For over a decade, Ada Palmer, a history professor at University of Chicago (and a science-fiction writer!), struggled to teach Machiavelli. “I kept changing my approach, trying new things: which texts, what combinations, expanding how many class sessions he got…” The problem, she explains, is that “Machiavelli doesn’t unpack his contemporary examples, he assumes that you lived through it and know, so sometimes he just says things like: Some princes don’t have to work to maintain their power, like the Duke of Ferrara, period end of chapter. He doesn’t explain, so modern readers can’t get it.”
Palmer’s solution was to make her students live through the run-up to the Italian Wars themselves. Her current method involves a three-week simulation of the 1492 papal election, a massive undertaking with sixty students playing historical figures, each receiving over twenty pages of unique character material, supported by twenty chroniclers and seventy volunteers. After this almost month-long pedagogical marathon, a week of analysis, and reading Machiavelli’s letters, students finally encounter The Prince. By then they know the context intimately. When Machiavelli mentions the Duke of Ferrara maintaining power effortlessly, Palmer’s students react viscerally. They remember Alfonso and Ippolito d’Este as opportunists who exploited their vulnerabilities while remaining secure themselves. They’ve learned the names, families, and alliances not through memorization but through necessity: to protect their characters’ homelands and defeat their enemies.
Then, one year, her papal election class was scheduled at the same time as a course on Machiavelli’s political thought. The teachers brought both classes together, so each could hear how the other’s class (history vs. political science) approached the things differently. Palmer asked both classes: “What would Machiavelli say if you asked him what would happen if Milan suddenly changed from a monarchal duchy to a republic?”
The poli sci students went first: He’d say that it would be very unstable, because the people don’t have a republican tradition, so lots of ambitious families would be tempted to try to take over, so you’d have to get rid of those ambitious families, like the example Livy gives of killing the sons of Brutus in the Roman Republic, and you would have to work hard to get the people passionately invested in the new republican institutions, or they wouldn’t stand by them when the going gets tough or conquerors threaten. It was a great answer. Then my students replied: He’d say it would all depend on whether Cardinal Ascanio Visconti Sforza is or isn’t in the inner circle of the current pope, how badly the Orsini-Colonna feud is raging, whether politics in Florence is stable enough for the Medici to aid Milan’s defenses, and whether Emperor Maximilian is free to defend Milan or too busy dealing with Vladislaus of Hungary. “And I think I’d have something to say about it!” added my fearsome Caterina Sforza; “And me,” added my ominously smiling King Charles. In fact, my class had given a silent answer before anyone spoke, since the instant they heard the phrase, “if Milan became a republic,” all my students had turned as a body to stare at our King Charles with trepidation, with a couple of glances for our Ascanio Visconti Sforza. It was a completely different answer from the other class’s, but the thing that made the moment magical is that both were right.
Both answers were right, but they hinted at different kinds of approaches to history. The political science students articulated general principles, the structural forces that make new republics unstable, the institutional work required to sustain them. Palmer’s students, by contrast, gave an answer saturated with particulars: specific cardinals, specific feuds, specific rulers with specific constraints. They weren’t describing general laws but a turbulent moment where small differences — whether Ascanio Sforza is in the pope’s inner circle, whether Maximilian is busy with Hungary — could deflect the course of events in radically different directions.
From a grand perspective, Palmer’s students’ insights may seem irrelevant. In physics, after all, particulars do not matter. Whether two molecules bump into each other doesn’t affect the overall thermodynamic state of a steam engine. Yet in the historical context, things are different. Because you yourself are one of those molecules and you care greatly about whom you bump into. Whether Ascanio Sforza is in the pope’s inner circle matters, because it can determine whether your city will be sacked and your family killed.
Inventing the Renaissance ranges widely across Renaissance history, historiography, and ethics. The simulated papal election is but one of many topics, but it raises an interesting question Palmer doesn’t directly address: how do you study history when particulars determine outcomes but those outcomes remain fundamentally unpredictable? Her students aren’t learning to predict what happened. They’re learning something else entirely. Understanding what that “something else” is reveals not only why her experiment succeeds, but how it reshapes historical methodology.
***
Palmer’s simulation transforms students into the political actors of Renaissance Italy. Some play powerful cardinals wielding vast economic resources and influence networks, with strong shots at the papacy. Others are minor cardinals burdened with debts and vulnerabilities, nursing long-term hopes of rising on others’ coattails. Locked in a secret basement chamber, students play the crowned heads of Europe, the King of France, the Queen of Castile, the Holy Roman Emperor, smuggling secret orders via text messages to their agents in the conclave. Still others are functionaries: those who count the votes, distribute food, guard the doors, direct the choir. They have no votes but can hear, watch, and whisper.
