The 2026 Commonwealth Short Story Prize-winning work, "A Snake in the Forest," features a character named "Aunt Marsha." She is a large woman whose laughter is loud enough to shake dust from the rafters, yet whose voice is gentle enough to coax a child down from a railing. The setup reads like exaggerated, unrefined wordplay. The author of this story seems never to have communicated with a single human being in the real world. Perhaps that is in fact the case.
The author's name is "Jameer Nazir," but the credible AI detection program "Pangram" judged the story to be 100% AI-written. The work contains metaphors that seem logical yet are somehow awkward. The Commonwealth Foundation, which runs the competition, did not cancel the announcement of the winning work. Regarding this, journalist Kelsey Piper analyzed: "The era when a literary magazine could be run on trust alone is over," adding that "the choice remains of whether to accept submissions of AI-generated works or to filter them out."
Piper's diagnosis is accurate. Many writers and readers would think the answer is obvious. If Pangram's judgment is correct, the Commonwealth judges rejected a human work and chose so-called "AI Slop." A recent blind test conducted by The New York Times also found that numerous readers preferred AI-written summaries over the writing of award-winning, established authors. This shows that countless AI critics have misidentified the essence of the problem. If the spread of AI writing is a problem, it is not because it is garbage too poor for humans to read. Rather, in some respects, it is a problem precisely because it is so excellent. AI works are like "fast food." They are convenient and cheap, maintain a certain level of quality, and are optimized to stimulate the human pleasure center.
Of course, AI often displays bizarre metaphors or hallucination phenomena that differ from fact. This is because it learns only from text data without real-world context. On the other hand, in terms of style and structure, it excels at optimizing readability. Novels written by AI tend to be generally smoother, more compact, and psychologically simpler than human novels. They clearly reveal their themes and build linear narratives without complex foreshadowing, while excluding morally ambiguous characters.
These characteristics fit the novel judged to be AI-authored. This is what the public expects from fiction, and the same goes for those who pride themselves on having high literary discernment. These characteristics of AI were not learned by chance but are the result of reflecting the preferences of human readers. If so, what exactly is the problem with AI literature that provides exactly what readers want? Just like fast food, what readers want now is not beneficial in the long term. It would not be desirable if most of our future reading material came from only a handful of AI models. This is because it could lead to a kind of "Intellectual Monocropping."
The Irish Great Famine that occurred in 1845 shows how dangerous monocropping is in agriculture. At the time, Ireland, which depended on only a single variety of potato, suffered the agony of more than 1 million people starving to death and 2 million leaving their homeland due to potato blight. This tells us that it is preferable to grow diverse crops rather than rely on a single high-productivity variety. Likewise, it is desirable for human thought to emerge broadly from many sources. Relying on only one "big thing" creates a single vulnerability where failure can erupt all at once. Intellectual monocropping does not apply only to AI. The recent phenomenon in our society of rapidly tilting toward a particular political bias is the same. Even amid totalitarian currents, humanity has always maintained original thinkers and healthy diversity. If we entrust a significant portion of intellectual labor to machines, we will lose this precious bastion of thought. This is because machines are entities that have been trained through human texts and adjusted to fit the narrow framework of legal regulations and cultural norms.
This is precisely the problem with AI fiction. By generating sentences as if taking a weighted average of past human writing without personal experience, one can easily obtain writing of readable quality. However, it is difficult to obtain sentences that readers will long remember. That is why a world in which AI takes over not only novels and columns but even doctoral dissertations is worrying. The reason we need human writers is not because AI lacks bizarre metaphors. It is because AI lacks "Wild Ideas."
