The New Orleans Jazz Museum archives trace the history of the genre all the way back to the first jazz song ever recorded for commercial release: “The Livery Stable Blues.”
That soundtrack is well-preserved – the medley of trombone and piano comes across clear to the delight of jazz historians and enthusiasts alike. But not every century-old artifact has fared so well. Newspaper clippings have creased, musical scores have smudged and old records have warped over time.
“[Archives] are crumbling and disappearing and we are losing our history,” said Katharine Elkins, humanities professor at Kenyon College.
Deteriorating archives are endangering historical documents across the country. A group of students and faculty at Kenyon College, in rural north central Ohio, believes artificial intelligence could be the key to rescuing them.
Kenyon is one of 23 research teams being funded by the Schmidt Sciences’ Humanities and AI Virtual Institute (HAVI) to advance humanities research. Their goal is to build an open AI system that can save endangered archives in small and underrepresented communities, like the collection at the New Orleans Jazz Museum.
How AI can preserve history
The group’s first meeting in January felt more like a startup than a lecture hall. Instructor Jon Chun, one of the leaders of the AI project, furiously filled up a white board as questions flew and ideas bounced around the room.
Many small museums and newspapers can’t afford the professional preservation equipment that ensures the archives survive, Elkins said. So, the team is mapping out an 18-month plan to create a handheld tool that would use artificial intelligence to digitize and restore fragile artifacts with nothing more than a smartphone photo.
The team chose New Orleans jazz history as the test case for building their tool, in part, because it’s so complex. There’s artifacts in all mediums: soundtracks, photos and videos. Some documents are in Creole and Cajun French, languages artificial intelligence models don't yet have much digital data to train on.
“Making our AI systems represent the full range of languages and communities and historical pasts is opening up these archives,” she said.
New potential for discoveries
That representation is especially important, Elkins said, because the tool won’t just stop at restoring dilapidated sheet music and jazz club posters. It also will allow AI to find connections across archives.
“In the past, archives would preserve each of these separately. You'd have the digital, the sound and the video,” she explained. “Now we have these AI models that can deal with all of that together. So it's a complete game-changer.”
In Elkins’ vision, what used to take researchers years to comb through would now take minutes. They could trace how a melody traveled from a handwritten score to a recording years later with just a couple keystrokes.
Kenyon senior and neuroscience major Adrian Mangene sees possibilities for museums all over the country.
“There's always been such a divide between STEM and the humanities. I think bridging that gap and thinking about these problems from so many different angles is one, now possible because of AI; and two, revolutionary,” he said.
A human-centered approach
The AI tool wouldn’t take humans out of the humanities entirely. Project assistant and recent Kenyon alumnus Hannah Sussman said input from jazz experts, for example, would constantly check the program’s work.
These so-called “humans-in-the-loop” determine the accuracy of the system’s conclusions and work to catch any hallucinations or errors the AI might introduce.
“It would be impossible to have AI just analyze it and just take that as kind of the truth,” Sussman said.
The tool simply speeds up the tedious parts of archival work, she said, so that social researchers like herself can dig in deeper into collections across the country.
Artificial intelligence would make the connections, but humans would still get to decide what they mean.
“When you have students who are trained in the humanities and who are thinking critically about both the data itself and then also the social implications of the data, how it's being used, the ethical components, that really allows AI to be used in the best possible way,” she said.
