The rise of ‘artificial historians’: AI as humanity’s record-keeper

In documenting and recording society’s collective data on an unprecedented scale, artificial intelligence is becoming humanity’s historian—changing the way we record information for posterity.

But AI’s inadvertent role as memory-keeper raises profound concerns for today’s historians. Unlike human historians who explicitly document their methodologies, AI systems are creating the historical archives of the future without crucial transparency around how sources are selected, weighted, and interpreted.

This undermines a fundamental principle of historical scholarship, that methodologies should be visible and contestable. In the new book “Artificial Historians”, historian Marnie Hughes-Warrington explores how AI systems are transforming historical records.

The author argues that AI is already deeply involved in history-making, generating ‘most of the histories made around the globe’ on a daily basis. Rather than seeing this as only a threat, the author encourages historians to see it as an opportunity to engage with AI development to ensure these systems reflect historical complexity.

History and nuanced understanding

Hughes-Warrington puts forward concerns around biases in data collection, specifically the “uneven and unfair collection of information about the past.” When AI systems train on these biased historical records, they risk amplifying and perpetuating historical inequities, potentially cementing problematic narratives for future generations.

As well as this, some historical information may simply not be computable or readable by AI tools, giving an incomplete picture.

As well as the concerns around information gathering and transparency, Hughes-Warrington points out that AI misses the nuances of historical storytelling that humans inherently accept.

She explains that historical claims made by scholars and historians are never fully or perfectly true but are “partially grounded,” meaning they refer to evidence outside themselves and invite testing. This complex understanding of historical information, or historical “truths,” presents a challenge for AI systems trained to provide definitive answers.

When asked about world history topics, AI platforms tend to give similar, conventional responses that present a limited view of history, demonstrating that AI systems lack the nuanced understanding of historical context that human historians develop through years of study, Hughes-Warrington suggests.

“Information from the past might not be available or even computable, or presented in ways that make the use or combination of datasets difficult or even impossible,” Hughes-Warrington explains.

“The contexts for data collection might also be ignored. If you knew that information was collected about people in financial or judicial distress, for instance, would you use it without thinking about their experiences? Most importantly, though, there may be an overconfidence in the development of algorithms or the detection of patterns.”

AI is here to stay

“This hollowing out of history and its absorption into future, fiction, or geopolitics means that the historical expertise needed to make AI more effective and fair is missed. AI is not a threat to history if we see the invitation to be involved in its making,” she explains.

“By bringing historical expertise to AI development, we can create more effective and fair artificial historians while preserving the critical thinking and contextual understanding that defines quality historical scholarship.”

The text ultimately suggests that history-making is a complex, interpretive process that cannot be reduced to simple algorithms or rules. Hughes-Warrington challenges historians and AI technologists to think more deeply about how we define and create history.

“If history is the problem, then history is also the solution,” Hughes-Warrington concludes.