Black Swan. How many times have you heard of Black Swan? Unfortunately it has nothing to do with Natalie Portman, but it is a much more “absolute” term. It describes an extremely unlikely event, but which if it occurs could cause gigantic upheavals. Two examples above all? The 2008 financial crisis and… everything we have seen since 2020.
By definition, no one sees a black swan coming, otherwise what kind of black swan would it be? Unpredictable. Point. But Stanford researchers are not the type to stop at the first (nor the hundredth) difficulty, and for this reason they are trying to change things. They are building a computational method to try to predict when the next “unpredictable” event will occur.
Can we predict a black swan?
“This work is exciting because it is an opportunity to take the knowledge and computational tools we are building in the lab and use them in reality. To better understand (and even predict) what is happening in the world around us,” he says Bo Wang, assistant professor of bioengineering at Stanford and senior author of the study.
Published in PLOS Computational Biology, the method is based on natural systems, and could be useful in environmental and health research. (Applications in other fields with black swan events, such as economics and politics, may come soon after.)
“Existing forecasting methods rely on past data to predict future data,” Wang says. “And that's why they tend to predict the predictable, not the unpredictable like a black swan.” The new method, inspired by researcher Sam Bray, who works in Wang's lab, inserts an unknown element into the equation. It assumes we're only seeing part of the world, and tries to figure out what's missing.
The science of the unpredictable
Bray had been studying microbial communities for years and in that time he had observed some events where a microbe exploded in the population, eliminating its rivals. Bray and Wang wondered if this could also happen outside the laboratory and, if so, if it could be predicted.
To find out, the two not only needed to find ecological systems in which this black swan had already occurred, but these systems would also have to have enormous and detailed amounts of data, both on the events themselves and on the ecosystem.
For the development of the method, three data sets from natural systems were chosen: measurements of algae, barnacles and mussels on the Kiwi coast taken monthly for 20 years; levels of plankton of the Black Sea taken twice a week for eight years; and a Harvard study he carried out carbon measurements net from a forest every half hour since 1991.
The researchers processed all this data using statistical physics. Specifically, they used models developed for avalanches and other natural systems with short-term, extreme, and unannounced physical fluctuations—the same qualities that distinguish a black swan event. Taking that analysis, they developed a method for predicting a black swan event.
A predictor of black swan! It works?
The method is meant to be open to variables such as species and time scale, allowing it to work even with lower quality data. Armed with fragments that showed only minimal variations, the method accurately predicted the black swan event. It worked, yes.
Wang and Bray hope to expand this “predictor,” bringing it into other fields where a black swan can occur: economics, epidemiology and physics. The work joins a burgeoning field of artificial intelligence algorithms and computational models geared towards extreme events, including those intended to predict wildfires, assist in marine search and rescue, and optimize emergency response.
