them. However, the more advanced large language models are no longer making the same amateurish, easily detectable mistakes that defined the earlier, free versions of these technologies. This leaves scientific and academic publishers locked in a relentless arms race between increasingly plausible synthetic slop and the sophisticated detection systems aimed at stopping it before it corrupts published literature.
Scholarly publishers are far from alone in their daily efforts to hold the line against the encroaching wave of artificial content. They are joined by another monumental guardian of human knowledge: Wikipedia, the global crowdsourced encyclopedia. In an interview for a monthly feature highlighting key digital preservation efforts, Strickland spoke with Ellia Lebleu, a founding member of WikiProject AI Cleanup. This dedicated volunteer initiative is tasked with detecting, flagging, and ultimately deleting AI-generated material from the platform. Lebleu and her team of volunteers find themselves confronting the exact same systemic problem that plagues commercial academic publishers, but they must do so with a fraction of the personnel and significantly fewer automated tools.
Wikipedia has long been recognized worldwide as a premier repository of human-generated information, and its community is fiercely determined to hold on to that distinction, Strickland noted. Yet, looking at the rapid evolution of generative models, observers are left asking how long any human-curated platform will realistically be able to defend those borders.
The implications of this struggle extend far beyond individual websites or publishing databases. If society can no longer collectively assume that the scientific record is fundamentally trustworthy, or that basic, foundational facts about the world remain accurate, our collective understanding of reality begins to unspool into deep uncertainty. Perhaps that kind of pervasive ambiguity delights quantum scientists who study the indeterminate nature of the universe, or the digital grifters who thrive in an information ecosystem where a shared, objective reality ceases to exist. But that very same uncertainty ultimately undermines public trust in the institutional pillars that exist precisely to curate the vast corpus of human knowledge and to promote the safe advancement of science and technology for the benefit of humanity as a whole.
Alternatively, some observers suggest a more radical long-term scenario. In a world where artificial intelligence is increasingly doing the actual science and writing the resulting research papers, it may ultimately cease to matter whether human beings trust the outputs or even understand them. As Strickland remarked, it feels genuinely surreal to contemplate a future where AI systems will be writing exclusively for other AIs, leaving human researchers entirely out of the loop and watching from the sidelines as machines converse with machines about a reality we no longer interpret ourselves.
