Schmidt Sciences Funded AI for Historical Research

The initiative supports 23 research teams applying machine learning to digitize and interpret complex historical archives.

Updated on Sept. 25, 2026 in Artificial Intelligence

Isometric editorial illustration showing a weathered parchment fragment on a stone plinth intersected by a geometric light lattice, representing AI-driven historical research. — Schmidt Sciences has awarded $11 million in grants to 23 research teams using artificial intelligence to analyze and digitize complex historical manuscripts. AI Illustration. Upload story photo >

Live Poll

Should major AI labs treat historical scholarship as essential infrastructure for developing safer AI?

Schmidt Sciences has announced $11 million in new grants to fund 23 research teams using AI to analyze historical manuscripts and records. This funding initiative seeks to integrate computational tools into historical scholarship, building on existing research models like Google DeepMind's Aeneas.

Why it matters

The collaboration aims to provide AI developers with high-quality primary source data, which is necessary to improve reasoning capabilities and mitigate hallucinations in large models. Conversely, historians gain the computational power to process massive archives that remain impractical for manual analysis.

Researchers are utilizing generative models and text restoration software to identify patterns across thousands of records, often comparing fragmented texts against known linguistic networks. Projects like the Aeneas model, released by Google DeepMind in 2025, specifically target the restoration of ancient inscriptions.

The players

Schmidt Sciences

A philanthropic organization that funds scientific research and supports the AI2050 initiative focused on long-term artificial intelligence development.

Google DeepMind

An AI research lab known for developing advanced machine learning models and large-scale computational systems.

Benjamin Breen

A historian and researcher who advocates for increased investment in the intersection of artificial intelligence and historical scholarship.

The details

The research teams utilize generative models—AI systems that can create new content based on learned data patterns—to process primary source documents. By inputting large quantities of digitized historical text, researchers can identify recurring patterns across thousands of records. Software for manuscript reconstruction, such as that developed by Paul Dilley at the University of Iowa, compares weathered or fragmented source material against established networks of known language examples to fill in missing gaps.

Google DeepMind released the Aeneas model in 2025.
Benjamin Breen published an essay on September 24, 2026.
Schmidt Sciences announced the $11 million in research grants during the week of September 25, 2026.

The Tech Race

This effort follows the trajectory set by the AI2050 initiative, which seeks to solve long-term challenges in artificial intelligence development. By formalizing historical research as a core testing ground for model reasoning, these grants directly compete with general-purpose benchmark efforts.

The initial phase of this initiative involves 23 distinct research teams across various geographies and historical eras. While these tools currently focus on academic archives, the resulting models will eventually influence how researchers and the public interact with long-context digital databases.

The takeaway

Historical archives are becoming a primary training ground for improving model long-context reasoning. Observers should track the upcoming results from these 23 teams to see if historical data proves more effective than synthetic data for reducing model hallucinations.