Each student receives a character packet detailing their goals, personality, allies, enemies, and tradeable resources: treasures, land, titles, holy relics, armies, marriageable nieces and nephews, contracts, and the artists or scholars at their courts. “I’ll give you Leonardo if you send three armies to guard my city from the French.”
The simulation runs over multiple weeks. Students write letters to relatives, allies, rivals and subordinates. If you write to a player, the letter will be delivered to that person and will advance your negotiations. If you write to a non-player-character, you will receive a reply which will also affect the game.
Palmer designed the simulation as alternate history, not a reconstruction. She gave each historical figure resources and goals reflecting their real circumstances, but deliberately moved some characters in time so that students who already knew what happened to Italy in this period would know they couldn’t have the ‘correct’ outcome even if they tried. That frees everyone to pursue their goals rather than ‘correct’ choices. She set up the tensions and actors to simulate the historical situation, then left it run its course.
The simulation captures how papal elections were never isolated events. While cardinals compete for St. Peter’s throne, the crowned heads of Europe maneuver for influence. In the Renaissance, Rome controlled marriage alliances and annulments, could crown or excommunicate rulers, distributed valuable benefices and titles, commanded papal armies. The pope’s allies shifted the political balance to their benefit and rose to wealth and power while enemies scrambled for cover.
War usually breaks out after the election. “Kings are crowned, monarchs unite, someone is invaded,” Palmer writes, “but the patterns of alliances and thus the shape of the war vary every year based on the individual choices made by students.”
Palmer has run the simulation many times. Each time certain outcomes recur, likely locked in by greater political and economic forces. The same powerful cardinals are always leading candidates. There’s usually a wildcard candidate as well, someone who circumstances bring together with an unexpected coalition. Usually a juggernaut wins, one of the cardinals with a strong power-base, but it’s always very close. The voting coalition always powerfully affects the new pope’s policies and first actions, determining which city-states rise and which burn as Italy erupts in war.
And the war erupts every single time. And it is always totally different.
Sometimes France invades Spain. Sometimes France and Spain unite to invade the Holy Roman Empire. Sometimes England and Spain unite to keep the French out of Italy. Sometimes France and the Empire unite to keep Spain out of Italy.
Once students created a giant pan-European peace treaty with marriage alliances that looked likely to permanently unify all four great Crowns, only to be shattered by the sudden assassination of a crown prince.
***
The assassination of that crown prince is telling. In this run of Palmer’s simulation, a single student’s decision — perhaps made impulsively, perhaps strategically — eliminated what looked like an inevitable unification of Europe. A marriage alliance that seemed to guarantee peace for generations evaporated in an instant. One moment of violence redirected the entire course of the simulation’s history. Small things matter.
Or as Palmer herself puts it: “The marriage alliance between Milan and Ferrara makes Venice friends with Milan, which makes Venice’s rival Genoa side with Spain, and pretty soon there are Scotsmen fighting Englishmen in Padua.”
This is the pattern that emerges from repeated runs: certain outcomes seem inevitable (a powerful Cardinal wins the papacy, war breaks out), but the specific path history takes turns on moments like these, moments where a single action cascades into consequences no one could have foreseen.
Palmer’s students aren’t learning to predict outcomes. That would be impossible in a system where a single assassination can shatter a continental peace. They’re learning something else: how to navigate a world where small causes can have large effects, where the direction of those effects remains unknown until they unfold.
***
This is what scientists call sensitive dependence on initial conditions, more popularly known as the butterfly effect. A small perturbation, the flutter of a butterfly’s wings, the assassination of a prince, can cascade into enormous consequences through chains of causation impossible to foresee.
Stand beside a river and watch the water flow. In some stretches it moves smoothly. Cast a twig into the flow and it drifts peacefully downstream. The water follows predictable patterns. This is what physicists call laminar flow. It’s orderly and stable and small disturbances quickly dissipate.
But look downstream where the river narrows to meets rocks. The water churns and froths. Whirlpools form and dissolve. Sometimes you feel like you recognize a pattern but no two whirlpools are ever exactly the same. Drop a twig at this place and you cannot predict where it ends. It might circle three times and shoot left, or catch an eddy and spiral right, or get pulled under and pop up twenty feet downstream. Small differences in exactly where and how it enters produce completely different paths. This is turbulence.
And this is what chaos theory studies. It looks at turbulent system and asks: What exactly can we say about it? What predictions are possible when prediction seems impossible? And given that history flows very much like a river — with political science studying its laminar aspect and Palmer’s students learning to navigate the turbulent moments — what can chaos theory teach us about history?
