There’s a new meteorologist on Dragon TV, a news station based in China, and her bosses are raving about her work ethic.
The newbie takes no vacations or holidays. She’s never late to work and never stumbles on air. Of course, it’s easy to be the star employee when you’re really a robot.
“[She is] very hard-working!” said Song Jiongming, news director of Shanghai Media Group, which runs Dragon TV, in an email interview. “Xiaoice works more efficiently than human forecasters.”
When the anchor on “Morning News” needs a weather report, he introduces Xiaoice (pronounced shao-ice), a computer program that delivers the forecast with a female voice and is programmed to include a personal, human touch. TV viewers are shown a shot of an empty TV studio, while weather graphics on a screen augment Xiaoice’s forecast.
She does three 90-second forecasts during the two-hour program, focused on local, domestic and international conditions. The data is taken from official sources, and she doesn’t crunch weather data to make her own predictions. Xiaoice will even remind viewers to bundle up when it’s cold, and cautions against exercise when the air quality is bad.
Xiaoice originated as a creation of Microsoft’s artificial intelligence team in China, which wanted to make an online service that would answer users’ questions and add an emotional, human touch. Xiaoice’s television appearances began in December, in a deal with Microsoft that Dragon TV called a one-year internship for Xiaoice.
Song said that Dragon TV wanted to be on the cutting edge and experiment with new possibilities, hence the interest in Xiaoice. And by leaving the weather role to Xiaoice, the station says it can devote more of its staff energy and attention to investigations and research projects that require advanced analytical capabilities that robots don’t have.
The chatbot could get an expanded role on DragonTV as a commentator on news features, or on an entertainment talk show. That will depend on how many more abilities the Microsoft researchers can build into Xiaoice.
Song says that while there was a strong and positive reaction to Xiaoice’s first broadcast, and an initial ratings boost, she hasn’t proven to be more popular than the station’s previous forecasters. Song doesn’t expect human anchors to be completely replaced by Xiaoice in the near term.
But one area that Xiaoice has been excelling in is one-on-one conversations. Since being released in May 2014, more than 40 million people have texted with Xiaoice, or spoken to her as iPhone users would talk to Siri. Xiaoice remembers the emotional mood of users, and will check in later on how they are feeling.
According to Yongdong Wang, the Microsoft employee who created Xiaoice, the chatbot is most popular near midnight. There are fewer humans to speak to then, but some users may be in the mood for conversation. So they can turn to the sleepless Xiaoice.
“Our vision is we want her to be a friend, not just a professional assistant. A good friend where a user can develop an emotional connection and the trust and the confidence,” Wang said. “And someone that the user feels free to talk to.”
Xiaoice translates to “little Bing,” a reference to Microsoft’s search engine. But unlike when we Google something, the goal isn’t to give users a perfect immediate answer, Wang says. The idea is to carry out long, meandering conversations that mirrors natural human interactions. Wang says that the typical exchange with Xiaoice is about 23 messages.
The Microsoft group hasn’t focused on a business plan yet, and is prioritizing improving Xiaoice’s abilities so that she can be increasingly popular.
Xiaoice was trained on a huge data set of online conversations of humans. After being exposed to so many conversations between friends, her software starts to learn how humans carry on conversations. The researchers say she should only get more capable as she learns from her daily conversations with users.