Microsoft: Our chatbots can converse like humans, too

Microsoft is touting the 'full duplex' conversational capabilities of its Chinese chatbot, Xiaoice, while promising similar functionality will come soon to its Zo.ai bot.
Written by Mary Jo Foley, Senior Contributing Editor

Video: Microsoft, um, joins sprint to perfect chatty machine

Google Assistant is getting lots of attention for its ability to converse like a human, but it's not the only game in town.


That was one of Microsoft CEO Satya Nadella's key messages at a May 22 talk in London focused on AI. According to The Verge, Nadella demonstrated Microsoft's Xiaoice chat bot during his appearance, showing how the bot can converse using voice, in addition to text. As The Verge noted, Xiaoice doesn't work the same way that the Google Duplex feature does; Xiaoice doesn't act on a user's behalf, but instead has a back-and-forth conversation with users.

Xiaoice, translated as "Little Bing," has been a big hit in China since it was launched in 2014. Xiaoice has similar counterparts available in India (Ruuh), Japan and Indonesia (Rinna) and the U.S. (Zo.ai), which was the successor to the ill-fated Tay.ai. But none of these are anywhere near as successful, in terms of number of users or conversations as Xiaoice.

According to Microsoft, Xiaoice now has 500 million+ users and more than 230 skills including content, service and tasks, all of which are focused on more on EQ, or "emotional intelligence," in Microsoft parlance.

Xiaoice has played four million hours worth of audiobooks, officials said. It generates financial news for WIND which covers 90+ financial organizations in China and it hosts more than 20 TV and Radio programs every week through its Content Creation Platform. Microsoft refers to these audiobook and cross platform media solutions as "Xiaoice for business."

Microsoft Executive Vice President of AI and Research Harry Shum blogged yesterday that Xiaoice has "talked" with more than 600,000 people on the phone since launching this skill last August. (And in a jab against Google and its Full Duplex I/O demo, Shum quipped "And most importantly, we made sure people were informed that she wasn't a real person.")

Microsoft is making Xiaoice's full duplex capabilities (the ability to have human-like verbal conversations) available for partners and developers to use in their own applications. Microsoft also is adding a feature to Xiaoice to allow it to create a 10-minute customized audio story for kids in about 20 seconds using trained deep-learning models for personalizing the background music and story characters. This service will be free and available across markets in Asia starting June 1.

A Microsoft spokesperson told me today that Zo.ai also will be getting full duplex voice conversation support "soon."

With all these new capabilities, what distinguishes Xiaoice or Zo.ai from Cortana, in Microsoft's view? Could Xiaoice and its derivatives ultimately overpower Cortana in a bot war and become the new personal assistants du jour?

A company spokesperson provided this feedback, when I asked:

"While our social AI chatbots and Cortana both leverage the Bing back end, they are different technologies with different purposes. Cortana is the world's first truly personal digital assistant designed to help you get things done throughout your day. With a focus on helping you get things done at home, at work and on the go, Cortana gives you the tools to achieve more no matter where you are. Xiaoice, Zo and our other social AI chatbots are designed for social conversation and entertainment.

"We remain deeply committed to Cortana and are excited for what our conversational AI technologies mean for the future of Cortana."

Microsoft purchased AI startup Semantic Machines this week to help bolster its natural-language AI efforts. Officials said the technology would help with full-duplex advances in both chatbots and Cortana.

Google Assistant's big update: All the new AI tricks and features, explained

Editorial standards