SEATTLE—Major tech firms—including Amazon, Apple, Google, IBM, and Microsoft—are all making big pushes into artificial intelligence these days. At the Build conference here, I spoke with Dave Forstram, Microsoft's Senior Director of Communications for AI, for insight into what Redmond thinks sets it apart from the competition and where the company sees the field going over the next few years.
Two Teams
Microsoft's AI efforts fall into two different teams: the research division, headed by Harry Shum, and the Azure AI services for developers, headed up by Scott Guthrie.
The first team's work shows up in things like Bing, Cortana, ambient computing, and robotics. The Azure side makes AI services available to developers, letting them build apps using customizable machine learning with speech, language, vision, and knowledge services. Tools offered by the division include a Cognitive Services and Bot Framework, and deep-learning tools like Azure Machine Learning, Visual Studio Code Tools for AI, and Cognitive Toolkit.
AI Hardware
Microsoft is also investing in AI-powered hardware. At Build 2018, Microsoft announced Project Brainwave, which uses field programmable gate arrays (FPGA) to perform AI calculations. The setup offers latency five times lower than Google's TPU hardware, Microsoft CEO Satya Nadella said during his Build keynote. Brainwave hardware is also reconfigurable compared with TPU's set design. Brainwave even runs Google's Tensorflow AI code as well as Microsoft's Cognitive Toolkit and Facebook's Caffe2.
AI on the Edge
Another key to Microsoft's strategy is edge AI—where the computation can happen on a local device rather than in the cloud. There are plenty of industries that can't afford to have everything stop if cloud connectivity is interrupted. Azure IoT Edge offers a way to bring AI and machine learning away from the cloud and to places like the factory floor. As with a lot of Microsoft's AI tools, IoT Edge is open source and available on GitHub.
Where Is AI, and Where Is it Going?
Forstram likened AI evolution to human development. "The first thing a child does is it sees the world. Next it has to learn speech. And the next thing is comprehension. AI tech has come pretty far in categorizing the visual world, but still need to make more sense of it," he said.
This represents a whole level beyond the utilitarian, brand-focused chatbots, which are currently far ahead of personal assistants like Cortana and Alexa in terms of conversational AI. XiaoIce, a chatbot Microsoft launched in China, "has more than 200 million users, has engaged in 30 billion conversations, and has an average conversation length of 23 turns, which averages out to about half an hour," Forstram said.
Smart home voice assistants are just now adding the ability to support back-and-forth conversations; Google announced it for the Assistant at I/O this week, while Amazon rolled out a similar feature for Alexa, dubbed Follow-Up Mode, in March. At this point, Cortana can only ask you followup questions.
With XiaoIce (as well as Japan-based Rinna and the US-based Zo), "it's a completely different way of thinking about conversational AI. It just interacts; it's natural and personal."
Forstram noted that XaioIce is working toward achieving human parity at translation from Chinese to English; it's also considered "full-duplex," meaning you can interrupt and XaioIce can anticipate what you're going to say. It's so realistic that she's even hosted TV shows and written an acclaimed book of poetry, based on machine learning of centuries of poetry.
"When I hear people talking about the AI vision, that's the closest I've seen," said Forstram.
This degree of language comprehension accuracy has business implications, too. Customer care, in particular, benefits from a bot that has some emotional intelligence and even empathy—"that's the holy grail," Forstram said. It could even encourage people to engage with bots rather than furiously pressing 0 or screaming "REPRESENTATIVE!!" into their phones get a human on the line.
So how does Microsoft's AI differ from those of Amazon and Google? "Everybody started with the same building blocks," said Forstram, referring to machine learning, deep learning, neural networks and the like. "But we use different algorithms, even though we often get similar output."
Google, Amazon, and IBM are building out cognitive services, but Microsoft is the first to let customers customize them. "In the past, if someone wanted to use our speech-recognition technology, we would just give them libraries that they'd have to train their data to use. Google is just getting to that."
Finally, Forstram sees the Microsoft Graph as one of the company's biggest differentiators in AI. If knowledge is broken down into world, work, and user data, work knowledge piece is a distinguishing strength of Microsoft, with Office, Windows, and LinkedIn data to draw on.