IMPA is among the 14 organizations selected by OpenAI worldwide
What conditions are necessary for researchers to use artificial intelligence (AI) in an efficient, reliable, and sustainable way to conduct mathematical research? To answer this question, OpenAI, the developer of ChatGPT, selected the PI Center (IMPA Center for Projects and Innovation) to participate in an international initiative aimed at investigating how AI can transform scientific research and expand opportunities.
“AI has proven to be a clear catalyst for scientific research. Research institutions such as IMPA need to understand the costs and infrastructure required to adopt these tools safely and quickly, and to remain resilient as these models evolve,” explained Marcelo Viana, director general of IMPA.
The project “Building Resilient Scientific Research in the Age of Intelligence” will investigate what is needed for mathematics researchers to incorporate this new reality into their research. Over the course of six months, the team, led by project scientist Daniel Yukimura, will work directly with mathematicians at IMPA to assess factors such as infrastructure, computational capacity, engineering support, training, governance, and the costs involved in using artificial intelligence in research.
In addition to Yukimura, the project team includes project scientist Francisco Ganacim, IMPA researcher João Pedro Ramos, and postdoctoral researcher Cynthia Bortolotto. The initiative is part of a group of 14 independent projects selected by OpenAI from different regions around the world. The call for proposals received more than 400 submissions from individuals and organizations interested in investigating the impacts and opportunities brought about by advances in artificial intelligence.
OpenAI’s support includes funding for the project’s development and API credits that will allow researchers to use artificial intelligence models during the study. The initiative will also help estimate how much it would cost for an institution to maintain this type of access on an ongoing basis and what infrastructure would be needed to make it feasible.
Rather than simply assessing access to these tools, the proposal seeks to understand the conditions necessary for them to be incorporated into research in a sustainable manner. According to Yukimura, there is a risk that advances in AI could exacerbate existing inequalities among scientific institutions.
“There is a risk that universities with more resources will be able to conduct research at a faster pace and on a larger scale, while institutions with fewer resources will fall even further behind. The idea is to understand how much it costs to use AI efficiently and what infrastructure is needed to ensure that this technology does not exacerbate existing inequality,” he explained.
Based on IMPA’s own experience, the project aims to develop a model that can be replicated by other institutions interested in assessing the costs and infrastructure required to incorporate AI into scientific research. During the project, researchers will use artificial intelligence tools, and the team will monitor this process to identify needs and improve the supporting infrastructure.
Ultimately, a report will be produced detailing the study’s findings, including an analysis of the costs associated with the use of AI and the infrastructure needed to support researchers. The initiative also provides guidelines for the responsible use of artificial intelligence in mathematics and a discussion on the right of access to AI and on distributed models of scientific discovery.
The Mathematical Community Faces New Questions
In addition to the practical aspects, the project comes at a time of transformation for the mathematical community itself. Technology companies have been investing more and more in the development of systems capable of solving complex mathematical problems, while researchers are beginning to experiment with new forms of interaction between artificial intelligence and scientific research.
“The mathematics community is wondering what lies ahead. A lot of money is being invested in AI for doing mathematics, and at the same time, companies are putting a great deal of effort into demonstrating the progress of these models precisely through the solving of mathematical problems. Part of this project is precisely to try to better understand how mathematical research might evolve in the coming years. This will be important so that IMPA and other institutions can be better prepared for possible changes,” Yukimura reflected.
The analysis ranges from the way mathematicians work to the conditions necessary for different institutions to participate in this transformation. In a scenario where access to advanced models also involves computational and financial costs, understanding these differences can be crucial to preventing the new technology from further concentrating scientific output in institutions with greater access to resources.
In this regard, the project aims to view AI not merely as a tool, but as a transformation that requires new research structures. The Brazilian experience could contribute to a broader debate on how scientific institutions—especially those outside major technology hubs—can prepare for this change.