Could the next great novel be written by AI? Experts are skeptical
Linguists and novelists debate whether AI can produce literature comparable to human writing, highlighting the difficulty of distinguishing machine from human language, and the limits of AI in crafting compelling narratives.
Difficulty distinguishing human and AI writing
According to Claire Hardaker, professor of forensic linguistics at Lancaster University, humans correctly identify AI-generated text only about 60% of the time. Her online test "Bot or Not" presents users with a series of reviews to spot fakes. While many rely on simple heuristics like clichés, dashes, and the "rule of three," these features are also common in human writing, appearing even in the works of Charles Dickens.
Literary world shaken by AI allegations
Accusations of AI use have haunted authors. Hachette withdrew Shy Girl, a debut horror novel, after rumors that the author used AI, which he denies. Steven Rosenbaum's "The Future of Truth" contained fabricated quotes, which the author acknowledged. Media outlets, including the Guardian, increasingly receive complaints about suspected AI-generated text, though such suspicions often stem from subjective impressions.
Linguistic patterns and AI's influence on human language
Researchers have noted AI's overuse of words like "delve," "showcase," "boast," and "underscore." Interestingly, the rise of "delve" may be due to human reinforcement learning feedback workers associating it with quality. AI also prefers attributive adjectives over predicative ones and uses fewer pronouns. Different models have distinct "dialects": Gemini often says "here's a breakdown," while Deepseek responds with "Certainly!" AI tends to standardize language towards an Anglo-American norm, a process called "cultural ghosting."
Studies show AI language has leaked into human speech: after ChatGPT's release, "delve" and "boast" increased in unscripted conversations. However, after social media scrutiny, "delve's" frequency in academic abstracts dropped.
Novelists weigh in on AI literature
Gary Shteyngart, a novelist and professor at Columbia University, recounted his students' strong negative reaction to a peer using AI. He believes literary fiction relies on a tacit contract between human author and reader, which AI breaks. Professor Peter Stockwell of the University of Nottingham argues AI excels at lower language levels (words, syntax) but fails at higher levels like narrative structure and emotional depth. Great novels require human embodied experience and social nature, which AI cannot replicate.
