Ask Onix
AI and the quest to foresee history's turning points
For centuries, sudden events-from a defenestration in Prague to a fruit seller's protest in Tunisia-have reshaped the world. Now, researchers are turning to artificial intelligence to detect the warning signs before the next crisis erupts.
Historical sparks and unforeseen wildfires
On 23 May 1618, Protestant nobles in Prague hurled two Catholic governors and their secretary from a castle window, an act that ignited the Thirty Years' War. The governors survived-either by divine intervention, as Catholics claimed, or by landing in a dung heap, according to Protestants. What seemed a minor feud spiraled into a continent-wide catastrophe, killing millions through war, famine, and disease.
History brims with such moments: a misworded press conference hastening the fall of the Berlin Wall, a wrong turn in Sarajevo triggering World War I, or a Tunisian street vendor's self-immolation sparking the Arab Spring. Each event carried clues, but identifying the spark before it ignites remains elusive.
Data-driven crystal balls
Pitirim Sorokin, a Russian-American sociologist, pioneered a data-centric approach in the early 20th century to explain societal collapses. He quantified instability by tracking micro-events like assassinations and macro-events like revolutions, arguing that Rome's fall stemmed from decadence and overindulgence.
Today, Peter Turchin's team at Oxford University's World History Lab builds on Sorokin's work. Their database spans 80,000 historical data points, from the Bronze Age collapse to the Habsburg Empire's dissolution. Using computational models, they've identified patterns in revolutions: a toxic mix of economic decline, elite overproduction, and state fiscal weakness. Turchin's 2010 prediction of 2020's chaos-fueled by a "dark triad" of social strains-proved eerily accurate amid the pandemic and political turmoil.
While the team currently uses AI for data collection, they aim to integrate machine learning into predictive modeling. "Algorithms could enhance our mathematical models," says research assistant Jakob Zsambok.
Military and governments lead the charge
Defense and intelligence agencies are already deploying AI to anticipate conflicts. In 2020, the U.S. military's Raven Sentry AI analyzed historical violence, weather data, and social media to predict Taliban attacks in Afghanistan with 70% accuracy-matching human analysts but at greater speed. Rhombus Power, a defense contractor, claims its generative AI forecasted Russia's invasion of Ukraine using satellite imagery and missile site activity, though these predictions weren't publicly verified.
The United Nations uses AI to assess disaster impacts, such as estimating damage after Afghanistan's 2023 Herat earthquake. Its Crisis Risk Dashboard combines real-time and historical data to flag potential hotspots, monitoring hate speech in Sri Lanka and migration patterns elsewhere.
Financial markets and the limits of prediction
Stanford finance professor Antonio Coppola leverages AI to map financial vulnerabilities. His model, trained on $40 trillion in shadow banking data, accurately forecasted 2020's liquidity crisis by identifying which investors would sell assets en masse. While 10 times more effective than traditional methods, Coppola stresses AI should complement-not replace-economic theory.
The Alan Turing Institute, however, urges caution. "AI struggles with fragmented data on conflicts like the Arab Spring," says senior researcher Anna Knack. Her report concludes AI's best near-term use lies in tracking risk indicators and modeling outcomes after shocks occur.
AI's double-edged sword
Eugene Chausovsky of the New Lines Institute uses AI to simulate crises, like blockades in the Strait of Hormuz, assessing ripple effects on energy and agriculture. His team's AI-driven simulations even include bots role-playing as world leaders-though they tend to avoid escalation, unlike their human counterparts.
Yet AI itself may pose risks. Economists warn of an AI-driven financial bubble, while tech leaders caution about societal disruption. When asked to estimate the odds of AI causing a global crisis this century, ChatGPT pegged the probability at 20-40%, while Gemini called it a "50/50 toss-up." Claude declined to speculate.
The future: between hype and hope
For now, AI excels in low-stakes forecasting tournaments, where startups are climbing leaderboards. But predicting revolutions or wars remains a work in progress. As Chausovsky notes, "Understanding ripple effects is almost as valuable as knowing when a disaster will strike."
"If we can see the warning signs, we might mitigate the damage-even if we can't stop the storm."
Samantha Holder, Oxford University World History Lab
