Can Advanced Language Models Mimic the Writing Style of an 18th-Century Fur Trader?
By Josh Dzieza, an investigations editor with a focus on tech, business, and climate change at The Verge, where he has received accolades such as the Loeb Award for feature writing.
Upon the release of ChatGPT in late 2022, Mark Humphries, a historian delving into the 18th-century fur trade, initially viewed the AI tool’s capabilities as mere parlor tricks. However, his perspective shifted when he realized the potential applications of artificial intelligence in his research. Humphries, a professor at Wilfrid Laurier University, had been exploring AI’s utility in transcribing antiquated scripts, a task that typically involved tedious error correction. Intrigued by ChatGPT’s ability to rectify transcription errors in a handwritten French letter, Humphries saw the opportunity for AI to streamline and enhance historical research processes.
The vast digitization of historical documents in recent years presented a challenge due to the diverse formats and vocabularies, hindering efficient automated search capabilities. Humphries envisioned leveraging AI to extract and analyze vast amounts of historical data, revolutionizing the way historians interpret and understand the past. By automating the extraction of relevant information from archives, AI could potentially unveil new insights and connections that would otherwise remain undiscovered.
While the integration of AI into historical research shows immense promise, it also raises concerns about authenticity and bias. Historians, like Cindy Ermus and Katherine McDonough, are cautiously optimistic about AI’s transformative impact on various aspects of historical research, from data collection to analysis. AI tools offer unprecedented access to historical archives, enabling researchers to uncover hidden narratives and perspectives that were previously overlooked.
However, the inherent biases in AI systems, such as gender and racial biases, pose challenges in accurately interpreting historical data. Lauren Tilton highlighted the issue of “presentism,” where AI models reflect contemporary perspectives and struggle to comprehend historical contexts accurately. Despite these challenges, historians like Humphries are actively exploring AI’s potential to enhance research efficiency and expand historical knowledge.
Humphries’ endeavor to develop an AI research assistant, named HistoryPearl, exemplifies the evolving relationship between AI technology and historical scholarship. By training AI models to analyze and synthesize historical texts, Humphries aimed to automate the process of extracting valuable insights from historical documents. Through iterative refinement and experimentation, HistoryPearl demonstrated significant progress in efficiently processing historical data, outperforming traditional research methods in terms of speed and cost-effectiveness.
As AI continues to advance, historians like Humphries are grappling with the implications of integrating these technologies into their research practices. While AI presents new opportunities for historical inquiry, it also underscores the importance of domain expertise and critical assessment skills in leveraging these tools effectively. Humphries’ journey with HistoryPearl underscores the evolving role of AI in historical research and the ongoing dialogue on the intersection of technology and academia.
