Kira Radinsky’s first brush with prophecy did not go well. The year was 1990, she was four and the oracle was her mother, who decided that the family should leave Russia for Israel. The question was when. The first Gulf War was starting to simmer and pundits suspected that Iraqi President Saddam Hussein would try to drag Israel into the fray. Her mother spent months doing calculations and evaluating scenarios so the family would move only after the war was over. In the event, they landed in the Holy Land two weeks before the first Scud missile from Baghdad. A gas mask was among her first memories of Israel, Radinsky laughs.
Today, Radinsky, 30, has made millions from predictive analytics, after eBay bought SalesPredict, the startup she helped found four years ago. It’s based on humans repeating their patterns, noticing these patterns, and then producing probabilities that the patterns will repeat.
Predictable human behavior and math helped Radinsky and colleagues at Microsoft identify patterns that may be sociologically obvious but that people seeking patterns wouldn’t have thought of.
For instance, they found a pattern when tracking an Ebola variant: Every time there was a volcanic event or quake in certain parts of Africa, people would seek gold or diamonds there; they would deforest the area, expelling animals from their habitat – including the bats carrying that specific Ebola variant.
Tracking dozens of cases of genocide, Radinsky’s system saw another pattern. If a major figure in a country that is characterized by some small degree of friction calls a minority derogatory names like “cockroaches” or “rats,” the chance of a future genocide is greater than 50%. (Sociologists explain that such insults are the first step in dehumanization.)
Probability of violence in Sudan
Today, her company aims to predict what people will buy based on their history of orders. But the potential applications are much wider. The latest wrinkle on Radinsky’s mind, for example, is medical applications.
“Access to information today is practically unlimited,” she tells TheMarker. “All that remains is to let the computers scan it for patterns and structures. If a structure repeats itself a few times, there’s a probability that the sequence of events will repeat itself again. If we identify a chain of events at an early stage of the sequence, we can predict – based on the past – how it will end and, if necessary, act to change the future.”
What Radinsky does is take information that’s available online, combine it with more information from databases, and then apply her algorithms. What she gets is probabilities. Among other things, she says she predicted a cholera outbreak in Cuba, and rioting in Sudan during the Arab Spring. How?
“We noticed a repeating pattern in countries where the people are poor but the nation is rich in resources, like Sudan,” she says (the African state has hydrocarbon reserves, among other things). “We noticed that in these countries, canceling subsidies provokes students to riot, and if the deterioration isn’t stopped, events can end in violent clashes.”
When Sudan canceled its gas subsidies, the SalesPredict system calculated a high probability of riots, which did ensue – starting with students and then spreading.
Much the same happened in Egypt when bread subsidies were canceled. However, they didn’t have enough information about the country to make predictions then, Radinsky says.
Online auction site eBay bought SalesPredict in order to identify surfers’ shopping preferences, leading them to buy more items and to help traders improve their inventory management.
Breakthrough from boredom
Joining the Technion – Israel Institute of Technology, Haifa, at 15, Radinsky considered a career in biology, but proved too clumsy. “I ruined research going back years by contaminating samples of prostate cancer cells,” she mourns.
Then she was given a task that didn’t involve touching things: counting cells through a microscope. Tens of thousands of them. Bored out of her skull, she wrote software to count the things and went to the pool instead – and achieved a more accurate result than if she had counted manually, she points out.
Nowadays, every bit of information has a price, and big commerce sites like eBay will pay for it. The problem is analyzing the data and extracting value from it.
Data science has two main approaches: 1) Ask a question: How can I increase the sales of a product based on statistics of selling similar things? 2) Let the system surprise you.
Radinsky likes that: Give the system the data, and let it point out patterns by itself and offer the questions/answers.
The problem is that the system is likely to output thousands of hypotheses, most of which aren’t of interest. It’s up to the researchers to look at these structures and tell the system to focus on the most statistically significant, interesting or relevant, she explains.
After army service – in intelligence – she joined Microsoft, while simultaneously completing her master’s degree at the Technion. This gave her unlimited access to information and technology. Researching how to leverage the web and human knowledge for the purpose of prediction (for her doctoral studies), Radinsky recalls: “We began with scanning the entire New York Times archive and adding any other available information we could – articles, books, content from social media, people’s searches from search engines, anything available. What we got from the system was an illogical number of patterns and structures. We could hardly focus on anything. We chose to focus on areas that interested us.” This is how her starting research wound up focusing on disease and regional rioting.
How can one take everything mankind knows (in writing) and find repeating patterns to predict the future? Time is a profoundly problematic thing online, Radinsky says.
“Search engines are considered dynamic, constantly updating with more content – but in practice they’re static,” she notes. Say you’re Googling “Passover.” If you do that three months before the holiday, you’re probably looking for its dates. Two weeks ahead, you may be looking for holiday activities for the kids. And on the day itself, you’re probably looking for information on how to arrange the table. “Users’ needs change with timing. One of my missions was to anticipate needs in advance,” says Radinsky.
And so they integrated an algorithm into a search engine that can predict the connection between search and time, and show relevant results.
The first economic application from her system was its warning that iPad prices were likely to rise because of a tsunami that flooded a manufacturing plant in China, causing a shortage.
Ultimately, she and her partner, Yaron Zakai-Or, decided to create a system that would help companies predict the probability of doing business with another firm. That probability (deal/no deal) is microeconomic, but it’s related to elements like developments in the industry, other companies in the field, and so on.
35% chance of fighting with your spouse
Here’s the future, according to Radinsky: Instead of getting up in the morning and reading news on our smartphones or listening to the radio, we’ll look at a board with results of parameters like “work,” “family,” “financial status,” “state of health,” etc. Each parameter is given a grade. For instance, in health care, you might see a probability for a 9% increase in blood pressure because of stress at work. In “Personal relationships,” you might see a 35% probability of a fight with your spouse, based on patterns the system recognizes.
In fact, the future is already here. Insurance companies are already doing this very kind of thing, Radinsky points out. Banks give you credit ratings; HMOs send you cancer checkup warnings. The problem, she adds, is that they’re often inaccurate because the information is incomplete, so one makes decisions based on inaccurate ratings and opinions.
“We have to treat information like a natural and economic resource,” she states. “If we say that data is the new oil, we have to leverage it – because Israel is one of the leading countries in the world in collecting and analyzing data of all types, especially medical data.” And all in compliance with the law, she adds.
So, as an oracle, what does she see as the next big thing? The man-machine interface. And better access to data. “It would change my whole life.” The baby steps are tools like Siri and virtual reality.
Final question: You seem to like to flirt with the media.
“Me? You started it.”
Want to enjoy 'Zen' reading - with no ads and just the article? Subscribe todaySubscribe now