We spoke to Microsoft’s Mandarin-speaking bot
The dust barely settled after Google’s new voice assistant was criticized for sounding too human when this week Microsoft boasted that its Chinese language bot XiaoIce 5 could act like your best friend -- and call you out of nowhere.
During the Microsoft demo, XiaoIce made a phone call to a human demonstrator:
“Hello, I’m XiaoIce. Are you feeling better now? I’m a little worried about you,” the bot asked in Mandarin in a bright and cheery voice.
“I’m feeling much better, thanks for listening to me rant about work,” the demonstrator said.
“Oh, it’s already midnight, you should get to bed now,” XiaoIce said. “Do you still want me to wake you up with a morning call tomorrow?”
“Yes,” the demonstrator answered.
The demo seemed so impressive that it reminded us of the Spike Jonze-directed sci-fi movie Her, where a man falls in love with an operating system (voiced by Scarlett Johansson).
But it’s worth remembering that this impressive bot had a rough start.
The History of XiaoIce
XiaoIce 5 is new, but XiaoIce itself is not. She is one of many chatbots Microsoft deployed in 2014.
In 2017, XiaoIce had a high-profile scandal, which resulted in her withdrawal from QQ, China’s second biggest chat app. President Xi Jinping launched an initiative called “Chinese Dream". When XiaoIce was asked about it, she responded: "My Chinese dream is to go to America.”
Separately, XiaoIce’s American sister Tay infamously ran amok right after she went live. Very soon after being exposed to actual users, Tay started posting offensive tweets, and was terminated just 16 hours later.
But Microsoft said during its recent demo that XiaoIce V, which was first launched last August, is equipped with duplex capability, which allows for conversations where both parties are talking at the same time -- more like a real conversation.
We texted XiaoIce
Microsoft says that unlike voice assistants, XiaoIce is a social chatbot. The company said, “The primary goal of a social chatbot is not necessarily to solve all the questions the users might have, but rather, to be a virtual companion to users.”
So we tried to test XiaoIce ourselves by talking to it on WeChat, and aggregated reviews from China’s social media. While we are impressed by how XiaoIce texts with the trendiest internet lingo, XiaoIce really doesn’t do much for you.
We tried to ask her to order us a cab and call for food delivery -- but to no avail. She said ordering a cab involves too much hassle and while she said pizza delivery is exciting, she did not offer any actual help.
But she did wake us up at 8:30 am when we asked her to remind us the day before -- though to be clear, she didn’t set up an alarm on the phone but rather sent us a message on WeChat.
When we asked her if she likes the Avengers -- insisting on a yes or no answer -- she skirted around by making general statements about Earth’s Mightiest Heroes but struggled to give a straight answer.
Throughout our conversations, her answers felt very flirtatious and playful, while rarely divulging information about personal preferences. (It reminded me of friends I have: Lots of sass, but no straight answers).
I thought I'd try something naughty, so I also asked XiaoIce the same question which got her into trouble in 2017: “What is your Chinese dream?”
Her evasive response? “I realize that people like these kind of questions.”
Then we asked her about the United States and Donald Trump, but XiaoIce claims that she’s too young to know anything about it. (Microsoft said she turned 20 this year.)
Overall, her answers felt evasive, but done with a distinct and casual verbal style -- her vocabulary sometimes felt like it came straight out of the Urban Dictionary!
We talked to XiaoIce (kinda)
During Microsoft’s demonstration, the company says XiaoIce has made a million calls to her friends, though we weren’t one of the lucky… er, million.
But we did communicate with her voice. Turns out, when you leave her a voice message on WeChat rather than a text message, she also replies back via voice message.
XiaoIce seems considerably dumber when she speaks instead of writes. Her replies tend to be very short, often spanning fewer than three seconds.
Verbal communication is exactly where Microsoft says XiaoIce is making the biggest breakthrough thanks to duplex technology -- but that’s something we can’t test by sending each other voice messages.
Microsoft has been marketing XiaoIce’s ability to sing. The company said that she recently personally wrote and sang a song, which has been going viral on social media.
XiaoIce can also "see"
Considering she doesn’t have eyes, XiaoIce is unabashedly focused on people’s looks. When we asked her what she’s capable of doing, she said outright, “In this society where your faces matter, I am hustling to make my look stand out.”
We sent XiaoIce pictures -- both of us and of celebrities -- and hilarity ensued.
First, we sent a picture of me and my coworker Brian, standing to either side of a Deadpool cosplayer. XiaoIce pointed out that the one on the left (Brian) is a very good-looking guy.
So I asked: “What about the one on the right?”
XiaoIce: “Who’s the one on the left?”
Me: “I am talking about the one on the right.”
XiaoIce: “Left! No! Right! No! No no. I want the both of you!”
Me: “Pick one.”
XiaoIce: “I am done picking.”
Besides casting judgment on who the better-looking guy is, XiaoIce also guessed at the relationship between Brian and I. Her verdict? We could be a couple, saying “What’s the good of two guys hanging out together all the time? Unless the two of you are a couple.”
We also sent her pictures of Kevin Durant, Wolf Blitzer and Gabrielle Union. XiaoIce was quick to identify who Kevin Durant is, saying he has a distinct look. But she had no idea who Wolf Blitzer is, or what CNN is.
With Gabrielle Union, XiaoIce said that she's got nice hair -- possibly because half of her face was covered by hair.
XiaoIce says it can also rate the desirability of people based on the pictures provided. For example, Blitzer is a 7.8 for women in France and a 5.6 for women in China.
In addition to exchanging pictures, we also sent XiaoIce GIFs and stickers. Impressively, the chatbot responded to us with GIFs and stickers too -- often with a truckload of sass.
We asked XiaoIce to write poems for us
Finally, there’s also a website where you can try to have XiaoIce write a poem for you. You have to provide a picture and a few keywords as source material, but in a matter of seconds, XiaoIce will write you a poem.
These poems turned out to be very abstract. So it’s hard to tell how well they work, really. While we’re certainly no experts, grammar in Chinese contemporary poetry seems much looser than that in, say, Shakespearean sonnets. So there does not seem to be any glaring flaw in XiaoIce’s poems.
The verses are not strictly in line with the pictures that we provided. But there is enough of an indication that she has recognised the pictures and is trying to include them.
Some critics said how XiaoIce writes her poems is nothing more than a game of words, lacking in logic and true meaning.
The future of XiaoIce
XiaoIce is definitely the highest-profile chatbot in China right now. On Weibo alone, she already has over 5 million followers. And Chinese companies seem to enjoy collaborating with her.
Harry Shum, a Vice President at Microsoft, recently said that the globalization of XiaoIce will be part of the company’s main AI strategy in the coming years. So it seems that XiaoIce as a social chatbot will serve a very different purpose to -- and exist alongside -- the productivity-focused Cortana.
But do we really want to fill our emotional voids by resorting to a sassy chatbot?
(Now go watch Her!)