Interview
The Head of the Unicorn Sciences Department
Data scientist Hagit Perry, a child prodigy and the first female programmer recruited to Unit 8200, wants to teach the next generation of entrepreneurs to turn AI-powered models into companies worth billions
10:4007.02.20
On the wall of data scientist Hagit Perry’s office at the Interdisciplinary Center (IDC) Herzliya hangs a moto: in God we trust, all others must bring data. Tongue in cheek, yes, but that is the rule that has guided Perry throughout her career.
Perry was a child prodigy. She started her BSc in computer science at the age of 17, was the first woman accepted to a programming course at the Israeli Defense Forces’ (IDF) Unit 8200, known as the country's NSA equivalent, and went on to receive a PhD from Berkeley’s school of business. Today, at the age of 36, younger than some of her students, she heads an MBA program for artificial intelligence and big data at IDC.
Hagit Perry. Photo: Orel Cohen
“This program will birth unicorns, no doubt about it,” Perry told Calcalist in a recent interview. The program, she explained, is in actuality a startup incubator. “We had a student that had investors knocking on his door the moment he left the lab. All he did was take data from emails exchanged by sales personnel and input it into Salesforce’s customer service management system, generating data. Imagine what he could produce with data from other sources.”
Perry has headed the program since it opened in 2016. Today, 76 students are enrolled, working on 22 projects between them. One project, for example, examined why a third of patients skip their pre-scheduled doctor appointments, no matter how many reminders they receive or the severity of their medical issue. The students created AI-powered algorithms that surveyed a wide range of data, including age, clinic location, medical specialty, and the way the meeting was scheduled and found that the main factor that determines whether people make their appointments or not was the duration that passed between when the appointment was made and when it occurred. The longer they have to wait, the bigger the chance people will skip an appointment. Other projects tested why people do not switch to more healthy products despite campaigns—high prices are usually the answer, but even a tiny discount will fix reluctance—and how nationality affects the choice of hotels (Israelis place a high value on interior design, Australians search for proximity to bars).
Companies and organizations are willing to shell out a lot of money for this kind of information, and that is where AI comes in. It is pretty obvious how one turns a multitude of data into one clear conclusion, but as Frida Polli, CEO of behavioral neuroscience and AI technology company Pymetrics, said it best: AI is like teenagers having sex—everyone is saying they're doing it, but no one actually has any idea what they are doing.
That is the goal of IDC’s AI.MBA program—to teach students how to turn amorphic ideas into applicable ones and how to answer business questions using big data algorithms of their own making. The choice of placing the program under the institute’s business school rather than the computer science faculty is intended to attract people with engineering capabilities that have a business-oriented mindset, Perry said.
Currently, the students are working on presentations for pre-seed investors, set to come and survey the various projects in April. Unlike most programs, the final score a project receives matters less than the amount of money these potential investors will bet on it.
AI is one of the biggest buzzwords of the past few years. At the 2020 World Economic Forum in Davos, held January 21-24, tech giants like Google and Microsoft placed AI at the top of their agenda. Google CEO Sundar Pichai stated humanity is more preoccupied with AI than with fire or electricity, and McKinsey & Company estimated AI applications will contribute $13 trillion to the global economy by 2030.
Intuition and the ability to extrapolate is inherent to most people, but computers need to be taught that ability. A two-year-old, for example, only needs to be shown a photo of a cat once before they will probably be able to identify a cat the next time they see one. A computer, however, needs to process around a thousand different cat images to learn the difference between a cat and a sheep, which also has a hairy coat and four legs. That is the intersection between big data, machine learning, and AI: the more computing power a computer has, the more raw data (cat pictures) it can process, the better it can learn to make generalizations (what do all cats have in common), and the faster it learns to identify anomalies (this is a sheep, not a cat). The answers to such generalization processes are what we talk about when we talk about AI technology. Turning those generalizations into actionable forecasts—now that is the holy grail of AI.
AI technology is improving fast. If, in 2017, a computer needed over 10 days to process and identify 1000 pictures with the same level of accuracy as the human eye, by mid-2019 the same task took less than three minutes. And still, the use of AI to improve decision making in the medical or business industry is at its initial stages. Businesses make use of AI aspects only on the fringes of their operations and for basic actions, because there are not enough people with the expertise to operate those systems, Perry said.
