The Quest for a Thesis

In my last post, I discussed the seven dimensions of historical analysis. Shortly after I wrote that post, I started reading The Methods and Skills of History: A Practical Guide by Michael Salevouris and Conal Furey. I was pleasantly surprised to discover that they devote the entirety of chapter four to the topic of multiple causality in history.

One of the key points Salevouris and Furey make in their book is that historical research begins with a thesis. A thesis, as they define it, is “the author’s central argument about the topic under discussion” (p. 135). Let’s flesh this out with a concrete example. (I’ll be using a wide variety of cases in future posts, most coming from the field of civilizational analysis.) In this case, the topic is sumptuary (fashion) law in early modern Venice, but I don’t yet have a thesis, an overarching perspective to structure the inquiry. I therefore turned to ChatGPT 4.o, the latest model from OpenAI, to ask for some help. My first prompt was broad where I asked it the following:

I need to develop a thesis about sumptuary law in Renaissance Venice. This thesis will serve as my argument in a history paper I need to write for a graduate seminar. Can you help me?

Interestingly, the system responded with a multi-dimensional reply that included four of the dimensions I had already identified – economic control, moral regulation, social hierarchies, and political power. Some of the research suggestions were interesting and worthy of further exploration. Pleased with the model’s response, I then asked ChatGPT to write a thesis statement specific to one of the dimensions. Here’s one of those prompts and the system’s response:

Can you write a thesis statement for reinforcing social hierarchies?

"Sumptuary laws in Renaissance Venice were strategically implemented to reinforce and preserve social hierarchies by regulating attire and luxury consumption, thereby maintaining clear distinctions between the patrician elite and the emerging wealthy merchant class. These laws served as instruments of social control, ensuring that visual symbols of wealth and status remained exclusive to the nobility, thus preventing social mobility and reinforcing the existing class structure."

ChatGPT 4.o (12.05.24)

Now that I’ve read some books on Venetian history and an article or two about sumptuary law, I can evaluate this response with some expertise. In a word, it’s mediocre. The problem is that the historical reality was much more nuanced. In 1297, for example, the Great Council was enlarged to 1,000 members. With that enlargement, some social mobility is evident in the historical record, a fact that contradicts, in part, ChatGPT’s assertion that sumptuary laws prevented social mobility.

What’s more interesting, however, is a comment made by the Venetian historian Antonio Foscari (2020), who notes that sumptuary laws forbade nobles from meeting together in private homes even if the event was a party. Here, the purpose of the law was to prevent gatherings where insurrections could be plotted. These were genuine concerns at the time. Renaissance Italians had mastered the art of the feud. The ongoing conflict between the Montagues and Capulets in Romeo and Juliet was a reality for most Renaissance Italian cities. How do we prevent civil strife from breaking out? That was the question of the day, and Venice answered it in a unique way, partially through the creative application of fashion law.

Although the thesis written by ChatGPT was mediocre, my reaction to it was useful. Disagreeing with a machine is not that much different from having one with a human being. I strongly disagreed and then went looking for evidence to support my argument. Sumptuary laws played a role in maintaining distinctions between social classes, but they had an equally important role in ensuring that powerful Venetian families did not descend into conflict.

So, what have we learned about the usefulness of AI from this experience? First off, there’s no substitute for reading and what I call “old-school” research. The historian must still read extensively and master the nuances of whatever they’re studying. One cannot correctly evaluate an AI-generated response any other way. Positively, a large language model can act as a useful foil, especially when one has done enough background reading to assess its responses.