Laurie Clarke, BBC Future / BBC Ukraine
Four hundred years ago in what is now the Czech Republic, making a political statement in a dramatic way involved throwing your enemy out of a window.
On May 23, 1618, in the chancellery of Prague Castle, a group of Protestant nobles accused two Catholic royal governors of ignoring their rights. The heated confrontation ended with the Catholic governors and their secretary being thrown out of a third-story window.
Amazingly, all three survived this defenestration — thanks to the outstretched hands of angels who flew to catch them, according to Catholics. According to the Protestant version of events, their fall was cushioned by a decidedly less glamorous pile of manure.
This scuffle might have remained merely a footnote in the centuries-old history of the long bloody feud between the two Christian sects in medieval Europe. But its consequences proved to be much more far-reaching.
The incident triggered the Bohemian Revolt led by Protestants against the Catholic emperor of the Habsburg dynasty, which escalated into one of the most devastating wars in European history — the Thirty Years’ War. Exhausting three decades of conflict involved over a dozen countries and claimed millions of lives due to widespread destruction, famine, and disease.
This is just one of many examples in history where an event had unexpectedly far-reaching consequences.
History is full of such instances.
An East German official who misspoke during a press conference and triggered a flood of thousands to the Berlin Wall, hastening the end of the Cold War.
The driver of Archduke Franz Ferdinand, who took a wrong turn in Sarajevo and crossed paths with an assassin, igniting a powder keg that led to World War I.
A Tunisian fruit vendor who set himself on fire after police confiscated his scales, sparking the Arab Spring uprisings that spread across six countries and led to the overthrow of four national leaders.
In hindsight, one might say the warning signs were visible. A range of factors can be seen as tinder for a catastrophic event. The problem is knowing in advance which spark will ignite.
However, researchers hope that one day complex artificial intelligence models will be able to predict how and when major global events might unfold.
They believe that with enough data, it will be possible to trace how waves from seemingly minor incidents develop into tides capable of moving financial markets, causing revolutions, or leading to war. Already, AI technologies are providing insights into what might be possible.
Using the past to predict future crises
The idea of predicting the future based on patterns found in the past is far from new. In the first half of the 20th century, sociologist Pitirim Sorokin pioneered a data-driven approach to explain why past empires had dissolved.
To do this, he attempted to quantify social instability across different epochs by gathering data on “micro-events,” such as political assassinations or riots, and “macro-events,” such as civil wars and revolutions.
In the case of Ancient Rome, he used his data to justify what he believed caused the empire’s decline: excessive materialism and hedonism, which led to decadence and “overripeness.”
Today, the spirit of Sorokin’s work is continued by complex systems theory scientist Peter Turchin at the Laboratory for World History at Oxford University in the UK. For over a decade, Turchin and his team of researchers have accumulated 80,000 units of qualitative and quantitative data about societies ranging from the Paleolithic, attempting to explain the past and predict the future.
“We are looking for moments of crisis,” says researcher Samantha Holder, who works on the project.
From the collapse of the late Bronze Age to the breakup of the Habsburg Spanish Empire, crises are assigned scores reflecting their geographical scale and intensity. This data is analyzed for patterns using predictive computational models—a process that currently does not use AI but may start to soon.
Turchin’s team has used their database to hypothesize why moments of crisis arise. They have found that revolutions typically stem from a confluence of factors, including impoverishment of part of the population and an increase in the number of elites competing for a limited number of leadership positions.
“If this occurs simultaneously and the state is experiencing a financial crisis, revolutions and civil wars become more likely,” says Holder, citing the French Revolution as a telling example.
Back in 2010, Turchin applied these methods to predict that the year 2020 would be particularly chaotic. He warned that a period of intense political instability would be caused by a “dark triad” of social ailments: an excess of elites competing for power, declining living standards, and a weak fiscal state. After a global pandemic shook the world economy and a period of acute political turbulence, Turchin seemed prescient.
Now, the team is using AI to aid in collecting and classifying vast historical datasets. But in the future, they hope to use artificial intelligence for forecasting as well.
“Machine learning algorithms… could enhance the mathematical modeling we undertake,” says Jacob Jambok, a research assistant who also works with Turchin. “We are looking in that direction.”
Critics, including the late anthropologist David Graeber, have questioned the idea that history can be used to predict the future. And chance, one-off events like “black swans,” which can trigger periods of upheaval, are by nature impossible to predict. But in many cases, there are warning signals that precede such triggers.
Modeling Chaos
Unsurprisingly, some of those who have paid the most attention to this area so far are governments and the military.
In 2020, a secret US intelligence project used AI called Raven Sentry to predict Taliban attacks in Afghanistan.
According to an article published in the US Army War College journal, the AI tool received data on historical violence in the region along with real-time intelligence, including weather data, social media posts, news reports, and commercial satellite imagery.
The model achieved an accuracy of 70%, roughly comparable to human analysts, “just with much higher speed”.
One of the defense contractors involved in the project, Rhombus Power, claims to have used generative AI to predict Russia’s invasion of Ukraine by analyzing open data, including satellite images, missile site movements, and local business transactions. However, these predictions were not published in advance, so the claims could not be verified.
