The fog settled over the city like a slow, deliberate lie. It was the kind of morning where the skyline dissolved into gray nothing, and even the familiar shapes of buildings seemed to waver, uncertain of their own edges. For a writer named Elena, staring out her apartment window, the fog felt fitting. She had just received an email from a publisher she had admired for years. The feedback was polite, professional, and devastating: ‘Your manuscript is well-structured and grammatically flawless, but we have concerns about its authenticity. It reads, frankly, as if it were generated by an AI.”
Elena had never used an AI writing tool in her life. She had spent two years researching, drafting, and revising that manuscript. She had interviewed experts in three countries, transcribed hours of recordings, and rewritten entire chapters by hand. And now, the very qualities she had been taught to cultivate — clarity, precision, polish — had become evidence against her. The fog outside seemed to seep into her thoughts. How do you prove you are human when being human is no longer the default assumption?
This is the paradox that has quietly taken root in the age of large language models. The machines have learned to write so well that good writing itself has become suspect. And in response, a strange inversion has occurred: error has become a credential. A typo, a grammatical slip, a slightly awkward phrase — these are no longer signs of carelessness. They are proof of a human hand. The defect has become the signal of authenticity. And that signal is already being weaponized.
The journey of this idea begins not in a Silicon Valley boardroom but in a small private college in New England, where a student named Allie Abusamra was reading novels and writing essays about them She graduated with a degree in English, a credential that for decades had been treated by the tech industry as a polite irrelevance. The dominant narrative was clear: STEM degrees led to high-paying jobs in technology; liberal arts degrees led to coffee shops and debt. But Abusamra took an unconventional path. She did not learn to code. She did not pivot to a computer science boot camp. Instead, she brought to the table exactly what her English degree trained her to do: read carefully, write precisely, and understand how language actually works.
That turned out to be exactly what Google needed. Abusamra landed a role at the center of the company’s AI efforts, working on projects where her responsibilities involved crafting, editing, and evaluating the language that large language models produce. She assessed whether AI-generated text read naturally, whether it conveyed meaning accurately, whether it captured the right tone. This was not a peripheral task. It sat at the core of what makes products like Gemini useful or useless to hundreds of millions of users. And it required a kind of expertise that no amount of Python proficiency could replicate — the ability to parse ambiguity, detect subtle failures in reasoning, and understand what humans actually mean when they string words together.
Her English degree became an edge rather than a liability. The close reading she had practiced on novels and poetry translated directly into the close reading of AI outputs. She was trained to notice when something was off — a misplaced emphasis, a logical gap disguised by fluent prose, a confident assertion that did not hold up under scrutiny. Large language models are exceptionally good at sounding right. Identifying when they are wrong requires a different skill set entirely. This is the hidden labor that makes AI products functional: the human evaluation of machine-generated text, the calibration of tone, the detection of subtle errors that would otherwise erode user trust.
But here is where the journey takes a darker turn. The same skills that make Abusamra valuable to Google are the skills that have become suspect in the wider world. The same fluency that she works to instill in AI systems is the fluency that gets human writers accused of being machines. The tools she helps improve are the tools that make it harder for other humans to prove their own authorship. The very system she is building is the system that casts doubt on her own kind.
This tension has been building for years, but it has only recently reached a critical point. Studies have shown that neither humans nor AI systems can reliably distinguish between human- and machine-generated writing. [1] When human- and AI-generated text is intermixed, performance becomes even worse. The result is a climate of uncertainty in which false positives — wrongly accusing someone of using AI tools — have become a serious concern. Many universities that had been using plagiarism-detection tools for AI detection have stopped using them due to concerns about reliability. [1] The tools designed to catch cheating cannot be trusted to tell the truth.
In this vacuum of certainty, some writers have reached for the only signal still available to them: the aptly named human error. A repeated word, a small grammatical slip, a slightly clunky phrase — these have started to function less as signs of carelessness and more as proof of a genuine human hand. The defect has become the credential. And errors are already being deployed strategically in competitive contexts — university submissions, job applications, professional correspondence. Recruiters have begun advising applicants to leave a single deliberate typo in a cover letter, precisely to signal that an interested human wrote it.