Well, not much, as it turns out. At least not directly.
Chaos theory was everywhere in the 1990s. Fractals adorned dorm room posters. Jurassic Park explained the butterfly effect to moviegoers.
Then chaos theory largely disappeared from public discourse. Not because it was wrong, the mathematics remains valid, the phenomena real, but because it proved remarkably difficult to apply. A recent survey of commonly cited applications by Elizabeth Van Nostrand and Alex Altair found that most “never received wide usage.”
The theory excels at explaining what cannot be done. You cannot make long-range weather predictions. You cannot predict where exactly a turbulent eddy will form. You cannot forecast the specific trajectory of a chaotic system beyond a certain time horizon. These are important insights, but they are negative and thus non-sexy. They tell us about the limits of prediction, not how to make it better.
So if chaos theory mostly tells us what we cannot do with turbulent systems, what use is it for understanding history?
The answer comes from the one domain where chaos theory achieved genuine practical success: weather forecasting. But not in the way anyone expected.
In the 1940s, when computers first made numerical weather prediction possible, the approach was deterministic: measure current conditions, run the physics forward, predict the future. But by the late 1950s, cracks appeared: a single missing observation could cause huge errors two days later. Then came Lorenz’s 1961 discovery: rounding 0.506127 to 0.506 caused his weather simulation to diverge completely, proving that precise long-range forecasts were impossible.
Chaos theory explains why long-range deterministic forecasting fails. But it doesn’t tell you what to do about it.
It took thirty years to achieve a breakthrough. It came from changing the question. Instead of asking “What will the weather be ten days from now?” ask instead what it could possibly be. Run the model not once, with your best-guess initial conditions, but many times, with slightly different starting points that reflect measurement uncertainty. Each run produces a different forecast. Together, they map the range of possible futures.
This is ensemble prediction. Instead of a single forecast, you generate an ensemble of forecasts. If all ensemble members agree, confidence is high. If they diverge into different patterns, uncertainty is high. You cannot predict which specific future will occur, but you can map the probability distribution across possible futures.
Since becoming used in practice in the early 1990s, the results have vindicated the approach. Ensemble forecasts consistently outperform single deterministic forecasts. They provide not just predictions but measures of confidence. They reveal when the atmosphere is in a predictable state (ensemble members cluster together) versus a turbulent one (ensemble members diverge widely).
Ensemble prediction doesn’t defeat chaos, it works along with chaos. It accepts that specific trajectories cannot be predicted beyond a certain horizon, but reveals that the distribution of trajectories can be. It’s a fundamentally different kind of knowledge: not “it will rain Tuesday” but “there’s a 70% chance of rain Tuesday, with high uncertainty.”
***
Palmer’s papal election simulation exhibits exactly the same structure, though she arrived at it independently and for different reasons.
Each run of the simulation starts from the same historical situation. The date is 1492. There are the same cardinals with the same resources, the same European powers with the same constraints. But Palmer populates these roles with different students, each bringing their own judgment, risk tolerance, and strategic thinking.
Run the simulation once and you get a history: one specific pope elected, one specific pattern of alliances, one specific set of cities burned. Run it ten times and a pattern emerges that no single run could reveal: certain outcomes consistently occur (a powerful cardinal wins, war breaks out, Italian city-states suffer) while others vary widely (which specific cardinal, which specific alliances, which specific cities). The simulation generates not a single counterfactual but a probability distribution across possible 1492s.
What emerges is a probabilistic model of the political situation of 1492. Not “Florence will be sacked” but “Florence survives in 70% of runs.” Not “France will invade” but “French intervention occurs with near certainty, though the target varies.” This is the kind of knowledge ensemble prediction provides. Not certainty about specifics, but clarity about the shape of the possible.
Interestingly, Palmer has independently arrived at both major methods weather forecasters use for ensemble prediction, though for entirely different reasons.
For one, she perturbs the initial conditions by moving historical figures in time. Cardinals who never overlapped now competing for the same throne, creating configurations that never actually existed. And she also runs multiple models: each time different students inhabit the same roles, bringing different judgment and risk tolerance. One student playing Cardinal della Rovere might ally with France; another might seek Spanish protection. Same constraints, different decision-making models.
Palmer developed these techniques for pedagogical reasons, to prevent students from seeking ‘correct’ answers and to explore the range of human responses, but the result is structurally identical to what meteorologists spent decades developing to work around chaos.
***
Military planners have long grappled with the same problem. Wargaming exists because commanders cannot predict how battles will unfold. Chaos, friction, and human decision-making make deterministic prediction impossible. But unlike meteorologists, military planners lack the resources to run true ensemble predictions. A major wargame is expensive, it involves hundreds of personnel and equipment over weeks and a single scenario can be executed once, rarely twice.