Other researchers are also developing neural networks aimed at predicting food crises, in some cases using only climate data. However, some scientists remain skeptical about the reliability of AI for such predictions.
For example, the British Alan Turing Institute for AI research assessed the maturity level of AI-based forecasting technologies. Their conclusion? Overall, probably not quite there yet.
“One of the challenges in creating something like this is that it’s not easy to get the right training data for AI to predict future conflicts,” says Anna Knaak, a senior researcher at the Turing Institute specializing in national security who conducted the analysis.
“The problem is that when we think about things like the Arab Spring, 9/11, Iran or Kashmir, all this information is scattered across fragmented sources within the intelligence community,” she adds.
“It’s very difficult to predict even the current thoughts of some of our state leaders,” continues Knaak.
Her report concluded that the two most promising ways AI can help now are more accurately tracking conflict risk indicators and determining possible consequences immediately after a shock occurs.
Almost as useful as knowing when a disaster will occur is understanding the potential ripple effects, says Eugene Chausovsky, senior director at the New Lines Institute in the US, a research and policy analysis center focused on forecasting.
“Where might this crisis echo—not just geopolitically, but also economically?” he says.
Over the past year, Chausovsky and his team have been modeling different versions of the crisis in the Strait of Hormuz, which we are currently experiencing. In partnership with AI startup Mantis Analytics, they used AI to enhance their analyses—assessing secondary impacts on energy markets, semiconductors, and agriculture.
AI tools “allowed us to significantly expand the data streams we work with,” says Chausovsky, from monitoring open sources and global news to “statistical databases from everything—from trade to energy and critical minerals.”
This helps improve the accuracy of the simulations they conduct. They also experimented with involving AI in conflict simulations together with human experts, where bots took turns playing the roles of state leaders.
Now, says Chausovsky, “some of the nuance and complexity that can be at the human level is lost.” Interestingly, AI also tends to be more cautious than humans—for example, refraining from escalatory actions.
The United Nations Development Programme is already using AI to help assess the impacts of major disasters and events. After the earthquake in Herat, Afghanistan, in 2023, it utilized its AI-based Rapid Digital Assessment tool to evaluate the extent of damage and debris at any location, allowing for more accurate targeting of rescue efforts.
The UN is also investing in AI-based early warning systems as part of so-called “proactive crisis management.” They combine historical data and near real-time data on the Crisis Risk Dashboard to identify potential hotspots of violence before the situation escalates.
For example, in Sri Lanka, hate speech and macroeconomic data are monitored, while in other places, they might analyze population movements or migration.
The Next Financial Crash
Financial regulators also hope AI will give them an edge in detecting potential problems. They have access to “exceedingly detailed, essentially real-time data on who owns what across the entire financial system,” says Antonio Coppola, an assistant professor of finance at Stanford University.
This, combined with AI methods like deep learning, can be used to better inform the regulation of financial markets.
Part of the role of a financial regulator involves considering policy interventions to prevent or mitigate financial crises. Instead of predicting the crises themselves, Coppola’s current work focuses on the question: “When this big wave of stress hits, where will the problems be? Who specifically will get into trouble?”
In a recent publication demonstrating the principle’s potential, Coppola created a model trained on a large dataset comprised of financial portfolios encompassing about $40 trillion of wealth in the shadow banking system.
Shadow banks provide services similar to commercial banks but exist outside ordinary financial regulation, accumulating significant financial risks. For example, the shadow banking system contributed to the liquidity crisis during the Covid-19 pandemic.
Coppola found that after training the AI model on 20 years of data up until 2019, it was able to accurately predict which markets experienced the largest sell-offs of financial assets in 2020 and which investors contributed most to the market downturn.
According to Coppola, the results were 10 times better than traditional methods based on economic theory. However, he is quick to add that AI should not replace traditional economic modeling; rather, it can complement it.
Coppola is further exploring how these AI models could include unstructured data, such as news headlines, to increase their accuracy.
Other researchers are already examining how AI can be used to predict financial crises themselves, but this field is still in its early stages.
While it may take several iterations before AI begins accurately predicting crises, it is already making strides in the less risky field of prediction tournaments, where participants wager on the likelihood of various global events from sports to politics occurring. AI startups are gradually climbing leaderboards, although humans currently still hold the top positions.
But there’s a possibility that AI itself could trigger the next global crisis. Many economists are already predicting an “AI bubble” which, if it bursts, could have devastating consequences for financial markets, while technology company executives warn of broader societal upheavals that this technology might cause.
In light of this, we asked AI chatbots ChatGPT, Gemini, and Claude about the likelihood that AI itself might cause a future global crisis. A 2024 study showed that, despite a tendency towards hallucinations — fabricated data — combining predictions from several AI chatbots can achieve accuracy on par with human forecasters.
In response to my queries, Claude refused to name a specific figure, while Gemini rated it as “50-50.” However, ChatGPT estimated the probability that AI “will contribute to a serious global crisis at some point this century” at “approximately 20-40%.” It assessed the likelihood of an existential crisis as less than 5%.
So, for now, it seems we can only wait and see.
Illustration: Coda
