Who becomes replaceable? Generative AI and the discursive stratification of work

When we hear that artificial intelligence is changing the world of work, the conversation often feels like a technical audit. We ask which jobs can be done by machines and which cannot, treating the answer as a fixed property of the job itself. But there is another layer to this story, one that happens not in code or factories, but in the news and public conversation. Before a machine ever replaces a human, people must first agree that the replacement is possible, necessary, or even desirable. This agreement is built through language. When news outlets describe a job as "at risk," they are not just predicting the future; they are actively sorting workers into categories of those who will be kept and those who will be discarded. This process of sorting through words is what researchers call discursive stratification. It is a way of understanding how society decides who is vulnerable to change and who is safe, long before any actual layoffs occur.

A new study by Hana Kim, a researcher at the Korea Advanced Institute of Science and Technology, investigates exactly how this sorting happens in the age of generative AI. The research focuses on a specific turning point: the release of ChatGPT in late 2022. The author wanted to know if the way news media talked about AI and jobs changed after this moment, and more importantly, whether that change treated different types of workers within the same profession differently. The study does not look at whether people actually lost their jobs. Instead, it examines the text of nearly half a million news articles to see how the public narrative shifted. The core question is simple: when the news says a job is being replaced by AI, who is the "who" in that sentence?

To answer this, the researcher gathered a massive collection of news passages from 2010 to 2026, focusing on mentions of specific occupations like lawyers, doctors, writers, and developers. The team used a computer program, trained by human experts, to read these passages and sort them into four distinct categories. The first category was neutral, where the job was mentioned but not linked to AI. The second was augmentation, where AI was described as a tool that helps the worker do their job better. The third was task automation, where specific parts of the work were said to be taken over by a machine, but the job itself remained. The fourth and most severe category was replacement, where the text suggested the job itself was becoming unnecessary or that the worker would be removed.

The analysis revealed that after ChatGPT arrived, the total amount of news coverage discussing AI's impact on jobs did go up. However, the way this increase happened was not uniform. The rise in "replacement" stories was not because every single job suddenly sounded more dangerous. Instead, the news began to focus its attention on a different set of occupations than before. The coverage shifted toward jobs that were already being discussed as risky, such as writers and coders, while paying less relative attention to others. This means the overall feeling of danger in the news was driven by a change in which jobs were being spotlighted, rather than a sudden hardening of the narrative for every profession.

The most significant finding, however, lies not in which jobs were discussed, but in how the people within those jobs were described. The study looked closely at the difference between junior workers—those just starting their careers, often called entry-level or trainees—and senior workers, who are experienced leaders or principals. The results showed a sharp and growing divide. After 2022, news articles became much more likely to mention junior workers when discussing the impact of AI. In fact, the gap between how often junior and senior workers were mentioned as targets of AI change widened by twenty percentage points.

But the difference went deeper than just how often they were mentioned. When a junior worker was mentioned in the context of AI, the story was overwhelmingly likely to frame them as replaceable. The analysis found that among the news stories that did talk about junior workers, nearly half described their roles as being eliminated entirely, and very few described them as being helped by AI. In contrast, when senior workers were mentioned in the same period, the narrative was almost the opposite. They were far more likely to be described as being augmented, with AI serving as a tool to enhance their judgment and productivity. The news rarely suggested that senior workers would be removed; instead, it suggested they would be the ones wielding the new technology.

This pattern held true even when the researchers tried to rule out other explanations. They checked to see if the difference was simply because junior jobs are more routine or repetitive, which is a common reason jobs get automated. The data showed that while routine work was indeed more likely to be discussed as automated, the specific gap between junior and senior workers was a separate phenomenon. It was not about the type of task, but about the stage of the career. The researchers also tested whether the computer program was just reacting to the words "junior" or "senior" as a shortcut. They ran the analysis again after hiding those specific words, and the pattern remained exactly the same. The computer was responding to the context of the story, not just the labels.

To be absolutely certain, the researchers brought in human experts to read a smaller, carefully matched set of stories. They paired a news story about a junior worker with a story about a senior worker in the same profession and the same year, then asked the humans to classify them without knowing which was which. The human readers confirmed the pattern with striking clarity. In the vast majority of cases where the two stories differed, the one about the junior worker was classified as being about replacement, while the one about the senior worker was classified as being about augmentation. The human agreement was so strong that the statistical chance of this happening by accident was virtually zero.

The study suggests that the arrival of generative AI has not just changed the technical landscape of work, but has reshaped the social map of who is considered essential and who is considered expendable. The narrative has shifted to treat the entry-level positions in many professions as the primary targets for automation, while preserving the image of the experienced professional as the master of the new tools. This is a critical distinction because entry-level jobs are not just low-cost labor; they are the training ground where new professionals learn the skills, judgment, and habits needed to become experts. If the public discourse treats these starting roles as disposable, it risks cutting off the very path that allows people to become the senior workers of the future.

The research does not claim that these news stories are the direct cause of job losses, nor does it say that employers are acting on these stories immediately. Instead, it identifies a powerful shift in how society is imagining the future of work. By consistently framing junior roles as replaceable and senior roles as augmentable, the public conversation is creating a hierarchy of vulnerability. It is deciding, through words, that the future of AI belongs to the experienced, while the novice is the one who must be replaced. This discursive stratification happens before the economic changes take place, setting the stage for a future where the opportunities to learn and grow are narrowed for those at the bottom of the career ladder, while those at the top are given new powers. The study concludes that understanding AI's impact on work requires looking not just at what machines can do, but at who we are willing to imagine as replaceable.