History, we are told, is more like wargaming than meteorology or physics. We cannot do experiments. What happens, happens once. There is no going back to try different initial conditions. There is no way to rerun 1492 with different actors to see how it plays out.
But Palmer’s approach suggests otherwise. Experimental history is possible. Not in the sense of manipulating the past, but in the sense of systematically exploring its possibility space. Her simulation is an experiment: controlled conditions, repeated trials, emergent patterns. It will never achieve the precision of physics, but it’s a genuine advance beyond purely descriptive history, as we know it.
The limitation is obvious: Palmer can run her simulation perhaps ten times over the years she teaches the course. But what if we could run fifty simulations per day, as weather forecasters do? What if we do that for an entire year? We’d end up with tens of thousands of simulations and a detailed probabilistic landscape of the political situation of 1492.
Enter history LLMs, large language models trained exclusively on texts from specific historical periods!
The idea emerged from a fundamental problem: modern LLMs cannot forget. A generic LLM knows what already happened. No amount of prompting can remove this hindsight bias, which, by the way, it shares with Palmer’s students. A historian studying the Renaissance cannot un-know what came next, and neither can a model trained on Wikipedia.
But what if you could train an LLM only on texts available before a specific date? Researchers at the University of Zurich recently built Ranke-4B, a language model trained exclusively on pre-1913 texts.
“The model literally doesn’t know WWI happened.” It reasons like someone from 1913 would have reasoned, with 1913’s uncertainties and 1913’s assumptions about the future. It doesn’t know that Archduke Franz Ferdinand will be assassinated. It doesn’t know about tanks or poison gas or the collapse of empires.
Due to the scarcity of texts, it probably won’t be possible to train a 1492 history LLM. But a 1913 one is clearly possible. So what does that mean?
Can we run simulations of the July Crisis? Populate the roles with LLM agents trained on pre-1913 texts, Kaiser Wilhelm, Tsar Nicholas, British Foreign Secretary Edward Grey, Serbian Prime Minister Pašić, and watch ten thousand versions of 1914 unfold? Would we see the Great War emerge in 94% of runs, or only 60%? Would we find that small changes, a different response to the Austrian ultimatum, a faster Russian mobilization, a clearer British commitment to France, consistently deflect the trajectory toward peace, or do they merely shift which powers fight and when?
These aren’t idle questions. They go to the heart of historical causation. Was the Great War inevitable, locked in by alliance structures and arms races and imperial rivalries? Or was it contingent, the product of specific decisions made under pressure by specific individuals who might have chosen differently? Historians have debated this for a century. Palmer’s simulation suggests a new approach. Don’t argue, simulate. Map the probability distribution.
But this raises a deeper question. Given the butterfly effect, can actors in chaotic systems achieve their goals at all? If small perturbations cascade unpredictably through chaotic systems, then perhaps historical actors are merely throwing pebbles into turbulent water, creating ripples they cannot control, in directions they cannot predict. They perturb the system, yes, but with unknown and unknowable consequences.
Palmer argues otherwise. Her students don’t just perturb the system at random. They achieve goals. Not perfectly, not completely, but meaningfully. As she observes: “No one controlled what happened, and no one could predict what happened, but those who worked hard [...] most of them succeeded in diverting most of the damage, achieving many of their goals, preventing the worst. Not all, but most.” Florence doesn’t always survive, but when Florentine players work skillfully, it survives more often. The outcomes aren’t predetermined, but neither are they purely random.
This is what Machiavelli asserted. In The Prince, Chapter XXV, he writes:
I compare [Fortune] to one of those violent rivers, which when swelled up floods the plains, sweeping away trees and buildings, carrying the soil away from one place to another; everyone flees before it, all yield to its violence without any means to stop it. […] And yet, though floods are like this, it is not the case that men, in fair weather, cannot prepare for this, with dikes and barriers, so that if the waters rise again, they either flow away via canal, or their force is not so unrestrained and destructive.
The flood comes, but prepared actors can channel it. They cannot choose whether it occurs, but they can influence where it flows, which fields it devastates, which cities it spares. Fortune, Machiavelli concludes, “is arbiter of half our actions, but still she leaves the other half, or nearly half, for us to govern.”
Experimental history, as outlined above, could test whether Machiavelli’s metaphor actually describes how history works. If history is pure chaos, if human action makes no predictable difference, then skilled and unskilled players should succeed equally often. But if Machiavelli is right, patterns should emerge. Players who build strong alliances, maintain credible threats, balance powers, and manage debts carefully should protect their homelands statistically more often than those who don’t. Not always, not with certainty, but measurably. The flood still comes, but the dikes matter.
