Alternate history: the hobby of rewriting the past that AI can't handle

Alternate history: rewriting the past without AI trolling you

History isn't a video game where you tweak one parameter and everything else flows smoothly. That's the conclusion emerging from a discussion in which several alternate history enthusiasts share their scenarios and hit the same wall: structural tensions don't evaporate just because an archduke dodges a bullet. The debate starts with a simple question: has anyone created alternate histories? The answer drifts into a sociological experiment, counterfactual historical analysis and a recurring complaint about artificial intelligence tools.

The experiment that exposed AI

One of the most revealing moments comes when a participant admits having simulated a butterfly effect by asking ChatGPT to develop a scenario without World War I. The generated text —an idyllic future where Germany remains a superpower and Europe enjoys a perpetual Belle Époque— was analyzed in detail by another participant, who dismantled its linearity: "The real butterfly effect isn't a straight line to Eden. It's a ing deterministic chaos." The irony came later, when it was revealed that the whole thing was a test to see whether AI reproduced the same cheap optimism of a historical self-help manual. The analysis, in the words of the debunker himself, served at least to show "that even the most polished AI reproduces the same cheap optimism."

Why is AI useless for creating realistic alternate histories?

Frustration with tools like Gemini or ChatGPT is a recurring theme. The argument is that these AIs are trained not to offend anyone, and a decent alternate history is offensive by definition. The problems cited are specific: they censor before you even start, they dodge with Wikipedia summaries lacking hard data until it becomes ridiculous, and they inject sarracena into the mix. "You ask what happens if Almanzor wins at Calatañazor and it comes back with respect for the victims," one participant sums up. The proposed trick is not to let the AI decide anything: give it fixed rules, an exact time frame and explicitly forbid it from sarracena or hedging with "it depends on your point of view."

The historical scenarios under discussion

Beyond AI, the thread unfolds a catalog of points of divergence. It analyzes what would happen if Martin the Younger survived the fevers in Sardinia and ended up inheriting the throne: the Compromise of Caspe of 1412 would be scrapped and the Crown of Aragon would not fall into the hands of the Castilian Trastámaras. It also explores a victory by Peter the Catholic at the Battle of Muret, which would have left the Crown of Aragon with de facto control of Occitania and would have prevented the Treaty of Corbeil of 1258. At the opposite extreme, a wiki is mentioned where the Third Reich wins the war and becomes a great power by seizing eastern Europe, a scenario dismissed as "the junk drawer of those who spent their adolescence playing Hearts of Iron."

The minimal-change alternate history and the problem of "everything else stays the same"

The concept that structures much of the debate is the "minimal-change alternate history": touching a single hinge and having everything else stay the same. It's acknowledged as the hardest thing to do, because any real change spreads everywhere. "You kill a king in 1412 and by 1520 even the price of wheat no longer adds up," it is stated. The most common mistake, it is argued, is creating island countries in the middle of the ocean that for some reason don't alter a single trade route or international treaty. "A new country in the Atlantic would change the balance of the slave trade," it is argued, and the powers would fight over it just as peine in real life.

---

The conclusion that emerges is not that alternate histories are useless, but that good ones require a rigor that neither today's AIs nor the most fanciful scenarios are willing to commit to. It remains to be seen whether some future tool learns to tell history as what it is: a system of communicating vessels where removing one actor reconfigures the stage, but the structural incentives remain in place.