The new artificial intelligence can predict future major disasters.

Predicting dangerous tipping points in complex systems has long been a headache for scientists. Now, a new AI system may be ready to do this work for them.

Artificial intelligence can help predict catastrophic tipping points

Researchers in computer science have created a new artificial intelligence (AI) program that can predict the occurrence of catastrophic tipping points – such as ecological collapse, major financial crises, pandemics, and global blackouts. “If a looming critical transformation can be predicted, we can prepare for its impacts or even prevent the transition, thereby mitigating the damages,” said Gang Yan, lead author of the study and a computer science professor at Tongji University in China, to Live Science.

“This prompted us to develop an AI-based approach to predict the onset of such sudden transitions, even before they occur,” added the professor. The researchers published their findings on July 15 in the journal Physical Review X. Tipping points are sudden shifts where a localized system or its environment transitions to an undesirable state that is difficult to reverse. For example, if the Greenland ice sheet were to collapse, it would reduce snowfall in the island’s northern region, drastically increase sea levels, and irreversibly disappear a large part of the ice sheet.

Starting from tipping points in simple theoretical systems

The science that investigates the mechanisms of dramatic transformations is often based on oversimplified models, making accurate predictions difficult. Previously, scientists used statistics to assess the declining strength and resilience of systems based on detectable, increasing fluctuations.

However, the results of studies using these statistical methods were contradictory.

To achieve a more accurate way of predicting dangerous transitions, the researchers behind the new study combined two different types of neural networks or algorithms that mimic the way the brain processes information.

The first broke down complex systems into large networks of interacting nodes before tracing connections between nodes; the second modeled the temporal changes of individual nodes. “For example, in financial systems, a node could represent a single company; in ecological systems, a node could represent a species; while in social media systems, a node could represent a user, and so on,” Yan said.

Since tipping points are difficult to predict, as is where to look for them, real data on sudden critical transitions is sparse. Instead, the researchers focused on tipping points within simple theoretical systems, such as ecosystem models, where there is enough time to map out signs of changes leading to the tipping point.

The new AI algorithm’s prediction turned out to be accurate

Once their neural network gathered enough data, the researchers gave the artificial intelligence a real-world problem: the transformation of tropical forests into savannas. They collected over 20 years of satellite data from three regions in Central Africa where this sudden transition occurred, then provided information about rainfall and tree cover to the algorithm, selecting two regions for modeling.

From this data, the artificial intelligence accurately predicted what happened in the third region,

even though 81% of the nodes of the systems (in this case, the impacted land areas) went unnoticed, the researchers said. After successfully predicting a tipping point, the researchers are now looking for a method to detail the perceived patterns with the algorithm. They hope to apply their model to other systems, such as forest fires, pandemics, and financial collapses.

Human systems pose the greatest challenge

One of the major challenges in predicting systems that involve humans is how we respond to them and whether our predictions feedback into our behavior, the researchers write. “For example, consider urban traffic: while identifying congested roads may be straightforward, disseminating real-time congestion information to all drivers can easily lead to chaos,” Gang said. “This is because drivers can immediately change their routes in response to the information, which may reduce congestion in some roads but cause it elsewhere.

Such dynamic interactions make prediction particularly complex,” said the computer science professor at Tongji University. The researchers believe that to avoid such problems, it is advisable to focus on parts of human systems that seemingly do not affect our intentions. In the case of road networks, this can be done by focusing on the routes that are considered busier due to their fundamental design, rather than the behavior of drivers. The application of artificial intelligence may be a highly valuable tool for creating more accurate crisis prediction scenarios than ever before.