And if patterns emerge, experimental history then becomes a laboratory for learning what works. Which kinds of dikes prove most effective? Does early coalition-building outperform late negotiation? Do transparent commitments work better than strategic ambiguity? The specific tactics of Renaissance cardinals won’t apply to modern crises, but the principles might: How to protect vulnerable positions between great powers, when commitments under pressure hold or collapse? What distinguishes successful from failed crisis management?
Palmer stumbled onto this through pedagogy, meteorologists developed it through necessity, historians and political scientists might adopt it to learn how much we can actually govern within the half that Fortune leaves us, and how to govern it well.
This is a cross-post from https://www.250bpm.com/p/ada-palmer-inventing-the-renaissance.
Papal election of 1492
For over a decade, Ada Palmer, a history professor at University of Chicago (and a science-fiction writer!), struggled to teach Machiavelli. “I kept changing my approach, trying new things: which texts, what combinations, expanding how many class sessions he got…” The problem, she explains, is that “Machiavelli doesn’t unpack his contemporary examples, he assumes that you lived through it and know, so sometimes he just says things like: Some princes don’t have to work to maintain their power, like the Duke of Ferrara, period end of chapter. He doesn’t explain, so modern readers can’t get it.”
Palmer’s solution was to make her students live through the run-up to the Italian Wars themselves. Her current method involves a three-week simulation of the 1492 papal election, a massive undertaking with sixty students playing historical figures, each receiving over twenty pages of unique character material, supported by twenty chroniclers and seventy volunteers. After this almost month-long pedagogical marathon, a week of analysis, and reading Machiavelli’s letters, students finally encounter The Prince. By then they know the context intimately. When Machiavelli mentions the Duke of Ferrara maintaining power effortlessly, Palmer’s students react viscerally. They remember Alfonso and Ippolito d’Este as opportunists who exploited their vulnerabilities while remaining secure themselves. They’ve learned the names, families, and alliances not through memorization but through necessity: to protect their characters’ homelands and defeat their enemies.
Then, one year, her papal election class was scheduled at the same time as a course on Machiavelli’s political thought. The teachers brought both classes together, so each could hear how the other’s class (history vs. political science) approached the things differently. Palmer asked both classes: “What would Machiavelli say if you asked him what would happen if Milan suddenly changed from a monarchal duchy to a republic?”
Both answers were right, but they hinted at different kinds of approaches to history. The political science students articulated general principles, the structural forces that make new republics unstable, the institutional work required to sustain them. Palmer’s students, by contrast, gave an answer saturated with particulars: specific cardinals, specific feuds, specific rulers with specific constraints. They weren’t describing general laws but a turbulent moment where small differences — whether Ascanio Sforza is in the pope’s inner circle, whether Maximilian is busy with Hungary — could deflect the course of events in radically different directions.
From a grand perspective, Palmer’s students’ insights may seem irrelevant. In physics, after all, particulars do not matter. Whether two molecules bump into each other doesn’t affect the overall thermodynamic state of a steam engine. Yet in the historical context, things are different. Because you yourself are one of those molecules and you care greatly about whom you bump into. Whether Ascanio Sforza is in the pope’s inner circle matters, because it can determine whether your city will be sacked and your family killed.
Inventing the Renaissance ranges widely across Renaissance history, historiography, and ethics. The simulated papal election is but one of many topics, but it raises an interesting question Palmer doesn’t directly address: how do you study history when particulars determine outcomes but those outcomes remain fundamentally unpredictable? Her students aren’t learning to predict what happened. They’re learning something else entirely. Understanding what that “something else” is reveals not only why her experiment succeeds, but how it reshapes historical methodology.
***
Palmer’s simulation transforms students into the political actors of Renaissance Italy. Some play powerful cardinals wielding vast economic resources and influence networks, with strong shots at the papacy. Others are minor cardinals burdened with debts and vulnerabilities, nursing long-term hopes of rising on others’ coattails. Locked in a secret basement chamber, students play the crowned heads of Europe, the King of France, the Queen of Castile, the Holy Roman Emperor, smuggling secret orders via text messages to their agents in the conclave. Still others are functionaries: those who count the votes, distribute food, guard the doors, direct the choir. They have no votes but can hear, watch, and whisper.
Each student receives a character packet detailing their goals, personality, allies, enemies, and tradeable resources: treasures, land, titles, holy relics, armies, marriageable nieces and nephews, contracts, and the artists or scholars at their courts. “I’ll give you Leonardo if you send three armies to guard my city from the French.”
