What do novels teach us? Using AI to trace the moral history of the novel

Stories have always taught us something. From Aesop’s fables to modern novels, narratives model values, decisions, consequences, and ways of being in the world. As the literary critic Wayne Booth famously put it: all stories teach.

But if that’s true, an intriguing question follows: what exactly have novels been teaching us over time? And how might we study those lessons at scale, across thousands of books, cultures, and decades?

In a recent paper for the Journal of Computational Literary Studies, I explored how large language models (LLMs) can help us sketch what I call a moral history of the novel—not by imposing a fixed moral framework on literature, but by surfacing the kinds of life lessons that stories themselves seem to encode, as interpreted by readers and models alike.

From Stories to Story Morals

The idea of a “story moral” is familiar from children’s literature, but it applies more broadly. While novels do not end with a tidy lesson, they still convey important ideas about loyalty, resilience, truth, love, justice, responsibility, or the cost of ambition. These values are rarely stated outright. Instead, they emerge from how events are selected, arranged, and framed.

Rather than asking what happened in a story, story morals ask: why was this story told in this way? What general lesson might a reader reasonably take from it?

Traditional computational approaches to literature have focused on topics, genres, or plot structures. Story morals operate at a higher, more interpretive level—closer to how readers reflect on stories once the details fade.

Wikipedia as Literary Resource

To study moral patterns at scale, I turned to an unconventional source: Wikipedia plot summaries.

At first glance, this might seem like a compromise. Plot summaries are reductive; Wikipedia is uneven; contributors are not representative of the world’s readers. But that’s precisely the point. These summaries reflect how stories are remembered and retold by a broad, non-academic public. They provide a perspective of the novel’s history.

Using a dataset of nearly 10,000 English-language Wikipedia summaries of 20th- and 21st-century novels, the project treats summaries not as imperfect substitutes for books, but as meaningful cultural artifacts in their own right—snapshots of collective interpretation.

In that sense, Wikipedia becomes part of a new, computationally mediated hermeneutic circle: stories are interpreted by readers, summarized by contributors, reinterpreted by models, and then analyzed again by researchers.

LLMs as Moral Interpreters

Large language models are particularly well suited to this task—not because they are “objective,” but because they are good at abstraction. Asking a model to infer a story moral is closely related to asking it to summarize, generalize, or label a narrative’s underlying logic.

In pilot experiments, I found that small changes in prompt wording—whether the model was framed as an “expert,” where the summary appeared, or whether the question invited interpretation rather than instruction—produced strikingly different moral keywords for the same story. On average, fewer than 40% of keywords overlapped across prompt variants.

Rather than treating this variability as a flaw, I embraced it as a feature. Interpretation, after all, is perspectival. Instead of searching for a single “correct” moral per novel, the project randomly assigned different prompt formulations across the dataset, aggregating many slightly different perspectives into a broader moral landscape.

The Moral Landscape of the (Wikipedian) Novel

When you step back from individual books and look at thousands at once, patterns begin to emerge. At the same time, those patterns depend on how we cluster morals together. Here are two looks at some of the major thematic clusters of morals accordng to the novel archived by Wikipedia.

One of the most surprising findings emerged when tracking these moral clusters over time. Despite dramatic historical, political, and cultural shifts across the 20th century, the relative prominence of major moral groupings remained remarkably stable.