- Artificial intelligence has revolutionized the capacity to produce content, and this has a direct impact on the production of knowledge.
- Higher volume of knowledge production does not automatically mean higher quality.
- Universities, think tanks, newsrooms, multilateral organizations, and other institutions traditionally responsible for generating knowledge, now face a different question: not how to produce more, but who answers for what gets produced.
Not long ago, I watched José Ignacio Latorre, director of the Centre for Quantum Technologies in Singapore, describe an encounter in his office. A postdoc had spent the weekend in front of a screen. He had paid $250 a month for access to a premium generative AI tool and fed it the research problem he had been working on for months. The machine returned a complete scientific paper: introduction, methodology, conclusions, references. Latorre asked if he had checked every line. "All of it," he said. "Impeccable. Not a single error." His question was no longer whether the machine could do the work. It was: "Do I publish it under my own name?"
That question is now being asked inside every institution whose authority depends on credibility: universities, think tanks, newsrooms, and multilateral organizations like the Inter-American Development Bank (IDB). For decades, their distinctive value was producing analysis that demanded time, money, and expertise.
But now, with the right prompt, an advanced user can produce something that resembles serious research in a weekend with the help of generative AI. Content is no longer scarce. Trust is.
Greater Volume, Declining Quality
That shift is not hypothetical. A few months ago, the AI Task Force for Organization Science analyzed nearly 7,000 manuscripts submitted over a five-year window, treating the two years before ChatGPT’s late 2022 release as a baseline against which to measure change. They reported a 42% surge in submissions and a decline in writing quality.
The papers most likely to have been written with AI were also the most likely to be rejected. The authors most likely to use AI, non-native English speakers and new entrants to the field, did not improve their chances of being published. I belong to the first group, so the findings hit close to home. The Task Force’s diagnosis: more versus better. The system is producing more research, but less of it can be trusted.
That is dangerous because trust is asymmetric. AI produces content that is plausible but not always reliable. An institution can publish hundreds of credible reports, but then one report with a hallucinated source is what the reader remembers. Credibility takes years to build and seconds to lose.
The Importance of Good Research for Public Policy
Working in knowledge management at a multilateral development bank, I have seen research become raw material for public policy in Latin America and the Caribbean. What decision makers need has changed. For a minister in Lima, Bogotá, or Montevideo, the challenge is no longer finding a paper on a topic. It is deciding which of the many papers found online can be trusted. Trust is where we now have to invest, if we want our work to remain relevant.
Knowledge is a cornerstone of economic development because it helps countries make better decisions, strengthen institutions, increase productivity, foster innovation, and ensure that scarce resources are used effectively.
For the IDB, knowledge is a strategic asset and a core source of value, enabling the Bank to generate evidence, inform public policy, design better projects, support innovation, and help countries address both emerging and long-standing development challenges across Latin America and the Caribbean.
But as artificial intelligence transforms how knowledge is produced and consumed, organizations whose value rests on expertise face a critical question: how can they maintain trust and relevance in an era of increasingly capable machines?
Three Recommendations to Build Trust in Knowledge-Based Institutions
This question brings me back to Latorre's office. After the postdoc left, Latorre tested the same AI subscription on a problem his team had been stuck on. Three hours later, after processing data, the answer came back: it would not work. He gathered his researchers. "The world divides into those who turn their heads away," he told them, "and those who look the problem in the eye." To face the situation, he proposed two rules and raised an open question:
- Rule 1. Require an AI check for every paper.
- Rule 2. Provide premium AI access for every researcher.
- Question: Would the center accept AI authorship? (a question that every knowledge institution should be analyzing).
Considering those points, I believe three shifts follow:
First, treat knowledge production as a process of verification, not as a race to publish. The value of an institutional publication is no longer that it exists; but that someone is accountable for its accuracy. That means making sources traceable and verifiable, so a reader can distinguish a real expert from a confident chatbot.
Second, slow down to speed up. Speed alone is not the goal. The goal is to move at the pace at which good work can be done and trusted. Every shortcut in research, citation, or review weakens the institution's reputation.
Third, fix the incentives. The Organization Science AI Task Force is clear about what drives the flood of AI papers: tenure rules and bonuses that reward how much you publish, not whether anyone trusts your work. The same trap exists in every institution that emphasizes outputs. Reward volume, and AI will deliver volume.
Challenges Facing Latin America and the Caribbean
Of course, many countries still face a real deficit of rigorous, locally grounded evidence. Latin America invests 0.57% of its GDP in research and has roughly 668 researchers per million inhabitants, against 2.55% and 4,358 in Europe and North America, according to 2023 data from UNESCO.
That gap is real, and governments cannot close it alone. Institutions like the IDB help close it, producing research that in 2025 was cited by more than 1,500 policy documents worldwide.
However, in a region where such research is scarce, the cost of a single unreliable publication is higher, not lower. When there are few credible voices, each one carries more weight. Responding to scarcity with unvetted volume is not progress. It is a category error.
Latorre closed his talk with one line worth remembering. Everyone wants to live through a defining moment in history, he said. We are living one. The question for institutions is no longer how much they produce. It is whether anyone will still trust what they publish.
In an era where AI seems to be revolutionizing knowledge production, it is more important than ever to ignite conversations with the aim of understanding the depth and value of the human factor.