The recent advancements in artificial intelligence (AI) have raised questions about the future of historical research.
A test conducted by Professor Benjamin Breen using OpenAI o1, the latest GenAI model from Sam Altman's company, showcased the impressive capabilities of AI in analyzing an eighteenth-century Mexican medical manuscript.
Not only did the AI perfectly transcribe the text in historical Spanish, but it also identified graphic details that had escaped the historian himself.
This outcome prompts discussions on whether researchers will still need to delve into libraries, analyze manuscripts, learn ancient languages, and grasp paleography, considering the remarkable abilities AI has demonstrated. Breen mentioned in a video call with Wired that while historians conducting field and archival research are essential, the tests he conducted revealed that technological advancements have propelled us further ahead than commonly believed.
New Language Models Enhancing Historical Analysis
Cutting-edge language models like Gpt-4 by OpenAI and Claude Sonnet by Anthropic are proving to be formidable tools for historical document analysis. These models can transcribe ancient texts in various languages, provide accurate historical context, and even propose new interpretations.
Their ability to analyze multilingual texts, such as those with Renaissance Latin intermingled with Hebrew phrases, showcases their versatility in handling diverse historical materials.
One notable application is the analysis of Jesuit missionaries' writings in the New World, composed in erudite but intentionally complex Latin. This extensive corpus of hundreds of books remains largely unexplored and understood due to the dwindling number of scholars capable of deciphering them.
AI's involvement becomes crucial in this scenario. The capability of cutting-edge models to process large datasets and unveil hidden connections is reshaping the way historians will operate. Breen, an expert in historical research, conducted a scientific experiment published in his Res Obscura newsletter.
By testing the abilities of Gpt-4o, OpenAI o1, and Claude Sonnet 3.5, he aimed to explore the present-day potential of these systems. These AI systems can concurrently analyze texts in multiple languages, identify hidden patterns, and suggest relevant bibliographies with an accuracy that was previously deemed unattainable.
A Glimpse into Historical Interpretations
Breen, who teaches History of Science and Medicine at the University of California, Santa Cruz, conducted three in-depth tests using the AI models. The first test involved analyzing Urbano Monte's sixteenth-century hand-drawn world map manuscript in italic Italian script. Gpt-4o's transcription was almost flawless, with only a few errors, showcasing the AI's attempts to correct words used by the original text's author.
Additionally, Gpt-4o provided a remarkably accurate analysis of the historical context, citing relevant academic sources unknown to the professor himself. In the second test, a medical text from 1770 Mexico was analyzed by OpenAI o1, which not only conducted a correct analysis but also identified details that eluded the researcher, such as a small element in the center of the physician's drawing.
Finally, in the third test, Breen compared the AI models in analyzing correspondence between William James and Francis Galton, two significant figures in the history of science spanning the nineteenth and twentieth centuries. While Claude Sonnet 3.5 exhibited caution and self-criticism, focusing on its limitations, Gpt-4 proposed eight different original interpretations of the scholars' relationship.
One interpretation stood out for its originality, portraying James as an "anti-panopticon" theorist in contrast to Galton's statistical approach, impressing Breen with its novelty within conventional interpretative frameworks.
Challenges and Opportunities
Despite the promising outcomes, AI systems are prone to converging towards "average" interpretations, lacking true originality. This challenge underscores the need for historians to maintain a balance between utilizing AI for mechanistic tasks such as transcription, translation, and preliminary document analysis while preserving the creativity and interpretative aspects of historical research.
The collaboration between human and artificial intelligence may herald the future of historical research, propelling the discipline towards new methodologies and insights. While AI's integration into archival research has immense potential in unveiling hidden connections and suggesting novel research directions, it requires a cultural shift within the historical discipline to embrace these innovative technologies effectively.
In conclusion, the true revolution lies in how historians will leverage AI to make discoveries that would otherwise be unattainable. The symbiotic relationship between human intellect and artificial intelligence presents a promising avenue for exploring the past, ultimately enriching our understanding of history.
