Smithsonian AI Project Connects Thousands of Revolution Artifacts

Smithsonian data scientist Mike Trizna worked at a computer in Washington on Sept. 29, 2026, a July 8, 1809 edition of the Delaware Gazette open on the monitor in front of him. The paper had been printed in Wilmington, Del. Between that printing and the room where he sat, the Smithsonian had been founded, in 1846, and grown into an institution of 21 museums and galleries, the National Zoo among them, with historical holdings of more than 157 million items. The Gazette, a newspaper from the early republic, was present again in the capital. He did not close it. Around that single issue, in other buildings and other catalogs, lay the newspapers, pension files, and museum objects that still held the same years.

Those stacks now fed Revolution Crossroads, a Smithsonian–Library of Congress partnership. Researchers had begun applying AI models in the spring of 2026, drawing on archives that had never shared a single catalog. Chronicling America, the newspaper project of the Library of Congress and the National Endowment for the Humanities, holds more than 13 million pages spanning 1690 to 1963. The team took a subset from 1770 to 1810. Beside it stood the Revolutionary War Pension Files at the National Archives: more than 82,000 files, over 2 million pages of veterans’ claims, service notes, and papers from widows seeking benefits after the war.

Natalie Buda Smith, director of digital strategy at the Library of Congress, measured the stakes in that scale. A person hunting materials might not know whether to begin at the Smithsonian, the Library of Congress, or somewhere else. The partnership was built so scholars and the public could search across both and draw links that once demanded extensive manual work through separate catalogs. The models were already returning those links.

One match tied a craftsman’s 1774 newspaper notice to a creamer held at the National Museum of American History. The silversmith had been running advertisements for a creamer around that year; the museum still held a similar piece. The system did not invent the resemblance. It made the two holdings visible together—a commercial notice from one archive and a silver object from another becoming parts of one trail without a person first knowing either existed. Links of that kind had once required extensive manual searches across multiple catalogs. After the models went to work, the connection arrived as a single, concrete join.

The creamer was not an isolated case. About 10,000 objects had already been identified, centered on people who lived from 1770 to 1810. The team digitized those holdings, and in some cases redigitized them, so the models could work with them. An open Smithsonian Revolutionary Era Collections dataset went onto Hugging Face, a platform for sharing research data: 12,667 records and nearly 4,000 linked images drawn from the National Museum of American History, the National Postal Museum, the Smithsonian American Art Museum, and the National Portrait Gallery.

The Smithsonian said it was not using the current frontier AI models that had drawn recent scrutiny. What the project models returned still went to trained historians.

“What they might say is, ‘actually that is a known nickname for so-and-so’ or ‘that is a generic term that’s used for X, Y, Z,’” said Becky Kobberod, chief digital and innovation officer at the Smithsonian. “That is the nuance that a historian or someone who’s spent their whole career studying that time period is going to know that a machine is not.”

“An ideal scenario is that we get to a place where we’re actually adding to the public record,” Kobberod said. “Maybe they’re not Thomas Jefferson or George Washington, but they did live in a community and build that community, and that name is now known.”

The project was built for those names—people whose lives had faded from the record over the centuries. The ambition was to put them back where the public could find them. Natalie Buda Smith treated a find at one institution as the start of a longer trail. It “illustrates that there’s a richness out there and you might find part of the story at the Smithsonian and part of the story at the library,” she said. “Open up those doors to people so they can understand, because they found it in the Smithsonian doesn’t mean it’s the end of the story.”

A newspaper notice could lead to a museum object. The two institutions had lined their holdings up so that a single match would send a reader forward, not home. The records are available for public search at si.edu/revolution-crossroads/search, putting the project’s linked holdings within reach of scholars and of anyone else who wants them. A reader can move from one institution’s material into another’s without first knowing which catalog holds the next piece.

“It’s not going away,” Kobberod said. “We do believe we need to lean in and figure out the part that this technology plays in our work.”

On that same Sept. 29, 2026, Becky Kobberod, Sara Snyder, and Mike Trizna gathered in Trizna’s office in Washington. Snyder is the Smithsonian’s director of Digital Experience. Trizna brought a copy of the first printed publication of the Declaration of Independence onto the computer screen. The printed page occupied the monitor.