Many language models (LLMS) have easily entered the scene of historical research. Their ability to process, annotate and make the texts changing scholar workflows. Although historians are unique to ask a deep question – owning tools that shape our understanding of the past?
The most powerful LLMS is now developed by private companies. While their investments are significant, their goals – focused on profit, platform growth or intellectual property control – rarely align with the values of historical scholarships: transparency, reproducibility, accessibility and cultural diversity.
It raises serious concerns with a) Opacity: We often have no visibility of training data and accessers, especially for non-residential resources: including less contacts, especially.
It’s time to build public, open access to people – trained in chanrated, multilingual, mimic history of money, museums. These models must be transparent, responsible for academic communities and supported by public funds. Construction of such infrastructure is difficult but important. As we do not outsource national archives or school curriculums to private companies, we cannot hand it to our strongest translation technologies.
People have a responsibility – and an opportunity – to create culture learned, academic basic artificial intelligence. We don’t just use LLMs responsible but they are responsible as well. The scholar’s integrity and future knowledge of public can depend on it.
Prof Prr Matteo Valleriani
Max Planck Institute for History in Science, Berlin, Germany
