It's Not Enough to Just Remember: Using Generative AI to Scale the Atlanta Student Movement Archives for Public Use
Disciplines
Artificial Intelligence and Robotics | Databases and Information Systems | Digital Humanities | Oral History | Social History | Social Justice
Abstract (300 words maximum)
This research examines a pioneering collaboration between first-year student scholars and the KSU Atlanta Student Movement Project, focusing on their work as “red team” trainers for a custom GPT-based archival tool. Tasked with augmenting the digital database of the Atlanta Student Movement (a local chapter in the struggle for Civil Rights) these student-scholars tested and refined a transformative model that allows users to explore historical records in an immersive and interactive manner. During this presentation, attendees will gain insights into the research and training process, from initial data curation to user interface design. The discussion will illustrate how the model was transformed into an academic archive, underscoring the potential of digital archives to amplify underrepresented voices and encourage ethical engagement with the past. First-year scholars will demonstrate red-teaming and discuss surprises/challenges of working with oral histories. The “red team” methodology involved systematically challenging the model’s outputs, enhancing its accuracy, mitigating biases, and ensuring historically faithful responses. By scrutinizing how the GPT model interpreted primary sources (such as protest documents, photographs, and news clippings) the students identified gaps in context and potential misrepresentations of the Movement’s goals. The GPT harnesses the Movement archive to reveal the lived experiences of activists who confronted segregation at lunch counters, schools, and other public facilities. Audiences can engage with this history as both an academic tool and a public platform for civic learning. Through red teaming the custom GPT model, researchers identified faults in its coding that initially misrepresented or oversimplified the lived experiences of activists. Testing revealed biases in how the model processed segregation-era struggles, sometimes failing to capture the depth of resistance in public spaces. These insights allowed researchers to refine the model, ensuring it accurately reflects the complexities of collective action and its role in reshaping civil rights in Atlanta and beyond.
Academic department under which the project should be listed
RCHSS - English
Primary Investigator (PI) Name
Jeanne Law
It's Not Enough to Just Remember: Using Generative AI to Scale the Atlanta Student Movement Archives for Public Use
This research examines a pioneering collaboration between first-year student scholars and the KSU Atlanta Student Movement Project, focusing on their work as “red team” trainers for a custom GPT-based archival tool. Tasked with augmenting the digital database of the Atlanta Student Movement (a local chapter in the struggle for Civil Rights) these student-scholars tested and refined a transformative model that allows users to explore historical records in an immersive and interactive manner. During this presentation, attendees will gain insights into the research and training process, from initial data curation to user interface design. The discussion will illustrate how the model was transformed into an academic archive, underscoring the potential of digital archives to amplify underrepresented voices and encourage ethical engagement with the past. First-year scholars will demonstrate red-teaming and discuss surprises/challenges of working with oral histories. The “red team” methodology involved systematically challenging the model’s outputs, enhancing its accuracy, mitigating biases, and ensuring historically faithful responses. By scrutinizing how the GPT model interpreted primary sources (such as protest documents, photographs, and news clippings) the students identified gaps in context and potential misrepresentations of the Movement’s goals. The GPT harnesses the Movement archive to reveal the lived experiences of activists who confronted segregation at lunch counters, schools, and other public facilities. Audiences can engage with this history as both an academic tool and a public platform for civic learning. Through red teaming the custom GPT model, researchers identified faults in its coding that initially misrepresented or oversimplified the lived experiences of activists. Testing revealed biases in how the model processed segregation-era struggles, sometimes failing to capture the depth of resistance in public spaces. These insights allowed researchers to refine the model, ensuring it accurately reflects the complexities of collective action and its role in reshaping civil rights in Atlanta and beyond.