Exercise 3
Using AI to “interpret” the past / Responsible use of generative AI
Why Historians Aren't Going Anywhere
I remember watching The Imitation Game, the film starring Benedict Cumberbatch as Turing. The film stuck with me, not just for the test but for showing Turing's own humanity that went unrecognized by the state, who also persecuted him for being gay. It doesn't make sense to me because he designed a machine that could imitate conversation, yet society couldn't see him.
When Dr. Humphries' lecture mentioned the Turing test, my mind kept coming back to the irony: if a machine can convince you it is human, should we consider it intelligent?
The Turing story asks a deeper question: whose humanity do we choose to see, and whose do we choose to erase?
From the class demo of testing a chosen AI with historical prompts, I navigate and reflect on this question. I used Claude AI (Sonnet 4.6, the free version) and saw for myself what AI can really do on par to historians, and what it missed.
What Learned from Claude AI
(What AI Can Do (And Does Well)
The topic of the French Revolution in and of itself was a daunting one but asking Claude to summarise the topic, produced a response that was chronological and neutral. Claude walked me through different schools of thought and named key historians, and incorporated a note that no single-cause model holds up today.
This was incredibly useful for someone in need of a quick orientation on an overwhelming topic.
I prompted about primary sources as a follow up and Claude delivered with a breakdown of the different forms. It provided options from cahiers de doléances (notebook for ordinary people's grievances), provincial archives (for studying counter-revolution and police reports (that tracked what ordinary people thought). I think this was helpful because it specified what would be the most "helpful" in interpreting voices from both bourgeoisie and proletariat lens.
I created a new chat, I uploaded texts by Lucretia Mott and Elizabeth Cady Stanton. I prompted Claude to find recurring themes, and for each of them, attaching a quote from the corpus. Claude provided me with seven: Innate Equality, Denial of Rights, and others.
In prompting about gender-history perspective, Claude produced an analysis that introduced new concepts I did not know about such as "coverture" (a legal doctrine where a woman's rights were absorbed by their husband) and the "separate spheres" ideology.
Claude AI was powerful in summarizing debates, pointed me towards multiple sources it found "relevant" or "useful". It spotted patterns across documents and applied academic frameworks.
But becuase I provided clear instructions, with specific context and also prompting it to take on a persona, it was able to give me a sophisticated output.
So, what is AI doing? AI is working with stuff that already exists summarising, retrieving, spotting, and applying. This is not new in research, it doesn't offer new ideas, genuine insights, nor is it surprised by evidence. It takes no stake in anything. It's like an RA that's read everything but never felt anything.
Answering the Three Questions
What kind of tool is AI?
Based on the demo, AI is an RA and a tool for finding patterns, not a co-author. When I pushed for an original interpretation, it still only gave me synthesized patterns. AI's power stops at execution, but all responsibility remains entirely ours.
What should we not ask AI to do for us?
To never outsource our core original research because once it is uploaded, your data trains future models, often without informed consent.
That work becomes a data point for the big tech companies. Additionally, never ask AI to genuinely understand perspective we have trained it to see as wrong. For example, our demo contained the prompt to analyze texts from a perspective skeptical of women's suffrage. Claude struggled with this response. It wants to perform the modern values it had been trained on. I think AI can identify the tension but cannot wrestle with it.
Who benefits from it?
The Big Tech Companies control the tools, training, data and pricing. So, each second, we use AI, it's helping to train the next version, almost like free labour to gain access.
The knowledge sourced by AI is from data that's widely available on the internet. This means it overlooks local, and non-digitized histories. This is confirmed by a commenter in Dr. Humphries article. Still, generative AI's convenience gives a huge amount of power to the big tech companies that own LLM.
Is this really the last generation of historians?
An analysis of Gans and Humphries
I think the Turing test forces us to critically assess if AI becomes indistinguishable from a human, will society still value human "detective work" (mentioned by Dr. Humphries in the comments)? Or is the output all that matters, thus making the human historian economically obsolete, such as Dr. Humphries' woodworking hobby competing with IKEA?
Both authors spent most of their post demonstrating how powerful AI has become.
Joshua Gans demonstrated that it was able to publish a peer-reviewed paper with the help of AI. Humphries watched as AI navigated across 21 archives on its own. So, the evidence is right there, AI can do research.
But I think in both of these posts, the things that the authors felt they had to add, almost as an afterthought, were relevant.
For Gans it was the disclaimer about taking full responsibility because in including that disclaimer, he and the journal knew that the accountability cannot be outsourced even if AI did the work. If someone questions the findings, they won't email OpenAI, they'll email Gans.
For Humphries he describes Deep Research hallucinating, fabricating quotes and even producing codes that don't work. His conclusion reminds us to treat LLM's critically.
Critical engagement...that's a historian's job.
Even when AI is the most impressive, it still needs to have humans catch its mistakes, take responsibility for its outputs and decide what questions matter in the first place.
The future of historical research looks like a partnership to me.
If we let AI do the thinking and taking the responsibility, we don't just lose our jobs, we also lose accountability, we lose human willingness to revise, to sit with the discomfort and to be wrong.
History becomes a product without a producer.
So, to answer the question: is the last generation of historians...?
That depends on whether we noticed what Gans and Humphries themselves almost seem to hide in plain sight, and that's the failures and the moments where AI broke and the human had to step in. They're not just footnotes, they're the whole point.
A Personal Response
The imitation of a true human craft cannot be the same as understanding it. Whether it is a table or its human conversations, if we value only the output, we lose what is irreplacable.
I think AI does not need to match the "best" historian, it just needs to be "good enough" for everyday tasks as IKEA can simply be "functional" for most people.
If it can do that, then the demand for historians could shrink.
I return to Turing. That story reminds us that history is more than facts. It is about identity that shapes lives and legacy. AI can summarize it in seconds, but the algorithm cannot comprehend it. The test can measure imitation, but history requires empathy, lived experience and a recognition of our own humanity. For that, we need human historians.
Final Thoughts
After running Claude through historical prompts and reading Gans and Humphries, I land on the fact that AI is powerful. It's able to summarize debates I know nothing about and point me towards archives I didn't even know existed, and spot patterns across texts I would probably miss on my own. It's a legitimately useful tool for research.
However it cannot take responsibility. It cannot wrestle with perspectives it's been trained to see is wrong and it cannot decide whose stories matter.
The Turing test measures imitation. Can a machine convince you that it's human? By that metric, AI is close.
But Turing’s own story asks that deeper question: whose humanity do we choose to see? Historians have to make that choice every day. They decide which stories deserve to be carried forward.



















