Leading mathematicians demand limiting the role of AI in scientific research
- Leading mathematicians demand to limit the role of AI in scientific research
- 25 Fields Medalists signed a letter against AI models
- OpenAI is accused of pressuring scientists and secretly using data
- The company withdrew its sponsorship of the Caltech event after criticism
Scientists fear the loss of open scientific culture
At the beginning of the week, 25 prominent mathematicians, all of whom are Fields Medal winners, published an open letter. In the document, they state that companies developing large language models threaten the traditional process of scientific creativity by trying to outdo each other in solving well-known mathematical problems.
Specific conflict with OpenAI
Professor Tristan Buckmaster from New York University accused OpenAI of pressuring him to exclude a colleague from Anthropic, with whom he had collaborated on an important problem, from the list of co-authors. Buckmaster also expressed concerns that OpenAI might have used materials uploaded by scientists into the Codex system to obtain its own breakthrough proof during lengthy computations.
In response to growing criticism from researchers at the California Institute of Technology (Caltech), OpenAI announced on Thursday that it would withdraw its sponsorship of a planned mathematical event organized by the university.
The letter notes that announcements of new solutions are often made in haste, without proper formatting, without highlighting key methods, and without referencing previous work. This raises serious questions about authorship and potential plagiarism, and also jeopardizes the knowledge transfer chain characteristic of mathematical culture.
Other researchers are increasingly questioning whether OpenAI could have used information about their interactions with Codex when training new models. The real danger is that the open research culture could be replaced by secrecy: a large AI lab, discovering a promising path to a proof, could invest tens of millions of dollars in developing language models and publish the result first, leaving the original researchers in the shadows.
The letter followed the so-called Leiden Declaration, published in June by a working group of researchers. The declaration discussed potential changes in the work of mathematicians related to proofs obtained using large language models, and offered recommendations for researchers, educational institutions, and government bodies.
Thus, the issue goes beyond a single conflict between OpenAI and individual researchers. It concerns the fundamental question of how to maintain openness, transparency, and the collective nature of the scientific process in an era when powerful AI systems can generate results faster and cheaper, but without human oversight, they may remain opaque and inaccessible for further development.
Source: 3DNews