The simulation runs over multiple weeks. Students write letters to relatives, allies, rivals and subordinates. If you write to a player, the letter will be delivered to that person and will advance your negotiations. If you write to a non-player-character, you will receive a reply which will also affect the game.
Palmer designed the simulation as alternate history, not a reconstruction. She gave each historical figure resources and goals reflecting their real circumstances, but deliberately moved some characters in time so that students who already knew what happened to Italy in this period would know they couldn’t have the ‘correct’ outcome even if they tried. That frees everyone to pursue their goals rather than ‘correct’ choices. She set up the tensions and actors to simulate the historical situation, then left it run its course.
The simulation captures how papal elections were never isolated events. While cardinals compete for St. Peter’s throne, the crowned heads of Europe maneuver for influence. In the Renaissance, Rome controlled marriage alliances and annulments, could crown or excommunicate rulers, distributed valuable benefices and titles, commanded papal armies. The pope’s allies shifted the political balance to their benefit and rose to wealth and power while enemies scrambled for cover.
War usually breaks out after the election. “Kings are crowned, monarchs unite, someone is invaded,” Palmer writes, “but the patterns of alliances and thus the shape of the war vary every year based on the individual choices made by students.”
Palmer has run the simulation many times. Each time certain outcomes recur, likely locked in by greater political and economic forces. The same powerful cardinals are always leading candidates. There’s usually a wildcard candidate as well, someone who circumstances bring together with an unexpected coalition. Usually a juggernaut wins, one of the cardinals with a strong power-base, but it’s always very close. The voting coalition always powerfully affects the new pope’s policies and first actions, determining which city-states rise and which burn as Italy erupts in war.
And the war erupts every single time. And it is always totally different.
Sometimes France invades Spain. Sometimes France and Spain unite to invade the Holy Roman Empire. Sometimes England and Spain unite to keep the French out of Italy. Sometimes France and the Empire unite to keep Spain out of Italy.
Once students created a giant pan-European peace treaty with marriage alliances that looked likely to permanently unify all four great Crowns, only to be shattered by the sudden assassination of a crown prince.
***
The assassination of that crown prince is telling. In this run of Palmer’s simulation, a single student’s decision — perhaps made impulsively, perhaps strategically — eliminated what looked like an inevitable unification of Europe. A marriage alliance that seemed to guarantee peace for generations evaporated in an instant. One moment of violence redirected the entire course of the simulation’s history. Small things matter.
Or as Palmer herself puts it: “The marriage alliance between Milan and Ferrara makes Venice friends with Milan, which makes Venice’s rival Genoa side with Spain, and pretty soon there are Scotsmen fighting Englishmen in Padua.”
This is the pattern that emerges from repeated runs: certain outcomes seem inevitable (a powerful Cardinal wins the papacy, war breaks out), but the specific path history takes turns on moments like these, moments where a single action cascades into consequences no one could have foreseen.
Palmer’s students aren’t learning to predict outcomes. That would be impossible in a system where a single assassination can shatter a continental peace. They’re learning something else: how to navigate a world where small causes can have large effects, where the direction of those effects remains unknown until they unfold.
***
This is what scientists call sensitive dependence on initial conditions, more popularly known as the butterfly effect. A small perturbation, the flutter of a butterfly’s wings, the assassination of a prince, can cascade into enormous consequences through chains of causation impossible to foresee.
Stand beside a river and watch the water flow. In some stretches it moves smoothly. Cast a twig into the flow and it drifts peacefully downstream. The water follows predictable patterns. This is what physicists call laminar flow. It’s orderly and stable and small disturbances quickly dissipate.
But look downstream where the river narrows to meets rocks. The water churns and froths. Whirlpools form and dissolve. Sometimes you feel like you recognize a pattern but no two whirlpools are ever exactly the same. Drop a twig at this place and you cannot predict where it ends. It might circle three times and shoot left, or catch an eddy and spiral right, or get pulled under and pop up twenty feet downstream. Small differences in exactly where and how it enters produce completely different paths. This is turbulence.
And this is what chaos theory studies. It looks at turbulent system and asks: What exactly can we say about it? What predictions are possible when prediction seems impossible? And given that history flows very much like a river — with political science studying its laminar aspect and Palmer’s students learning to navigate the turbulent moments — what can chaos theory teach us about history?
Well, not much, as it turns out. At least not directly.
Chaos theory was everywhere in the 1990s. Fractals adorned dorm room posters. Jurassic Park explained the butterfly effect to moviegoers.
