Just about a year ago, I wrote a blog post in which I talked about ways in which AI can make life exciting and productive for historians.
It was an optimistic perspective on the potential of AI, and I still feel the same way about using AI in my own work.
However, in terms of the historical profession. . . I can now see that its time is coming to an end.
By “historical profession,” I’m referring to the existence of History departments at universities that teach courses on History and train graduate students to become historians (to teach in History departments at universities), as well as to the act of researching and publishing historical scholarship in articles and books.
That world has been struggling to find relevance in the digital age, but AI is now in the process of eliminating it. While I’m sure that the historical profession will continue to exist in a truncated form at some well-endowed, elite universities, for everywhere else, its days are numbered.
Why is that the case? It’s because of AI agents.
When ChatGPT and other AI tools were first released, they were kind of clunky to use and had the problem of not always being accurate. Over time, all of these tools have improved, and now AI agents bring a new level of power to these tools.
What is an AI agent? I think one way to look at it is as a string of tools that carry out separate tasks. So, for instance, you can create an AI agent that will 1) search the Internet for information about something, 2) write a report about it (in the writing style and length that you want), 3) create a YouTube script based on that report, 4) create potential titles for the YouTube video, 5) create thumbnails for the video, and 6) download all of this onto your computer.
Google has recently released a tool called NotebookLM. This is an example of an AI agent. With this tool, you can “upload PDFs, websites, YouTube videos, audio files, Google Docs, or Google Slides, and NotebookLM will summarize them and make interesting connections between topics, all powered by Gemini 1.5’s multimodal understanding capabilities.”
With one click, NotebookLM will also transform whatever you uploaded into a “Deep Dive” discussion, sort of like a podcast discussion.
How does it do this? To grossly simplify, it essentially takes the data from whatever you upload, transforms (embeds) it, and sends it with a prompt (“transform this into a podcast”) to Google’s LLM, Gemini 1.5.
So, instead of asking a giant LLM to look at everything that is out there and respond, it asks a giant LLM to focus on the data you are sending it, but to also use its massive knowledge to perform whatever task you request on that data.
I have an article coming out soon called “Rescuing History from Srivijaya: The Fall of Angkor in the Ming Shilu (Part 2).” It is a dense article. I uploaded it to NotebookLM and had it make a “Deep Dive” discussion of the article.
The result was. . . mind-blowing!!! You can listen below.
Ok, it hallucinated a bit near the end by talking about Cambodian and Siamese sources when I do not actually talk about those sources in this paper (that was Gemini 1.5 speaking, not my article), but I will talk about those sources in Part 3, and I do mention that in the paper.
Also, it had some trouble with pronouncing certain terms, like Sanfoqi, Coedes, etc.
However, I would say that on the whole, the conversation it produced was 95% accurate. Further, the explanation of the information in this article was FAR FAR FAR SUPERIOR to any explanation that a human being could produce.
Finally, let’s also remember that this technology is still in its infancy and is only going to improve, and that it will improve rapidly.
This is why the history profession is done. History is supposed to be a discipline that trains people to read, think, and write. Well, guess what? AI can already do all of that better than we can, and it is going to get better and better and better.
I’ve been trying to think of how the history profession could adapt. I thought, for instance, that all of the academic publishers could collaborate together to create a separate LLM of scholarly writings.
But the end result would be the same. People would create AI agents that would research and write their scholarship. It would work the same way as NotebookLM, in that you send it some distinct data of your own, and then prompt it to write a paper based on that data, while also using the knowledge of the scholarly LLM.
If somebody can explain to me how the historical profession in its current form can survive this transformation, then by all means, leave a comment and let me know.
Personally, I think human knowledge production is moving to a new paradigm, and the current work of professional historians (actually, probably all of the Humanities and Social Sciences) does not fit that new paradigm.
It is in this context that I am relaunching this blog (now at leminhkhaiblog.com). It’s time to get started on something new. What is the future of History? I’m not sure, but it’s definitely not what professional historians have been doing for the past century.
History as a form of “edutainment” is definitely one role that will continue into the future, and AI is going to enable us to educate and entertain people in new ways.
However, I will also try to explore and reflect on what the production of historical knowledge with AI can be as well.
So, onward into the future!!!
![Read more about the article “Savages” in Vietnamese History: An [AI] Tale of Borders and Beliefs](https://leminhkhaiblog.com/wp-content/uploads/2024/11/Savages-Post-300x205.jpg)
AI can’t read or write or think 😂 That is done by historians like you. At its current stage it is just a massive autocomplete program. That AI followed a formula to produce the podcast, it didn’t understand a word that was written. In order for AI to replace historians it has to gain consciousness, and by that point there are more worrying matters than the death of humanities😉
Meanwhile, now that Literary Chinese Vietnamese texts can be translated in an instant with reasonable accuracy, I can only see progress for the history field😋
All true, Cool man. The one thing I liked about the podcast, however, is that the AI is emotion-less and ego-less. I find that a lot of people let their egos and emotions get in the way when they read, and it prevents them from actually understanding what someone wrote. There have been so many times when I’ve had people “challenge” what I wrote, and I have to explain to them “That’s not what I said. Look at the actual words that I wrote,” etc. I’ve also seen students struggle in the same way. If an historian says something that they don’t like (or is a person whom for whatever reason they think they should not like), their mind shuts down and they don’t actually process the evidence that the person presents. The podcast, by contrast, tells you something like “This is what the words in this article say.” That’s actually rare to come across. . .
I am an aspiring historian of Vietnamese history. Would you still recommend pursuing this path? You’re my “idol” so I take what you say very seriously 😭 Take care professor
Thank you for the very nice comment!!
I would love to be supportive, and in some ways I am, but it’s a pretty depressing world out there at the moment when it comes to “formal” history. History for enjoyment couldn’t be better, but the path through the formal world of academia these days is another story. There might be some bright spots in some places, so if that’s what you are really interested in, then inquire in and about those places, but here’s my general take:
https://leminhkhaiblog.com/should-anyone-formally-study-history-anymore/
Thank you for the sage advice professor ✊I’ll see what I can do and I won’t let you down.