The irony here is almost too sharp to bear. Alan Turing, the father of modern computing, suggested in the 1950s that machines could appear more convincingly human by sprinkling in a few deliberate typographical errors. [1] He was addressing that advice to machines. Seventy years later, humans are following the same playbook. The machines have learned to write perfectly, so humans must learn to write imperfectly. The defect has become the credential. But this is not a stable equilibrium.
The currency of the error signal is on borrowed time. Once imperfection becomes a recognized sign of authenticity, it immediately becomes available for imitation. Users will ask AI systems to sound rougher, less polished, and more human. The systems will comply and soon become adept at performing calibrated incompetence. The machines will learn to make mistakes on purpose. And then the arms race will begin again: humans will need to find a new signal, and the machines will learn to mimic that too. The defect will lose its value.
This is where the journey reaches its most troubling insight. The problem is not that AI can write well. The problem is that we have lost the ability to trust the evidence of our own reading. We have outsourced our judgment of authenticity to tools that cannot reliably provide it. And in doing so, we have created a world in which the very act of writing well has become a source of moral suspicion.
Consider what this means for a young writer like Elena, whose manuscript was rejected not because it was bad, but because it was too good. She is not alone. Across universities, publishing houses, and hiring departments, the same dynamic is playing out. Writing well, once a mark of skill, has become a source of suspicion. The skills we once used to signal intelligence and effort — clarity, precision, a well-turned sentence — are starting to lose their meaning. The signal has been drowned out by noise.
The path ahead is unclear. Perhaps some situations will demand more direct proof of authorship without the assistance of AI: face-to-face, unmediated assessments, handwritten submissions, and real-time explanations. Some universities have already begun allowing students to use AI in exams, so long as they submit their prompts as part of the assessment. The logic is that transparency can substitute for authenticity. But this approach has its own problems. If every piece of writing must be accompanied by a record of its creation, then the act of writing becomes a bureaucratic process, subject to audit and verification. The spontaneity and intimacy of language — the qualities that make it feel human — may be lost.
Others argue that the decisive skill will simply be knowing how to use AI tools well. In this view, the ability to craft effective prompts, to evaluate AI outputs, and to combine machine-generated text with human judgment will become the new literacy. The English major who becomes Google’s secret weapon is the model for this future. But this vision has a dark side too. It assumes that the value of human writing lies primarily in its utility to AI systems. It reduces the writer to a trainer, a calibrator, a quality assurance worker for machines. The act of writing becomes a form of labor that serves the system, rather than an expression of the self.
There is a deeper question here, one that the fog of technological change has obscured. What do we actually mean when we say a piece of writing is authentic? Is it the trace of a particular human hand, with all its quirks and imperfections? Or is it something more intangible — the weight of experience, the texture of thought, the sense that another mind is reaching out to communicate? If the latter, then perhaps authenticity cannot be faked by machines, no matter how good they get at simulating it. But if the former, then we are in trouble. Because the machines are getting very good at simulating imperfection.
The human cost of inaction on this front is already visible. Writers like Elena are being penalized for their skill. Students are being accused of cheating when they have done nothing wrong. Hiring managers are rejecting candidates whose cover letters are too polished. The suspicion has become a tax on competence. And the burden falls disproportionately on those who write well: the careful, the precise, the fluent. The very qualities we once rewarded have become liabilities.
If we ignore this finding — if we continue to rely on unreliable detection tools, if we continue to treat fluency as evidence of fraud, if we continue to demand that humans prove their humanity by making mistakes — then we will create a world in which the best writers are driven out of the profession. The manuscript that is too good will not be published. The cover letter that is too polished will be ignored. The essay that is too clear will be flagged. We will have trained ourselves to prefer mediocrity, to reward error, to distrust excellence. And the machines will have won not by surpassing us, but by making us afraid of our own best work.
The fog that morning was thick, but it was not permanent. It would burn off by noon, and the city would reappear, sharp and clear. But the fog of suspicion that has settled over writing may not lift so easily. Elena closed her laptop and walked away from the window. She had a decision to make: learn to write worse, to sprinkle in errors, to signal her humanity through imperfection, or refuse the game entirely and hope that somewhere, someone still valued writing that was good for its own sake. The machines were learning to write like humans. The humans were learning to write like machines. And somewhere in the middle, the truth about what it means to be human was getting lost.
Sources
1. Google