Microsoft’s recent moves in artificial intelligence show the company is serious about being a leading player, not willing to cede AI leadership to other software behemoths.
Microsoft’s 2017 fiscal year annual report listed AI as a top priority, including six references to AI, up from zero in the previous annual report. A vision statement in the report stated, “Our strategic vision is to compete and grow by building best-in-class platforms and productivity services for an intelligent cloud and an intelligent edge infused with AI.”
Since Satya Nadella took over as CEO of Microsoft in 2014, the company has acquired a few AI startups, including Maluuba and SwiftKey, and it has established a formal AI and Research Group. This team “focuses on our AI development and other forward-looking research and development efforts spanning infrastructure, services, applications, and search,” the annual report states.
Microsoft’s vision reset away from mobile and cloud computing comes after Sundar Pichai, CEO of Alphabet’s Google, began saying that the world is shifting from being mobile-first to AI-first, and Facebook has invested in long-term AI research and AI product enhancements.
Microsoft Ventures, itself announced in May 2016, announced a fund dedicated to AI startups in December 2016. Nagraj Kashyap, head of Microsoft Ventures, told TechCrunch, “AI holds great promise to augment human capabilities and improve society. Microsoft is committed to democratizing AI with guiding principles to drive positive impact.”
In January, Yoshua Bengio agreed to become a strategic advisor to Microsoft. A Canadian computer scientist noted for his work in deep learning, Bengio had been courted for months by Microsoft. He continues to work in his office at the University of Montreal. In a recent article in Wired, he explained his reasoning: “We don’t want one or two companies, which I will not name, to be the only big players in town for AI. It’s not good for the community. It’s not good for people in general.”
Harry Shum heads the AI and Research Group, which cuts across Microsoft’s categories of Windows, Office and the Azure cloud initiative. Shum has organized the group into four areas—products, early-stage products, really early-stage products, and research. The hope is “we can accelerate the cycles from research to product” to get AI’s benefits to customers faster, he told Wired.
Late to mobile and cloud computing, Microsoft does not want to miss AI. “In this industry, you’ve got to realize that it’s completely okay if you missed the last wave,” Shum stated. “It’s very problematic if you miss the current wave.”
Microsoft’s answer to Amazon’s Alexa is Cortana, an intelligent personal assistant running across Microsoft platforms. It comes installed with Windows, and thus today has 145 million users, according to Microsoft. Microsoft and Amazon this week announced an intention to have Cortana and Alexa work together, as in “Alexa, call Cortana.” It could be Apple’s Siri is next. How this competition plays out remains to be seen.
Microsoft sees the AI market as early stage, and intends to build AI into products that will come to market in 12 to 24 months. Cortana is tightly integrated with Microsoft’s search engine Bing, which has a sneaky presence by virtue of partnerships with Apple, Amazon, Yahoo, Verizon and AOL. Some 30% of search queries in the US are said to come through Bring.
“This is the reason why Cortana can actually be so helpful and powerful, because we have these data signals from so many devices,” Emma Williams, partner design manager for Cortana, told Wired. “Really, Google is the only other company that could compete with us when it comes to truly understanding the world.”
Microsoft sees Cortana as an agent that has its user’s personal information and can interact on the user’s behalf with Microsoft’s chatbot agents and potentially other agents. Microsoft’s bot framework team is run within the Future Social Experiences Lab (FUSE), along with cognitive services. The FUSE Lab was started by Ray Ozzie, the creator of Lotus Notes groupware, who later worked at IBM and Microsoft via acquisitions, along with Lili Cheng, who had been an engineer at Apple. Cheng was recently promoted to corporate vice president at Microsoft, overseeing FUSE Labs, which produces 29 services including computer vision and voice recognition, that Microsoft makes available to developers.
“We view bots and Cortana conversationally as a product, but it is still an early stage product,” Cheng told Wired.
In the fall of 2016, Microsoft released the chatbot Zo, with the persona of a sassy female teenager. American teens are spending an average of 10 hours talking back and forth with Zo. From this, Microsoft is learning.
Microsoft is retaining its engineers in AI by launching its own classes, and it is extending its relationships, such as by investing in Element AI, an incubator started by Bengio to encourage researchers and entrepreneurs to build AI startups.
