AI and History Research

AI is all the rage these days. I am finding that even when I travel to speak at colleges, I get questions from faculty and students about AI in teaching and research. Mind you, I have no claim to AI expertise, so these questions indicate just how buzzy the topic has become.

I never want to be the “old guy” who doesn’t know how the latest technology works, so I have indeed begun making some initial forays into using AI for history research and writing. As I have written before, creative workers have to draw a hard line when it comes to using AI to create work that you present as your own writing. This is a form of plagiarism.

Plus, I will need a lot of convincing before I could believe that AI is doing anything more than synthesizing knowledge. It can’t create; you still need humans made in God’s image to do that.

However, I have been training myself to ask, whenever I need to do non-creative tasks, “could AI do this for me?” Or, “could AI help me do the work that only I can do more efficiently?” Once you get used to thinking this way, you find increasing numbers of instances where AI can be of assistance. Let me give three examples:

Summarizing manuscripts I need to read and assess. I regularly need to read and comment on book-length manuscripts. These include doctoral dissertations, books that publishers are reviewing for possible publication, and published books that I am reviewing. When I am reading a book for the first time, I ordinarily see if I can find a review or two of the book. I read those before I get started on my own read, just so I can get my bearings on what kind of book it is, how it struck other readers, etc.

You won’t find reviews of books or dissertations that haven’t been published, however, but if you upload the manuscript to an AI program, it can generate a summary of the book of (say) 1000 words that functions somewhat like a book review. As usual, I still need to read the manuscript myself, but an AI-generated summary can give me a much stronger sense of what the manuscript is before I start reading.

Finding and documenting sources. I was recently writing a magazine article based on one of my current book projects. I assumed correctly that the publisher did not want footnotes. What I did not know was that the publisher’s fact-checker was going to want to see sources to confirm every factual claim. It’s a type of situation that comes up regularly - no creativity is required, but if I had to re-track the sources down manually it would certainly take an additional day or more of work.

So I plugged the copyedited manuscript of my Word document into AI (Copilot, since that’s what we subscribe to at work). I told it to give me a published source reference every time the fact-checker asked for one. I had to answer a few more questions for AI to grasp what I wanted, but once it started, Copilot generated a solid list of dozens of reliable sources and page numbers in just a few minutes.

Transcribing handwritten sources. This is the function I have explored the most. For my current book project on Adoniram Judson, I have hundreds of handwritten letters and journals to process. Many of these are documents that were published during Judson’s lifetime, but I never know when the publisher in the 1800s might have left out portions of the letter (often the most interesting sections because they deal with controversial topics). Most AI programs now have some Optical Character Recognition functionality that can convert handwriting (more or less reliably) into text.

I am quite confident that Copilot is not the best option for the transcription of handwriting. However, it is integrated well with the rest of the Microsoft 365 suite, so that (for example) it is easy to convert the transcription into a Word document and save it in a file of transcriptions. I have heard that Transkribus is the best option for AI transcribing historical handwriting. However, I find it difficult to use and quite slow in the free version of the software, so for the time being I will stick with Copilot.

To transcribe a document, you upload a pdf of the letter/journal and tell AI to transcribe it. You can develop a “baseline” of the person’s handwriting by uploading more documents and correcting the transcription errors that AI makes. Judson’s handwriting is often pretty legible, but damage to the document, oddities in his script, or crossed out lines can make some words difficult to recognize via OCR. So far it averages getting the transcriptions around 85% correct.

If I was trying to publish an edition of Judson’s papers, of course, we would need to shoot for 100% accuracy, but that’s not essential for my purposes. I just need to get a good transcription, and be able to compare it to the published edition when one is available. If one is not available, I need to read the transcription more closely to see if there’s any new information I want to use in the book.

When using AI in research, we want to make sure we are not cutting any ethical corners or having AI do our creative work for us. One test of whether the use is ethically sound is whether you would be willing to go public with the ways you’re using AI (say, for example, in a Substack newsletter!). If you are willing to tell everyone what you’re doing, it at least means that you’re not trying to “get away” with something unethical.

AI may have many ramifications, but at a minimum, it is a new technology (like the internet, digital sources, Google Books, etc.) that has valid applications in history research. It will undoubtedly make certain tasks (like document transcription) vastly more efficient than doing it manually. But there are certain things it can never replace, including the human creativity behind good writing.

It’s a busy time for the writing pipeline! The second edition of my American History textbook is due out in July, in time for Fall ‘26 courses.

September 1 is the release date for The Death of Religion? Nones, Others, and the Flourishing of Faith, co-authored with my friend and colleague Byron Johnson of Baylor’s Institute for Studies of Religion.