Then chaos theory largely disappeared from public discourse. Not because it was wrong, the mathematics remains valid, the phenomena real, but because it proved remarkably difficult to apply. A recent survey of commonly cited applications by Elizabeth Van Nostrand and Alex Altair found that most “never received wide usage.”
The theory excels at explaining what cannot be done. You cannot make long-range weather predictions. You cannot predict where exactly a turbulent eddy will form. You cannot forecast the specific trajectory of a chaotic system beyond a certain time horizon. These are important insights, but they are negative and thus non-sexy. They tell us about the limits of prediction, not how to make it better.
So if chaos theory mostly tells us what we cannot do with turbulent systems, what use is it for understanding history?
The answer comes from the one domain where chaos theory achieved genuine practical success: weather forecasting. But not in the way anyone expected.
In the 1940s, when computers first made numerical weather prediction possible, the approach was deterministic: measure current conditions, run the physics forward, predict the future. But by the late 1950s, cracks appeared: a single missing observation could cause huge errors two days later. Then came Lorenz’s 1961 discovery: rounding 0.506127 to 0.506 caused his weather simulation to diverge completely, proving that precise long-range forecasts were impossible.
Chaos theory explains why long-range deterministic forecasting fails. But it doesn’t tell you what to do about it.
It took thirty years to achieve a breakthrough. It came from changing the question. Instead of asking “What will the weather be ten days from now?” ask instead what it could possibly be. Run the model not once, with your best-guess initial conditions, but many times, with slightly different starting points that reflect measurement uncertainty. Each run produces a different forecast. Together, they map the range of possible futures.
This is ensemble prediction. Instead of a single forecast, you generate an ensemble of forecasts. If all ensemble members agree, confidence is high. If they diverge into different patterns, uncertainty is high. You cannot predict which specific future will occur, but you can map the probability distribution across possible futures.
Since becoming used in practice in the early 1990s, the results have vindicated the approach. Ensemble forecasts consistently outperform single deterministic forecasts. They provide not just predictions but measures of confidence. They reveal when the atmosphere is in a predictable state (ensemble members cluster together) versus a turbulent one (ensemble members diverge widely).
Ensemble prediction doesn’t defeat chaos, it works along with chaos. It accepts that specific trajectories cannot be predicted beyond a certain horizon, but reveals that the distribution of trajectories can be. It’s a fundamentally different kind of knowledge: not “it will rain Tuesday” but “there’s a 70% chance of rain Tuesday, with high uncertainty.”
***
Palmer’s papal election simulation exhibits exactly the same structure, though she arrived at it independently and for different reasons.
Each run of the simulation starts from the same historical situation. The date is 1492. There are the same cardinals with the same resources, the same European powers with the same constraints. But Palmer populates these roles with different students, each bringing their own judgment, risk tolerance, and strategic thinking.
Run the simulation once and you get a history: one specific pope elected, one specific pattern of alliances, one specific set of cities burned. Run it ten times and a pattern emerges that no single run could reveal: certain outcomes consistently occur (a powerful cardinal wins, war breaks out, Italian city-states suffer) while others vary widely (which specific cardinal, which specific alliances, which specific cities). The simulation generates not a single counterfactual but a probability distribution across possible 1492s.
What emerges is a probabilistic model of the political situation of 1492. Not “Florence will be sacked” but “Florence survives in 70% of runs.” Not “France will invade” but “French intervention occurs with near certainty, though the target varies.” This is the kind of knowledge ensemble prediction provides. Not certainty about specifics, but clarity about the shape of the possible.
Interestingly, Palmer has independently arrived at both major methods weather forecasters use for ensemble prediction, though for entirely different reasons.
For one, she perturbs the initial conditions by moving historical figures in time. Cardinals who never overlapped now competing for the same throne, creating configurations that never actually existed. And she also runs multiple models: each time different students inhabit the same roles, bringing different judgment and risk tolerance. One student playing Cardinal della Rovere might ally with France; another might seek Spanish protection. Same constraints, different decision-making models.
Palmer developed these techniques for pedagogical reasons, to prevent students from seeking ‘correct’ answers and to explore the range of human responses, but the result is structurally identical to what meteorologists spent decades developing to work around chaos.
***
Military planners have long grappled with the same problem. Wargaming exists because commanders cannot predict how battles will unfold. Chaos, friction, and human decision-making make deterministic prediction impossible. But unlike meteorologists, military planners lack the resources to run true ensemble predictions. A major wargame is expensive, it involves hundreds of personnel and equipment over weeks and a single scenario can be executed once, rarely twice.
History, we are told, is more like wargaming than meteorology or physics. We cannot do experiments. What happens, happens once. There is no going back to try different initial conditions. There is no way to rerun 1492 with different actors to see how it plays out.
