At a Microsoft executive retreat during his first month as chief executive officer, Satya Nadella saw a research project that captured his attention. The demonstration in February 2014 used speech recognition and artificial intelligence to translate a live conversation into another language. Nadella told the team he wanted the tool combined with Skype and ready in time to show off at his first public speech three months later.
This is not how Microsoft typically works. As Nadella, a 24-year veteran of the company, would have known, the process of turning a Microsoft Research project into a product would often happen slowly, if at all. That’s partly by design. The company’s research group was set up in isolation from the product teams to allow researchers to envision the future without worrying about how their inventions will make money or fit into the company’s mission.
But Nadella’s tight deadline left executives with no time to debate the separation of church and state. “We did not have a formal team working on this when he made that statement,” said Lilian Rincon, the Skype group program manager. So they assembled one and immediately went to work on what would eventually become Skype Translator.
Without Nadella’s direct intercession, the translation work might have amounted to little more than a talking point among academics. While Skype Translator would set an important precedent for the company, relying on the CEO to personally vet every lab project isn’t a sustainable business plan. That’s why Microsoft is overhauling its research arm and the way it works with the rest of the company. The goal is to quickly identify technology with the most potential and get it into customers’ hands before a competitor replicates it.
To break down the walls between its research group and the rest of the company, Microsoft reassigned about half of its more than 1,000 research staff in September 2014 to a new group called MSR NExT. Its focus is on projects with greater impact to the company rather than pure research. Meanwhile, the other half of Microsoft Research is getting pushed to find more significant ways it can contribute to the company’s products.
Besides Skype, other services that have benefited from the recent transformation include cloud productivity tools in Office, faster and more power-efficient servers running Bing, and the augmented-reality headset HoloLens. The latest to come out of this initiative is a new feature for Cortana. Microsoft plans to release an update to the digital assistant on Monday that relies on work from the corporate research group. It will give Cortana the ability to scan e-mails for tasks the user has agreed to accomplish and automatically set reminders to do them.
Microsoft is in a race with Google and Facebook to establish the strongest hold over people’s digital lives. The changes at Microsoft Research resemble how its younger Silicon Valley rivals have operated for years. “Microsoft totally separated its research arm from the rest of the company and almost made it optional to contribute to the rest of the company,” said Ahmad Abdulkader, an engineer on the applied machine learning team at Facebook who previously worked at Microsoft and Google. “Google took the exact opposite approach.”
Google researchers work very closely with product groups, and almost everything they produce is visible to the rest of the company, said Jeff Dean, a senior fellow at Google. “We don’t have this really isolated group that is doing stuff without any regard to what might be useful in products,” he said. “We have this very porous connection between research and products.”
Researchers and developers on the search engine or Gmail teams share many of the same tools, including the company’s open-source AI framework TensorFlow, Dean said. That kind of close collaboration has helped produce impressive features, including Smart Reply, which suggests e-mail responses based on the content of a message. The feature, released in November 2015, was based on about a year of AI research at Google. Once the company decided to deploy the tech in Google’s Inbox app, it took around four months to produce a prototype, said Jason Freidenfelds, a Google spokesman.
A similarly ambitious effort at Facebook, aimed at developing a conversational AI assistant called M, was born from research that began in 2014. The company published a paper on its work in October 2014, and by summer 2015 the tech was ready to be tested in Facebook Messenger. Alex Lebrun, who leads development of Facebook M, meets weekly with the company’s top AI researchers to explore which lab developments are ready for the world. Facebook employees can track research-in-progress using a tool named FBLearner Flow. It lets them access, copy, and adapt the source code, and then deploy their own versions of the software, said Abdulkader, the Facebook machine learning engineer. “This is how experiments are exposed and shared.”
Facebook Chief Technology Officer Mike Schroepfer said the cooperation between its research labs and those working on the social network is an effective recruiting tool. “The promise I made to all the artificial intelligence folks that joined us is we’re going to be the best place to get your work to a billion people as fast as possible,” Schroepfer said during an event last year at the company’s headquarters in Menlo Park, Calif.
One potential drawback of this approach is it could encourage scientists to ignore projects that don’t have obvious financial potential. Each company is trying to strike a balance to avoid this breed of short-term thinking. For instance, Facebook assigns some staff to focus on long-term research, and Google’s DeepMind group in London conducts pure AI research without immediate commercial considerations.
The bigger problem at Microsoft was that often the most promising research never made its way into the company’s products until a rival built something similar. Jim Gray, the late Microsoft Research scientist and A.M. Turing Award winner, designed one of the first modern digital mapping programs in the late 1990s. Microsoft co-founder Bill Gates showed off Gray’s TerraServer to applause at a conference in 1998 but didn’t do much with it after that. When Google Maps debuted in 2005, Gates ordered the company to build its own version in 100 days.
Gates and former CTO Nathan Myhrvold started Microsoft Research in 1991 amid the decline of onetime tech powerhouses Bell Labs and Xerox PARC. They hired Rick Rashid from Carnegie Mellon University to create a lab in the school’s image, hiring bright minds and letting them do their thing. “The bad news was they have an internal technology transfer problem,” said Ed Lazowska, a computer science professor at the University of Washington who has been on the lab’s advisory board since its founding.
Today’s geniuses often prefer impact over independence, which is partly why Microsoft is rethinking its approach. “The product groups are now eager because whatever they’ve been doing hasn’t been working,” Lazowska said. “Part of Satya’s mantra is, ‘We need to be open to new ideas, and Microsoft Research is where they will come from.'”
Jeannette Wing, a Microsoft Research vice president, pitched Lazowska’s students in a speech last month on how the company’s changes make it an attractive place to work. “We have enjoyed a stellar reputation in academia because of our scientific impact,” she said. “Now there’s an emphasis to have as much company impact as we’ve had scientific impact.”
The new reminders feature in Cortana was the product of a series of regular meetings between top researchers and product heads. Marcus Ash, a group program manager for Cortana in Windows, said his team is working with researchers on expanding the feature to track when someone else asks the user to take on a job. “There’s been a more deliberate focus on leveraging some of the best ideas from Microsoft Research to enhance Microsoft services and products,” said Eric Horvitz, managing director of Microsoft’s research group who worked with Ash on Cortana.
These changes won’t happen immediately, but Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence in Seattle who sold a company to Microsoft in 2008, said the company is well-positioned to make the transition. “You just can’t turn it on a dime. It’s a process,” he said.
A command from the CEO certainly helps. The team behind Skype Translator managed to prepare a prototype in time for Nadella’s onstage demonstration at a Re/code tech conference in May 2014. “I would have never believed we could move that fast,” said Vikram Dendi, who helped bring the product to life. Today, Skype Translator is available in seven languages, and Dendi is the strategy director for Microsoft Research. It’s his job to keep track of MSR NExT projects and identify ones that are ready for prime time. The Skype and research teams now talk every day.