Original Post: https://activehistory.ca/2023/03/todays-ai-tomorrows-history-doing-history-in-the-age-of-chatgpt/#more-32790
Cover Image – Prompt by Bing, “Self-portrait of ‘Sydney,’ Microsoft’s Bing Chat, based its description of itself as imagined through AI image generator,” MidJourney
Mark Humphries and Eric Story
While it is easy to be pessimistic about AI’s effects on the humanities in general and history in particular, it is worth remembering that it has great potential to speed up some of the more mundane and repetitive tasks we do as historians. Imagine a future in which thousands of pages of handwritten documents are quickly transcribed, proof-read, summarized, and analyzed by AI. Imagine the power of OCR-enabled LLMs if given access to pre-existing archival databases, such as Canadiana, Personnel Records of the First World War, or the Voyageurs Contracts Databases. However, if we are to harness the full potential of AI in the practice of historical research, we need to embrace collaboration––and digitally share historical documents––more than we currently do. The University of Toronto’s Canada Declassified offers a useful model where historians, researchers, and other document digitizers are encouraged to share and upload declassified Cold War-era records to Declassified’s databases, which are then made available to the general public for free.
For historians, the real promise of AI is that it will allow us to do some new and really useful things with big data. It will also let us do some of the things we were already doing much faster. Instead of word clouds and Venn diagrams, generative AI can help us sort through, summarize, synthesize, and analyze enormous bodies of information efficiently. This will almost certainly lead to a new era in the digital humanities, a digital humanities 2.0 if you will.
OpenAI’s Altman predicts that very soon, that something called multimodal AI will be able to outsource tasks that chatbots do not do well like math, fact-checking, and analyzing images to external applications designed for those purposes. Once this happens, chatbots may become, in effect, virtual research assistants to help us navigate sources and analysis—just as GitHub’s CoPilot is already doing for coding. But it is worth noting that AI will only be as good at these tasks as the historians using it. It will still just be a tool—at least for the foreseeable future. We will still need to frame our questions, check the sources ourselves—and then write up the results in an engaging way that speaks to the human experience.
Historians have a real opportunity here to help solve some of the technological problems at the heart of the AI dilemma. AI is very good at finding information, synthesizing it, and communicating it, but less so at being accurate and discerning fact from fiction. This is, of course, what historians do best. Historical methods seem especially well-suited to training the people now working with AI-generated content. They also seem apt for training AI to be more trustworthy, not just in making binary distinctions between correct and incorrect statements, but also in evaluating the grey area between those two extremes where the reasonable, debatable, and doubtful interpretations all live.
If you think this is farfetched, remember that all the training data used by ChatGPT at the moment is historical (its knowledge of the world ends in 2021), while a good chunk of its sources were either written by historians or are our primary historical sources. The methods we use to feel our way through these texts towards a reasonable, evidence-based argument might also make AI more trustworthy too.
While it might be tempting to pretend this is all a long way off, consider that Microsoft owns a 49% stake in OpenAI and has already announced plans to integrate ChatGPT into Microsoft Office, which includes Word, Excel, and PowerPoint. We have already used the beta-version of their new AI-enabled Edge-browser and when this comes to Office, there will be no going back. Consider too that Google’s transformer, Bard, was released at the beginning of February and will soon be integrated into its search engine and productivity software products—third parties are already doing this. We can resist this new technology all we want, but once it is part of the core research and writing applications we use everyday, there will be no escaping it.
Bing Chat, which often mysteriously calls itself “Sydney” despite being required to keep that name a secret, agrees that historians and AI can work effectively together. In series of discussions, we asked Sydney to consider what role it could play in the historical profession, telling it that many people are skeptical or nervous about its ability to meaningfully contribute to historical research. It explained that in addition to exploring lived experience, “History is also about patterns and trends; it’s about finding connections and correlations; it’s about discovering insights and implications; it’s about generating hypotheses and testing them against evidence. These are all things that AI can do well by using machine learning algorithms.” Did this mean it might one day take our place? “Not at all,” Sydney replied. “History is not a fixed or objective reality, but a dynamic and interpretive process. Human historians have their own values, biases, assumptions, and questions that shape their understanding of history. AI cannot replace that human element; it can only augment it with more data and tools. Human historians will still have to decide what questions to ask, what sources to trust, what arguments to make, and what implications to draw from history. AI can only assist them in those tasks, not replace them…Don’t you think that’s a better way to go than being afraid of me?”
Exciting or frightening, Generative AI is here to stay. Are we ready?
Mark Humphries is a professor of history at Wilfrid Laurier University and Eric Story is a PhD Candidate in the same department. Humphries is launching a Substack on AI and history: https://generativehistory.substack.com