The Problem with Ranking History

Can we really rank the "most important" people in history? The idea seems almost absurd on its face. How do you compare a religious founder to a scientist, a conqueror to a poet, a philosopher to an inventor? By what measure? Toward what end?

These are not merely rhetorical questions. They cut to the heart of what we mean by "importance" and whether such a concept can be meaningfully quantified at all.

What Do We Mean by "Influence"?

The prompt we give to models asks for "historical influence," but this term conceals profound ambiguity. Consider how differently these dimensions might rank the same figures:

• Causal influence: Did this person's actions directly change the course of events? By this measure, Genghis Khan ranks highly—his conquests reshaped the demographics and politics of Eurasia for centuries.
• Ideological influence: Did this person's ideas spread and persist? Confucius, Jesus, and Marx score well here, though they commanded no armies.
• Scientific influence: Did this person advance human knowledge or capability? Newton and Darwin are obvious examples—but how do we weigh discovery against application?
• Cultural influence: Did this person shape how we see and represent the world? Shakespeare, Homer, the Buddha—their influence is pervasive but diffuse.
• Negative influence: How do we weigh those whose influence was primarily destructive? Hitler "influenced" history enormously. Should that count? How?

Different models appear to weight these dimensions differently. Some lean toward "ideas that persisted"; others toward "actions that changed events." Neither is wrong. Both are incomplete.

We have not attempted to resolve this ambiguity. Instead, we let models interpret "influence" as they will, and we make their different interpretations visible through comparison. The disagreements are often more illuminating than the consensus.

Sunstein's Critique

In 2013, the legal scholar Cass Sunstein published a devastating critique of attempts to quantify historical significance. Writing about Skiena and Ward's book Who's Bigger?, which used Wikipedia metrics to rank historical figures, Sunstein raised fundamental objections that apply equally to any such project—including this one.

"We lack standards and tools to measure social contributions, certainly across time and across diverse fields and enterprises."

— Cass Sunstein, "Statistically, Who's the Greatest Person in History?" (2013)

Sunstein's critique begins with a useful analogy: baseball. In baseball, we can at least agree on what we're measuring—a player's contribution to winning games. Statisticians have developed sophisticated metrics like "Wins Above Replacement" (WAR) that attempt to isolate each player's contribution. But even here, disagreements persist.

"Who was the greatest baseball player of all time? Some people say Willie Mays. They emphasize that he had all of baseball's 'five tools': he could run, hit, field, throw, and hit with power. Other people insist on Ty Cobb, who had the highest career batting average in baseball history... People vigorously disagree about the relationship among those particular rankings and overall 'greatness.'"

If we cannot even agree on how to rank baseball players—where the goal (winning) is clear and the data is abundant—how much harder is it to rank historical figures, where the very definition of "importance" is contested?

"And what about religious leaders, scientists, philosophers, artists, and novelists? Can they be ranked as well—in terms of greatness or importance? Might we be able to play some kind of Moneyball with Joyce Carol Oates, Stephen King, James Joyce, Charles Dickens, and Thomas Hardy? Can cultural figures from diverse fields be ranked against each other? How might we compare Einstein, Plato, Descartes, Hume, Michelangelo, Suzanne Farrell, and Bob Dylan? True, it might be ridiculous, or even a bit crazy, to try."

Sunstein identifies a fundamental problem with Wikipedia-based metrics: they measure interest, not importance.

"It would make no sense to measure significance only by reference to the number of hits, page length, and number of edits; such an approach would denominate today's popular entertainers, and other celebrities, as history's most significant figures. Measures of that kind tell us about the interests of Wikipedia readers and editors, and knowing about those interests tells us something about popular tastes. But it does not inform us about significance."

He also notes the cultural bias inherent in using English-language Wikipedia:

"If the goal is to learn about worldwide fame or significance, it is more than a bit strange to rely exclusively on the English-language version of Wikipedia. At most, the resulting rankings reflect only the preoccupations of the English-speaking world, and mainly the United States. Surely Jesus would not have done so well in China, to say nothing of all those American presidents."

Sunstein concludes with a warning that applies to any attempt at quantification:

"Human beings can measure countless things... Our ability to measure is growing exponentially. It can be fun to rank people in terms of what we are measuring. But before doing that, we should be clear on what we have measured, and we should avoid nutty extrapolations. Bishop Butler, the eighteenth-century theologian, famously cautioned that 'every thing is what it is, and not another thing.'"

