Credential farming in the age of generative AI: relocating trust from outputs to communities
This paper argues that generative AI has undermined the reliability of traditional academic credentials as proxies for competence, necessitating a shift in epistemic trust from polished outputs to the relational communities and formative processes that genuinely cultivate and witness scholarly judgment.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
For decades, the scientific world has operated on a simple, trusted assumption: if you can produce a polished research paper, you must understand the work inside it. This idea relies on credentials—degrees, published articles, and job titles—which act as shorthand for a person's ability to think critically and contribute to shared knowledge. These documents were never perfect measures of a person's mind, but they were reliable enough because creating them required genuine effort and deep learning. However, the rise of generative artificial intelligence has begun to break this link. Now, a person can use a computer program to write a flawless research article, complete with data analysis and citations, without actually knowing what the words mean or how the conclusions were reached. This shift creates a dangerous gap where the document looks perfect, but the person behind it may have learned nothing.
This new reality threatens the entire system of how we select scientists and doctors. If institutions continue to judge people solely by the papers they submit, they risk hiring individuals who have mastered the art of asking a computer for answers rather than the skill of finding them. A group of researchers, including students and professors from universities across the globe, has argued that the solution is not to build better tools to catch people using AI, but to change where we place our trust. They suggest that instead of looking at the final product, we must look at the process of how a person learns and thinks within a community.
The authors of this paper, a diverse team ranging from high school students to medical researchers, argue that the traditional way of evaluating competence is collapsing. They describe a phenomenon they call "credential farming," where individuals extract the social value of a degree or a publication without actually contributing to the scientific enterprise that gives those titles meaning. In the past, producing a research paper was proof that someone had spent years mastering a subject. Today, generative AI allows anyone to generate that same proof instantly, regardless of their actual knowledge. The authors point out that trying to fix this with detection software is a losing battle. These tools are often unreliable, can unfairly penalize writers who do not speak English as their first language, and defend a boundary that no longer separates the knowledgeable from the unknowledgeable.
Instead of fighting a war of detection, the researchers propose moving the focus from the output to the formation of the scientist. They argue that true competence is not stored in a document but is built through a long, witnessed process of learning within a group. This concept draws on the idea that we learn by doing work alongside others who can see our mistakes and guide our thinking. In a healthy scientific community, a member learns not just by reading facts, but by engaging in conversations where they must explain their reasoning, admit when they are wrong, and revise their ideas based on feedback from peers. This kind of learning is difficult to fake because it happens over time and in relationship with others, whereas a computer can instantly generate a perfect-looking essay that reveals nothing about the writer's mind.
To test this idea, the authors describe three practical steps that research communities can take to rebuild trust. First, they suggest selecting new members based on the depth of their thinking rather than the clarity of their resume. This involves asking open-ended questions that have no single correct answer, such as asking a candidate to describe a time when a new idea unsettled their worldview. A computer can generate a plausible answer to such a question, but it cannot replicate the genuine hesitation, reflection, and personal history that a real human brings to the conversation. Second, they recommend building small, informal groups where people can talk freely without the pressure of producing a final report. These small circles allow for the kind of honest exchange where a person can say, "I do not understand this," and receive help, mirroring the way learning happens in real life. Finally, the authors urge these communities to resist the urge to turn these new methods into new metrics. If they start counting how many times someone speaks or how quickly they reply, they will simply recreate the old problem of optimizing for a number rather than for genuine understanding.
The paper highlights that this shift is especially critical for researchers in low-resource settings, who often learn through informal networks that are invisible to traditional institutions. These individuals may have deep curiosity and real skills but lack the polished vocabulary to present themselves in the way elite universities expect. By focusing on formation and community relationships, institutions can find these talented people who are currently overlooked. The authors acknowledge that this approach is not a perfect solution and carries its own risks, such as the potential for bias in subjective conversations. They admit that character is hard to observe and that their own team includes students who are still learning, which means they cannot fully certify their own formation from the inside.
Ultimately, the paper concludes that machines can write papers, but they cannot become people worth trusting. The task for scientific institutions is no longer to verify that a document exists, but to determine if a person has truly learned to think. This requires a move away from judging the final product and toward witnessing the journey of learning itself. It is a call to rebuild the system so that it values the messy, difficult, and human process of becoming a scientist, rather than just the shiny, manufactured result.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.