With data centers springing up nationwide and impacting American neighborhoods, it’s challenging to focus on the positives without the concern that one may soon be nearby. Moreover, as a historian primarily concerned with African diasporic history, the chance for a generative AI model to misunderstand or misinterpret a prompt has made it an unattractive tool to use. Nonetheless, several historians have made convincing arguments for implementing AI cautiously to improve archival practices, create engaging content for students, and perhaps even build stronger media literacy skills as models continue to advance.<\/p>\n
As someone who works in the archives, it is incredibly difficult for me to imagine a Large Language Model (LLM) making some of the decisions I do. Currently, I am working with records from the university\u2019s African and African American Research and Resource Center. While I have biases as a historian and sometimes find it difficult to see the usefulness of every document, I can counter (some) of these biases by considering the user or researcher who might be searching for the very document I might disregard. After surveying and processing, AI has shown that it is a useful tool for refining finding aids and descriptions with strict supervision. As Alex Houston\u2019s \u201cBringing hidden histories to light\u201d establishes, at its best, \u201cAI can surface names that were never indexed, connections that were never documented, and records that were never available simply because no one had the time to describe them. But AI does not correct for bias on its own\u2014it can also replicate and reinforce exclusions already present in archival data. That\u2019s why human intervention is critical.\u201d[i]<\/a><\/p>\n It is more feasible for me to envision using LLMs to transcribe materials. This process, too, would require human oversight, but there aren\u2019t as many expectations for the AI to make the decisions of a trained archivist. Transcribing can be a tedious process that might prevent users from accessing the materials.[ii]<\/a> Furthermore, there can be a learning curve to understanding the handwriting, and AI transcription can be a useful tool for honing these skills.<\/p>\n The rise in youth reliance on ChatGPT and other generative AI models warrants some concern. As I was growing up, Wikipedia was my close ally, much to the chagrin of my teachers. However, I refuse to believe that Wikipedia and ChatGPT are on the same playing field. As Wikipedia rose in prominence, younger readers were encouraged to be skeptical of its content and question everything they find. While I believe we were told this because people copy and paste content from Wikipedia without citing sources (we were in fourth grade), it helped us develop critical thinking skills. Nowadays, young students are turning to ChatGPT, a generative AI model that, despite the empty warnings to question what it regurgitates, presents itself as an omniscient being to their impressionable audience. Rob Nelson\u2019s idea of incorporating AI into classrooms to make learning more engaging and help students develop their critical thinking skills was truly inspiring.[iii]<\/a> It\u2019s disheartening to see students struggle to write 5-paragraph essays when AI tools like GenAI do much of the work for them.<\/p>\n I recognize this post might sound somewhat pessimistic, as my personal views on AI are evident. However, there’s hope for public historians and society if we prioritize media literacy. Grasping how these models operate, their development, ethical usage, and limitations is crucial for navigating an AI-driven future.<\/p>\n ——<\/p>\n [i]<\/a> https:\/\/about.jstor.org\/blog\/bringing-hidden-histories-to-light-an-archivist-reflects-on-ai-archives-and-the-future-of-digital-stewardship\/<\/p>\n [ii]<\/a> https:\/\/cblevins.github.io\/posts\/llm-primary-sources\/<\/p>\n