In the future, Perry believes, AI will extend to many aspects, such as marketing, where creative directors will be replaced by data specialists. For that reason, alongside the startup incubator, Perry has established a data lab for external projects commissioned by government organizations and companies like Intel, Teva Pharmaceutical Industries Ltd, and Amdocs Inc.
Perry’s love for data dates back to the coding summer camp she attended in middle school. Having graduated high school at the age of 17, she spent the gap year before her mandatory military service working as a programmer at a startup. The army was so impressed with her achievements that her recruitment was pushed back so that it could change the rules and recruit her into a top programming unit that, until then, had only recruited men. Another notable entrepreneur that served in the same unit after it began recruiting women is Kira Radinsky, formerly eBay Israel chief scientist and director of data.
Perry’s military service launched her interest in AI. After her release from the army, she worked at Israeli IT company Aman Group, one of the first Israeli companies to utilize predictions, entered university, and was handpicked by Dov Moran, the co-founder and CEO of USB flash drive developer M-Systems, for his new company Modu.
The year was 2006, and Moran was the superstar of Israeli tech, having just sold his company M-Systems to SanDisk Corp. for $1.6 billion. Modu developed a new modular phone concept, a basic “naked” phone on which various applications could be added, and Perry was hired as a product manager. “In startups, the product manager is, in essence, the CEO, and my objective was to predict how a smartphone will look before one actually existed. I was Dov’s right hand and he gave me a lot of freedom,” she said
Modu was founded in 2007, on the eve of the financial crisis of 2008. Despite the circumstances, Moran managed to raise $120 million, but his gamble proved unfruitful: Apple’s iPhone was released to the market in June 2007, and the rest, as they say, is history. Modu shut down in 2010.
Modu’s main problem was that intersection between a global recession and competition with a tech giant like Apple, which necessitated financial reserves the company just did not have, Perry said. “It was the shattering of a dream, not of money but of zionism—to create another giant invention with Dov after the disk-on-key,” she said. The experience taught her to trust her own instincts, she said—Modu tried to go too big, and despite doing the necessary market research, she folded in the face of older and more experienced people.
Despite Modu’s failure, there were several silver linings—such as the fact that Perry registered eight patents that were later sold to Google, or that Modu’s ex-employees went on to found over 30 tech companies. Perry herself banded with some army friends to found the ultimately unsuccessful tigiGo, an online shopping platform that utilizes recommendations from social networks for a more personalized experience.
“We had two clients, but raising funds during the recession was hard, and we were also too young,” Perry explained. “I was 25 and we made small mistakes, but it was enough to fail at fundraising.”
Perry decided to return to school and was accepted to a PhD program at Berkeley. It was at this point, when she was taking a stop from being an entrepreneur and had just given birth, that the turning point came. Her doctorate became a breakthrough study about the way AI could be integrated into product pricing smart models. “In economics, you try to model human behavior, and optimize company operations based on that,” she said. “I tried to do it using computer models.”
Her model used machine learning to figure out what makes people buy a certain product, “out of an understanding that human behavior in this context is based mainly on inertia, that is, our tendency to keep buying the same brand.” Her case study was the private coffee brand of U.S. supermarket chain Safeway, which tried to take on heavyweights like Starbucks. She built a system that helps plan pricing in the long-term and relied on more than basic data like sales made at a specific date or price. “Data does not tell the whole story, as it is blind to the psychological aspects of an acquisition,” she said. AI, on the other hand, can identify something the human eye cannot, she explained—for example, that people tend to forget about a product they sampled once during their next shopping trip, even if they enjoyed it.
Her study, she said, proved that despite the belief that loyalty steers choices in products like coffee brands, in truth, price outplays loyalty, which is why consumer preference can be altered via pricing. When introducing a new brand the original pricing should be similar to premium brands, but in actuality, you should seldom sell at that price, she said. “You need to give many many discounts and yell ‘only today, only now at half-price,’” she said. “Even though it goes against the intuition and the message of many marketing guides, it worked.”
Perry’s model turned her into a persona grata not just in academia but also in the business sector, and earned her a reputation in the AI sector. But Perry made the choice to stay in academia and train the next generation, instead of setting up her own business again. “I heard many tech entrepreneurs say that more knowledge and a better academic background would have saved them many mistakes. That is why I was excited about opening an academic program for the new reality—it was an opportunity to connect both worlds into one useful framework.”
