Why you should care
Oracles have guided human behavior for thousands of years. Now there’s some science behind it.
She thought it had to be a glitch. Still, in 2012, Cuba experienced a severe cholera outbreak, the first of its kind in 130 years. And while it surprised nearly everyone, a petite 27-year-old Israeli named Kira Radinsky says she saw the signs — thanks to software she built. She says it can foresee events months into the future.
Is there anything harder than predicting when a country will break out into civil unrest, or knowing where the next pandemic will strike? Based in the seaside city of Haifa, Radinsky may have cracked some of the code. Along with Eric Horvitz, director of Microsoft Research’s main lab, she’s created software that trolls reams of Web data — everything from archived news reports to real-time tweets — to predict future geopolitical events, claiming a 70 to 90 percent accuracy rate. Listed as one of MIT Technology Review’s 35 Innovators Under 35, Radinsky is now trying to parlay all this into a new company aimed at helping businesses predict consumer behavior.
In her world, nothing is random. “History tends to repeat itself,” she says.
In layman’s terms, the software predicted Cuba’s cholera troubles by ferreting out an obscure pattern: Elsewhere and in the past, severe droughts that came two years after a hurricane had led to outbreaks. Other eerily accurate predictions include student protests in Egypt and riots in Sudan. She found patterns from countries with rising gross domestic product, growing poverty and loss of subsidized products. In her world, nothing is random. “History tends to repeat itself,” she says.
Of course, prediction is a proven hazard. Just ask the weatherman, or seismologists or, for that matter, economic forecasters, whose reams of data and supercomputers don’t seem to help much. And then, data-based prediction has a serious limitations: It’s only as good as the data. “An algorithm will never be able to make predictions about what no one is reporting on, and there are flagrant gaps in news coverage,” says David Lazer, professor of political science and computer science at Northeastern University. Paradigm-changing events with no precedent — like the rise of the Islamic State or the spread of Ebola — are beyond the algorithm’s range, too. Professor Shaul Markovitch, Ph.D. supervisor at Technion, the Israel Institute of Technology, calls Radinsky’s work unique — and risky. “No one had even tried before her,” he says. “I would have discouraged her from going in that direction.”
Radinsky tells OZY her work just automates and makes prediction more efficient. “I’m not doing anything that a human analyst couldn’t do,” she says. Sitting in her office near Tel Aviv, Radinsky shifts uncomfortably in her chair and speaks with a speed and conviction that makes everything sound simple. Radinsky enjoys risk. She’s a karate black belt, loves extreme sports and deeply admires Elon Musk, the billionaire CEO of SpaceX, co-founder of PayPal and CEO of Tesla Motors. Emigrating from Ukraine to Israel at the age of 4, she grew up with a mathematician mother and a software engineer aunt who taught her — at age 5 — how to write a command to pass a video game level. “I remember thinking, ‘That’s much easier than actually playing!’ ” she says, laughing. A decade later she began her bachelor’s at the Technion. At 20, she won the prestigious Israel Defense Prize, after working for the country’s secret service.
Radinsky’s passion for prediction came about partly thanks to her husband, also a computer scientist and her best friend since the age of 8. She was showing search peaks on Google Trends to him when he said that what would be truly interesting was predicting the next peak. “I thought, ‘Yes! That would be much more interesting,’ ” recalls Radinsky.
… there wasn’t much point to be warning about cholera or riots if … that country didn’t have the money to do something about it.
— Kira Radinsky
Radinsky’s innovation was to develop the algorithm to tap into data never before available and then create useful applications. The algorithms need to spot correlations and then compare them with similar instances in different contexts, all of which requires using colossal volumes of text — from the New York Times’ archives to Wikipedia’s data — while teaching the computer how to identify cause and effect. The answers can be wrong. After the 2011 protests in Sudan, the algorithm said the Sudanese government would fall; it didn’t. “It all comes down to not having enough data,” she says. “For example, we thought a company deal would work, but it didn’t because one of the parties didn’t have enough funds and we just didn’t know it.”
She occasionally collaborates with U.N. agencies and medical nonprofits in Israel. But the hyperactive scientist is now trying to focus on business. “I realized there wasn’t much point to be warning about cholera or riots if the people and companies in that country didn’t have the money to do something about it,” she says. Radinsky is now applying her online premonitory skills to her startup, SalesPredict. The company, which has 20 employees and has raised more than $5 million in funding, promises to help businesses boost revenue by predicting customer behavior. Her competitors include companies such as Lattice Engines and Infer.
Radinsky believes she’s only scratched the surface. More data is the key, from governments, academic institutions and companies. Meanwhile, she’s trying to improve the explanations so the mere human brain can grasp the results.