In April, Conflict Forecast, a project that uses artificial intelligence to predict where outbreaks of violence will occur in the world, noticed a surge in discussion of military activity in Russian media.
About 500 days into the Russia-Ukraine war, levels of media chatter on the conflict were significantly higher than has been normal since the invasion on February 24 last year.
One researcher behind the project said it might have signalled the mutiny by the Wagner paramilitary group, which happened two months later.
Conflict Forecast tracks news events, dividing six million news articles from the past 40 years into economic and security categories to spot historical patterns that showed when countries are on the edge of war.
The AI system, created by professors Hannes Mueller and Christopher Rauh, showed a surge in Russian discussion of conflict several months before the 2014 invasion of Crimea.
It is just one of several cutting-edge projects using AI to spot the next global crisis, crunching down vast amounts of data that would take teams of analysts months to sift through, from satellite imagery to climate data that could predict conflict over land.
In theory, decision-makers armed with this early warning can act promptly. Generals and politicians can assess options from sanctions to a military response, or increase aid and diplomacy to avert the worst, with months instead of weeks or days to prepare.
Stanley McChrystal is a retired general who commanded US forces in Afghanistan and formerly the US Joint Special Operations Command. He is now an advisor at Rhombus Power, a Silicon Valley company that has created Guardian, which it describes as an "artificial Intelligence platform for national and global security."
“All logistics or preparation take time, so the earlier you have high probability or high confidence indicators, you're in a position [to ask] ‘Should I take action?’,” he tells The National.
“The problem is, as we get more and more data, with some clear indications, decision-makers are put in a more difficult position because they can't dodge it. They can't pretend that it's not likely to happen, because the indicators are there.”
But the big problem of prediction is timing. Recent research suggests forecasters are getting much better at predicting where conflict will erupt, but the “when” is still a challenge.
The 'hard problem'
Conflicts that erupt in countries that were previously in a state of relative peace are extremely rare.
They are seen as the “hard problem” for conflict researchers to predict, with June’s near-miss civil war in Russia being one example.
According to Conflict Forecast, that is because about 90 per cent of conflicts begin within two years of a previous conflict ending in any country. It is what researchers call the “conflict trap”, where damage caused by conflict to infrastructure, the economy and human life leads to growing inequality and grievances – factors that lead to more conflict.
In Libya, Sudan and the Democratic Republic of the Congo for example, political, ethnic, tribal and religious fractures across communities have occurred against a legacy of colonialism and dictatorship. Add enduring poverty, foreign intervention and available weapons and the result is a nation crashing from one war to the next.
By contrast, if a country can sustain 10 years of peace after a war, the chance of a new conflict drops dramatically, placing a premium on early intervention to stop violence.
It is these unexpected new conflicts outside of the “conflict trap” countries that can derail a nation’s progress, massively setting back societal development, and they're very hard to predict.
Decades of conflict data outlined below illustrate factors likely to cause conflict, including rival armed factions, sudden economic declines and sharp ethnic or religious divisions, often inflamed by inequality.
Some researchers also point to very specific risk factors. Monty Marshall was one of the first members of the Political Instability Task Force, a CIA-commissioned, multi-decade academic project to identify countries most at risk of a collapse into violence.
“PITF reported that factional competition was found to be a powerful, leading indicator for the onset of political instability, especially for adverse regime changes towards greater autocratic authority but, also, for onsets of political violence,” he tells The National.
Armed with this theoretical framework, recent research has tried to use AI to process vast amounts of data, from rainfall in countries where locals have fought over cattle grazing rights, to tracking government arms purchases or those by militia groups and, in one project, even looking at changing patterns of mobile-phone use.
Predicting Middle East conflict
Henry Wilkinson is the chief intelligence officer at Dragonfly, a risk consultancy.
He agrees that predicting new conflicts in countries that had been relatively peaceful for years is difficult, giving the Arab uprisings in 2010 as an example.
He says he visited some of the affected countries, spoke to people on the ground and made observations to try to work out what was next.
“A lot of people who were sort of Middle East experts quite didn't predict the Arab Spring, they didn't predict the contagion effect of how it swept across the region,” he says.
AI conflict prediction programs would “probably have given us slightly more objective diagnoses of the systems and the nature of those systems, compared to the assessment of a lot of people who were long-standing Middle East watchers”, he says.
“Whereas in objective analysis, the data would say ‘this country is at serious risk’, so there's a subjective bias element,” he adds.
Gen McChrystal agrees. “AI is going to help analysts and people who can use the data to reduce their internal biases,” he says.
Both Gen McChrystal and Mr Wilkinson do not see AI conflict forecasting replacing human experts, but rather, multiplying the power of human analysts.
AI from Iraq to Ukraine
Dr Anshu Roy is the founder and CEO of Rhombus Power, the company that created the Guardian AI platform.
He describes how being able to process years worth of data helped his team assess Russia’s intentions in late 2021, warning of invasion months before it happened, getting as detailed as examining local business transactions where Russian forces were based.
“Guardian was able to pick up certain locations of strategic importance that seemed to be glowing particularly hot in terms of activity and changes,” he says.
“With this discovery, the machine was able to drill down into the depth and breadth of its repository of all-domain real-time information. It is the anomalies at these locations that the AI can get specific about, much faster and with greater precision than ever before.
“In seconds, you're aware, through open source, who are the key individuals that matter in this context, and their presence starts to show up. Examining their purchasing and other patterns of life is the natural next step. The AI-derived insights, once interpreted in context, reveal that folks are doing things that you tend to do when you're preparing for the long haul.”
Gen McChrystal says AI data analysis can also help understand highly complex conflict environments, such as civil wars and insurgencies.
“In Afghanistan, one of the ways we would track Taliban activity in the south was prices of produce in the markets. And that's because Afghanistan lacked refrigeration,” he says.
“So many of their grapes and pomegranates had to be trucked to Pakistan, stored under Pakistani refrigeration and then trucked back in. And the number of Taliban checkpoints, including some that were just run by criminals, that would extract bribes affected the prices in the markets.
“There are just countless examples of this kind of information that are very hard to collect and to collate and to make sense of. But this gets to the heart of what AI has some potential with.”
Gen McChrystal says having this capability today could help spot trends in crisis-hit countries, sifting through thousands of signals to understand political currents.
This could prove vital for conflict forecasters because one of the difficult things to predict, in addition to new conflict, is changes in conflict intensity, from a manageable situation to a major regional crisis.
“I think back to 2005 in Iraq, probably the low point of the conflict, we were cutting edge at that point, because we were gathering – particularly signals intelligence – more than ever before. And we had very rudimentary capability to link it together and to derive conclusions. If we could have done this with AI, we could have first understood that [enemy] network.
“But much more broadly, we could have understood what was happening in Iraq in a way we just didn't. We knew it was terrible, but it was difficult to make sense of the different forces and things that were happening across Iraq. I believe that the capability is already at the point to make much greater sense of a similar crisis today.”
Mr Wilkinson points to a similar need to judge the effects of conflicts as they occur, for example, considering the impact of the Ukraine war on commodities and supply chains.
“If this scenario were to occur, what would we see in terms of the impacts and dealing with the bewildering complexity of all the different permutations of things? And if this were to happen, would the stock market shift up or down and how does it impact supply chains? You can start to link all the dependencies together.”