Announcing a New Grant: AI for Cultural and Historical Reasoning
I’m delighted to share that I am part of a team that has been awarded funding through the Schmidt Sciences Virtual Institute for Humanities and AI (HAVI), an ambitious initiative designed to bridge the gap between humanities scholarship and the next generation of artificial intelligence.
Our project, Artificial Intelligence for Cultural and Historical Reasoning, takes aim at a central limitation of today’s AI systems: their difficulty in reasoning across distinct cultural contexts, historical periods, and forms of humanistic knowledge. While large language models have become impressively fluent, they often flatten cultural difference, struggle with historical specificity, and apply a single mode of reasoning where plural perspectives are essential.
What the project does
The first phase of our work focuses on benchmarking. We will develop rigorous evaluation frameworks that measure how well language models capture culturally specific patterns of reasoning and whether they behave in ways that are appropriate to particular historical contexts. These benchmarks are not just diagnostic tools; they are intended to shapehow models are trained, evaluated, and deployed.
By creating metrics that make cultural and historical sensitivity visible—and measurable—we hope to contribute to a broader ecosystem of context-aware, pluralistic AI. A core goal of the project is to ensure that these benchmarks are widely adopted and shared with both AI researchers and humanities scholars.
Once we have confidence in historically grounded language models, the project opens onto a second, equally exciting horizon: using AI to study cultural and intellectual change over time. This will allow us to explore historical questions experimentally—asking “what if” questions about cultural reasoning in ways that have previously been possible only through observational or interpretive methods.
Why this matters for the humanities and for AI
HAVI’s mission aligns closely with the intellectual stakes of this project. Humanities research requires sustained attention to ambiguity, narrative form, metaphor, and historically situated judgment. Yet current AI tools often privilege uniformity and scale at the expense of precisely these qualities.
At the same time, AI development faces persistent challenges in multilingual, multimodal, and culturally diverse settings—the very domains where humanistic scholarship has built deep expertise. By bringing these traditions into conversation, HAVI supports work that advances both fields simultaneously.
From the humanities side, AI-assisted methods promise to expand access to cultural archives and enable more systematic, data-informed inquiry without replacing interpretive judgment. From the AI side, humanistic knowledge offers models of reasoning, evaluation, and excellence that can help move beyond narrow benchmarks toward richer forms of intelligence.
Looking ahead
I’m grateful to be working with such an outstanding interdisciplinary team on a project that treats cultural and historical difference not as noise to be smoothed away, but as a core feature of human intelligence. Many thanks to Schmidt Sciences Virtual Institute for Humanities and AI for supporting this work, and to the broader HAVI community for building a space where humanities scholarship and AI development can genuinely inform one another.
More updates soon as the project gets underway.