But Palmer’s approach suggests otherwise. Experimental history is possible. Not in the sense of manipulating the past, but in the sense of systematically exploring its possibility space. Her simulation is an experiment: controlled conditions, repeated trials, emergent patterns. It will never achieve the precision of physics, but it’s a genuine advance beyond purely descriptive history, as we know it.
The limitation is obvious: Palmer can run her simulation perhaps ten times over the years she teaches the course. But what if we could run fifty simulations per day, as weather forecasters do? What if we do that for an entire year? We’d end up with tens of thousands of simulations and a detailed probabilistic landscape of the political situation of 1492.
Enter history LLMs, large language models trained exclusively on texts from specific historical periods!
The idea emerged from a fundamental problem: modern LLMs cannot forget. A generic LLM knows what already happened. No amount of prompting can remove this hindsight bias, which, by the way, it shares with Palmer’s students. A historian studying the Renaissance cannot un-know what came next, and neither can a model trained on Wikipedia.
But what if you could train an LLM only on texts available before a specific date? Researchers at the University of Zurich recently built Ranke-4B, a language model trained exclusively on pre-1913 texts.
“The model literally doesn’t know WWI happened.” It reasons like someone from 1913 would have reasoned, with 1913’s uncertainties and 1913’s assumptions about the future. It doesn’t know that Archduke Franz Ferdinand will be assassinated. It doesn’t know about tanks or poison gas or the collapse of empires.
Due to the scarcity of texts, it probably won’t be possible to train a 1492 history LLM. But a 1913 one is clearly possible. So what does that mean?
Can we run simulations of the July Crisis? Populate the roles with LLM agents trained on pre-1913 texts, Kaiser Wilhelm, Tsar Nicholas, British Foreign Secretary Edward Grey, Serbian Prime Minister Pašić, and watch ten thousand versions of 1914 unfold? Would we see the Great War emerge in 94% of runs, or only 60%? Would we find that small changes, a different response to the Austrian ultimatum, a faster Russian mobilization, a clearer British commitment to France, consistently deflect the trajectory toward peace, or do they merely shift which powers fight and when?
These aren’t idle questions. They go to the heart of historical causation. Was the Great War inevitable, locked in by alliance structures and arms races and imperial rivalries? Or was it contingent, the product of specific decisions made under pressure by specific individuals who might have chosen differently? Historians have debated this for a century. Palmer’s simulation suggests a new approach. Don’t argue, simulate. Map the probability distribution.
But this raises a deeper question. Given the butterfly effect, can actors in chaotic systems achieve their goals at all? If small perturbations cascade unpredictably through chaotic systems, then perhaps historical actors are merely throwing pebbles into turbulent water, creating ripples they cannot control, in directions they cannot predict. They perturb the system, yes, but with unknown and unknowable consequences.
Palmer argues otherwise. Her students don’t just perturb the system at random. They achieve goals. Not perfectly, not completely, but meaningfully. As she observes: “No one controlled what happened, and no one could predict what happened, but those who worked hard [...] most of them succeeded in diverting most of the damage, achieving many of their goals, preventing the worst. Not all, but most.” Florence doesn’t always survive, but when Florentine players work skillfully, it survives more often. The outcomes aren’t predetermined, but neither are they purely random.
This is what Machiavelli asserted. In The Prince, Chapter XXV, he writes:
The flood comes, but prepared actors can channel it. They cannot choose whether it occurs, but they can influence where it flows, which fields it devastates, which cities it spares. Fortune, Machiavelli concludes, “is arbiter of half our actions, but still she leaves the other half, or nearly half, for us to govern.”
Experimental history, as outlined above, could test whether Machiavelli’s metaphor actually describes how history works. If history is pure chaos, if human action makes no predictable difference, then skilled and unskilled players should succeed equally often. But if Machiavelli is right, patterns should emerge. Players who build strong alliances, maintain credible threats, balance powers, and manage debts carefully should protect their homelands statistically more often than those who don’t. Not always, not with certainty, but measurably. The flood still comes, but the dikes matter.
And if patterns emerge, experimental history then becomes a laboratory for learning what works. Which kinds of dikes prove most effective? Does early coalition-building outperform late negotiation? Do transparent commitments work better than strategic ambiguity? The specific tactics of Renaissance cardinals won’t apply to modern crises, but the principles might: How to protect vulnerable positions between great powers, when commitments under pressure hold or collapse? What distinguishes successful from failed crisis management?
Palmer stumbled onto this through pedagogy, meteorologists developed it through necessity, historians and political scientists might adopt it to learn how much we can actually govern within the half that Fortune leaves us, and how to govern it well.