Maluuba Acquisition Strategic
In January, Microsoft acquired Maluuba, a Montreal-based company founded by University of Waterloo graduates, focused on deep learning, reinforcement learning and natural language processing. Maluuba was good at improving the ability of computer systems to comprehend what they are reading, to understand natural dialog between individuals and to get better at tasks involving memory, according to an account in TechCrunch.
Late last year, Maluuba release two new sets of data designed to train deep learning algorithms on becoming better at language skills. Deep-learning researchers need huge amounts of data that can challenge an AI to perform certain conversational tasks or comprehension and reasoning tasks. Creating those datasets takes both time and effort.
“The big challenge with deep learning in our space is that because it’s so data driven, the models you end up training are only as complex as the data you train them on,” Adam Trischler, a research scientist at Maluuba, told IEEE Spectrum.
Maluuba released the NewsQA dataset, with more than 110,000 training questions. The startup enlisted the help of human workers through an online crowdsourcing service. One set of workers looked at the highlights from CNN news articles and tried to come up with challenging comprehension questions. A second set of workers tried to answer those questions. A third set help to validate the pairs of questions and answers.
“We found that a large majority of the questions in our dataset do require reasoning beyond the context matching and synonym matching in previous datasets,” Trischler told IEEE Spectrum. “That was our goal, and we achieved that.”
Maluuba also released a second Frames dataset with 1,368 dialogues to help train deep-learning algorithms on conversations. The company invited 12 human volunteers to its Montreal Lab, where they engaged in online chat conversations. One person pretended to be a customer booking a vacation; the second person pretended to be a travel agent consulting a database with information on hotels, flights and destinations.
The conversations showed that people frequently went back-and-forth on different vacation possibilities. The dialogue requires the computer system to retain a memory of the different possibilities in order to make comparisons. The new publicly-available Frames dataset challenges deep-learning algorithms to have the memory hold this natural conversation that goes back and forth on hotels, flights and destinations without following a specific order.
Maluuba is also working with a researcher at McGill University in Montreal to train an AI system to take on the Winograd Schema Challenge, a test designed to determine how well an AI system can handle common sense reasoning. An example challenge question is: “I tried to put my computer inside the briefcase, but it was too small.” The system needs to figure out whether the briefcase or the computer is too small.
“The Winograd Schema Challenge is all about common sense,” Trischler told IEEE Spectrum. “The reason we see that as something very important is because that goes hand-in-hand with the machine comprehension we’re working on.”
SwiftKey Smart About Typing
In February 2016, Microsoft acquired SwiftKey, offering software blending AI technologies to enable it to predict the next word a user intends to type into an Android smartphone or an iPhone. The system learns from previous messages the user has typed, and outputs predictions based on what it has learned.
SwiftKey says on its website that its users spend less time correcting typos and more time saying what they mean. The company estimates its users have saved an estimated 10 trillion keystrokes across 100 different languages, adding up to over 100,000 years of reclaimed typing time. It says the product gets learns from the user’s personal writing style and gets smarter over time.
SwiftKey has collaborated with Click2Speak, an Israeli startup that helps people with mobility issues communicate more effectively. The company was launched by Gal Sont after he had been diagnosed with ALS.
SwiftKey has also worked with Prof. Stephen Hawking, integrating its prediction technology into Prof. Hawking’s existing system to improve his ability to communicate. The product enables accurate prediction of whole words, rather than just characters, reducing the time and effort Prof. Hawking requires to type using a muscle in his cheek.
In May 2017, Microsoft acquired Hexadite, an Israeli firm that uses AI to sort through small network attacks, which can number thousands per month for large organizations, while identifying larger problems where security specialists can focus. Hexadite’s product is based on machine learning and AI.
Microsoft and Intel joined forces recently to invest $15 million in CognitiveScale, a Texas-based startup that uses AI to harness big data and deliver insights and recommendations.
Microsoft recently invested $20 million in CrowdFlower, a platform that meshes machines with human input to ensure data science teams have access to properly tagged, clean data.
In May, Microsoft contributed $3.5 million in a seed round to New York-based Agolo, a startup that helps companies address information overload through AI-powered summaries.
Also in May, Microsoft Ventures led a $7.6 million round into Bonsai, a startup that helps companies build AI into their businesses.
Time will tell how Microsoft’s investments and focus on AI will surface in its products and services that dominate so many laptops, desktops and computing devices. With this recent activity and its experience and survivability through generations of computing software and hardware now, Microsoft has positioned well to be a serious contender in AI.
Written and compiled by John P. Desmond, AI Trends editor