OVERVIEW: We will discuss how to harness the potential of generative AI to support exploration and discovery in archival record collections. Our session will report on our progress on an IMLS-funded project called GenAI-4-Arch where we partner with the Maryland State Archives (MSA) and focus on a subset of the Legacy of Slavery “Domestic Traffic Ads” (DTA) collection. The DTA collection encompasses newspaper advertisements from 1824 to 1864, which shed light on the abhorrent practices of chattel slavery—where buyers and sellers engaged in transactions involving human beings.
We use Large Language Model (LLM) interfaces to interact with records conversationally. The use of LLMs can carry significant disadvantages:
- the potential to transmit sensitive and potentially offensive content back to the models with the risk of it being incorporated into their training data
- the potential for falsehoods and distortions (aka hallucinations) to be generated
To mitigate these shortcomings, our project actively engages with collection content experts at the MSA in Annapolis, Maryland, and a focus group of local cultural and historical experts and community members at the Kennard African American Cultural Heritage Center in Centreville, Queen Anne’s County, Maryland (located on the Eastern Shore, where many of the DTA Collection ads originate).
We also use non-AI tools to extract some ground truth from the DTA collections. We compare results from multiple LLMs. We ask the LLMs the same questions multiple times. All of this is done in an effort to further measure and mitigate the risks associated with using generative AI with archival records.
In this session we will demonstrate how to:
- Make archival records AI-ready
- Tune LLMs to incorporate community feedback
- Evaluate the resulting outputs based on the propensity of LLMs to hallucinate
GARA CERTIFICATE CORE COMPETENCY AREA: "Record Considerations for Emerging Technologies"
TARGET AUDIENCE: Federal, Tribal, State, Local, University
FOCUS AREAS: Archives, Records, Technology
PRESENTERS: Richard Marciano, Director, Advanced Information Collaboratory, Rajesh Kumar Gnanasekaran, Asst. Director of AI Solutions, Division of IT - University of Maryland, and Christopher Haley, Director, Study of the Legacy of Slavery in Maryland, Maryland State Archives