These criticisms are valid, and we take them seriously. HistoryRank does not claim to have solved the problem of measuring historical importance. But we believe there are reasons to attempt this project nonetheless—reasons that emerge precisely from the peculiar nature of the tools we are using.

A Counterpoint

Sunstein's critique was directed at human attempts to quantify history using Wikipedia metrics. But HistoryRank does something different: it asks large language models to make these judgments, then compares their answers to human-generated rankings. This changes the nature of the exercise in important ways.

We are not claiming to have found the "true" ranking of historical importance. Instead, we are creating an experimental platform that reveals how different forms of intelligence—human and artificial—make judgments about the past.

LLMs as Foreign Intelligences

Large language models are, in a sense, foreign intelligences. They are not human. They engage with human history at a remove, processing it through training on an enormous corpus of text spanning many languages, cultures, and perspectives. This gives them a peculiar vantage point.

When a human scholar compiles a list of "the most important people in history," their judgment is inevitably shaped by their native language, nationality, education, and cultural assumptions. An American historian will tend to overweight American figures. A Chinese historian will have different blind spots and different emphases. This is not a flaw—it is simply the nature of human perspective.

LLMs, by contrast, have been trained on text from across the world. They are inherently multilingual. They have "read" millions of books, articles, and documents in dozens of languages. While they certainly have biases—reflecting the biases of their training data and the choices of their creators—these biases are different in kind from human biases.

This makes LLM rankings an interesting counterpoint to human rankings. Where do they agree? Where do they diverge? When an LLM consistently ranks a figure higher or lower than human sources, what does that tell us about the assumptions embedded in each?

A Test of Alignment and Reasoning

Asking an LLM to rank historical figures by importance is, it turns out, a remarkably good way to probe how these models reason about issues of deep significance. The task requires the model to weigh competing values, to make judgments about what "matters," and to articulate its reasoning.

When you click on a figure in HistoryRank, you can see the explanation each model provided for why it ranked that person where it did. These explanations are a window into model reasoning:

• How do models understand "influence" versus "fame"?
• Do they weight moral judgments? (Is Hitler "important" because of the magnitude of his crimes?)
• How do they handle figures from non-Western traditions?
• Do different models systematically favor different domains (science vs. religion vs. politics)?
• What hidden assumptions emerge in their justifications?

This makes HistoryRank a potential resource for researchers working on AI interpretability and alignment. The rankings themselves are less important than the patterns they reveal: how models reason, where they converge, and where they surprisingly diverge from human intuitions.

Future directions might include: Do Chinese-developed models differ notably from American ones in their assessments? What if you ask the same model the same prompt, but in French? German? Turkish? Mandarin? Each variation becomes a probe into how language and training shape judgment.

The Biases We Know About

Through our Compare tool, several systematic biases have become visible. We name them here not to excuse them, but to make them explicit:

We surface these biases through the Compare page, which visualizes how models diverge. The goal is not to eliminate bias—that may be impossible—but to make it visible and therefore available for critical examination.

An Essay, in the Original Sense

The word "essay" comes from the French essai, meaning "trial" or "attempt." When Montaigne invented the form in the sixteenth century, he meant his essays to be experiments in thought—provisional, exploratory, open to revision.

HistoryRank is an essay in this original sense. It is not an effort to definitively answer the question "who were the most important people in history." Such a question may not have a meaningful answer. Instead, it is an experimental platform for exploring how different intelligences—human and artificial—approach this unanswerable question.

The value lies not in the rankings themselves but in what the exercise reveals: about model behavior, about cultural bias, about the hidden assumptions that shape how we think about the past.

An Invitation

We offer HistoryRank in two spirits. First, as an unconventional but hopefully useful benchmark for researchers thinking about AI interpretability, alignment, and cross-cultural bias. The dataset of model explanations, the patterns of agreement and disagreement, and the comparisons across models may prove valuable for understanding how LLMs reason about complex value-laden questions.

Second, as a starting point for students of history interested in the grand sweep of the human past. Who were the people who shaped the world we live in? What did they do? How do different sources—academic historians, Wikipedia readers, artificial intelligences—weigh their contributions? These are questions worth exploring, even if they have no final answers.

We make no claim to have produced "the" list of history's most important figures. We have produced several such lists, from different sources, using different methods, and we have made the patterns of agreement and disagreement visible. What you do with that is up to you.