Inside the new Microsoft, where lie detection is a killer app

Microsoft was early to machine learning but is now competing with Google and Amazon to commercialize the technology. Photo / Bloomberg

Jennifer Marsman recently tested a lie detector of her own design on her boss at Microsoft. Do you work for the best company in the world, she asked. Yes. Oops! According to the software, that was probably a fib. Will she get a promotion this year? Yes! This time her manager was most likely telling the truth.

No, Microsoft isn't getting into the law-enforcement game. Marsman, 37, is a "principal developer evangelist," whose job is to tirelessly advocate for machine learning-a form of artificial intelligence that uses data to make predictions about everything from quarterly sales to when a cow will get pregnant.

The lie detector, cobbled together from algorithms and a 14-pronged headset that measures brain waves, is a kind of party trick Marsman deploys to show software developers how to use Microsoft's Azure Machine Learning tools.

Boisterous and known to spout Harry Potter references, Marsman plays a crucial role for a company that was early to machine learning but is now competing with Google and Amazon to commercialize the technology.

The stakes are high. In the coming years, machine learning will change the world-making computers exponentially smarter and helping companies cut costs, figure out where to invest and a whole lot more. Bloomberg Intelligence analyst Anurag Rana describes the technology as "one of the biggest differentiators for software companies for years to come." Without it, he says, "you're not going to be able to sell your product."

The reaction of many people there was 'We know how to do things, why are you questioning my views with your data'.

Though Microsoft has been working on machine learning for at least 20 years, divisions like Office and Windows once harnessed its predictive qualities only sparingly. "The reaction of many people there was 'We know how to do things, why are you questioning my views with your data,'" says Pedro Domingos, a University of Washington computer science professor who wrote a book on machine learning called "The Master Algorithm."

Microsoft truly embraced the technology when it started Bing in an attempt to catch up with Google. Satya Nadella ran engineering and technical strategy for the search division before becoming chief executive officer two years ago and has been sprinkling machine learning like fairy dust on everything his company touches.

"Microsoft is now in this place where they have machine learning very deeply embedded," Domingos says. "They're investing a lot in making machine learning less Wild West."

Amy Hood is a big fan of this. She can sleep nicer knowing that a machine learning model predicted her quarter.

Like Google and Amazon, which have both used the technology to improve their own products, Microsoft is weaving machine learning into its own operations. This isn't simply about helping the company save money and function better; the more Microsoft uses the technology itself, the easier it is to explain and sell.

"Customers are confused," says Joseph Sirosh, lured from Amazon in 2013 to oversee engineering for Microsoft's machine learning efforts. "Cutting through that noise has been a bit of a challenge. It has been also hard for our own field and sales people to go talk to customers and educate them about all the use cases."

CFO Amy Hood's finance department has come to rely on algorithms-using them to help forecast sales and how many licenses the company will sell in a given period. "It turns out to be very, very accurate for that application," Sirosh says. "Amy Hood is a big fan of this. She can sleep nicer knowing that a machine learning model predicted her quarter."

Microsoft also uses algorithms to predict how many servers it needs to buy for its rapidly expanding datacenters and to help salespeople predict which clients to focus attention on. Even the company's older products, like the accounting software it acquired in 2002, are getting a machine learning facelift, Sirosh says. Microsoft's Cortana Analytics Suite lets customers build some of these tools in-house.

Ram Shankar Siva Kumar, 25, is a data cowboy-he chose his own title-on the Azure Security Data Science team. He uses machine learning algorithms to predict suspicious behavior on Microsoft's networks. Security teams are already pretty good at finding an attack once they know what they're looking for; Kumar has to find them before anyone has that information.

To train his algorithms to recognize malicious behavior, he feeds them actual attacks by Microsoft's Red Team of internal hackers paid to break into the company's networks as well as threat reports coming in from Microsoft's security center. That allows him to build models that can recognize real exploits.

How do people find out that Microsoft has really cool machine learning capabilities that they can use? They've got to get their story out there.

All kinds of companies and industries are already using Microsoft technology.

Japanese farmers are tracking cows, which walk more when ready to conceive, so they can inseminate the cow at the optimal moment. An Australian wine company is using similar algorithms to predict grape yields. A hospital about an hour from Microsoft is using Azure tools to help figure out which cardiac patients are most likely to require re-admission. Norway's eSmart Systems uses Azure Machine Learning to forecast energy grid usage and turn down home heating when demand is high.

Matt McIlwain, a managing director at Madrona Venture Group in Seattle, says Microsoft's machine learning is as good or better than rival technology. But he says Microsoft still suffers from the perception that it's playing catch-up. "How do people find out that Microsoft has really cool machine learning capabilities that they can use?" he says. "They've got to get their story out there."

That's where Jennifer Marsman comes in. She travels the world demonstrating her lie detector and sparking conversations about potential uses for machine learning. Medical applications come up a lot. People have asked about using the technology to predict seizures, monitoring the elderly at assisted living facilities and deciding whether a football player injured during a game should go to hospital or back onto the field.

Says Marsman: "I have the coolest job in the company."