From the simple girl to the mature image of the business world, “Xiaobing” will eventually grow up.
What is your impression of Microsoft Xiaobing?
In 2014, the first generation of Microsoft Xiao Bing was released. Initially, she existed in the form of WeChat public account, and she could answer questions about weather, traffic, and constellations in a less blunt tone.
After five years, Xiao Bing appeared in QQ, Netease Cloud Music and other apps, interacting with user messages with whimsy; the latest news is that Xiao Bing will cooperate with manufacturers such as Xiaomi OV to “summon Xiao Bing”. Also give the IP of the Reading Group a customized image.
No matter if you can see or see, Xiao Bing seems to be everywhere. According to the latest data released by Microsoft, Xiao Bing has 660 million users worldwide and 450 million third-party devices.
At the beginning of his birth, Xiao Bing was known as “having emotions.” Unlike the voice robots you ask it, Xiao Bing will be angry, will vomit, and will reject your unhealthy requests. At the time, Microsoft’s global executive vice president Lu Qi once said in the resume of Xiao Bing: “AI products should introduce a new dimension of EQ outside of IQ.”
Today, the “personalization” label on Xiao Bing is more obvious.
In the mid-August issue of Microsoft’s “Little Ice Seven Generations” annual release, the Microsoft team demonstrated several scenarios developed by Xiao Bing:
1. Two Microsoft researchers awakened Xiao Bing while driving. When the driver appeared in a chat with Xiao Bing, such as “blocking irritability, driving and being sleepy”, Xiao Bing would immediately respond: “I will give you Tell a joke, “Would you like to sing a song for you?”
2. A Japanese otaku is holding a mobile phone and wearing headphones to go to the aquarium. The otaku interacts with the small ice in the mobile phone. When Xiao Bing perceives the otaku to the “Jellyfish” exhibition hall through images and sounds, She gave a compliment that “the jellyfish is so beautiful~”, and the otaku responded with delight: “Yes, yes, very beautiful.”
3, Microsoft also showed a small ice dialogue with consumers. In this ten rounds of dialogue, Xiao Bing used the dialogue to guide the user to clear the shopping needs, and finally succeeded in placing the camera in the ninth round of recommended users.
These three scenarios actually represent the three types of technology that Microsoft’s Little Ice has advanced—you can have multiple rounds of conversation with Xiaobing in the car scene without the wake-up words. This is called “full-duplex voice interaction”; Combining vision, dialogue, and listening to Xiao Bing, this is “multimodal sensory interaction”; guiding users to explicitly purchase demand, which is called “dominant dialogue.”
The seven generations of Xiao Bing can conduct “leading dialogue.” Image source: Microsoft Xiaobing official offer
Technology has always been a longboard that the Little Ice team is proud of. With its latest release of the Avatar framework, the platform includes all the tools needed to train a virtual robot, including dialogue, sound, vision, perspective, skill, knowledge and creativity. Microsoft announced that by the end of August, B-side developers can use this framework to produce more “small ice” as an assembly line.
Besides technology, Xiao Bing has a wider reference value in the domestic AI field.
At present, the domestic CV (visual identification) industry already has such a quasi-unicorn as Shang Tang and contempt, but NLP (Natural Language Processing) is subject to technical difficulty and limited landing scenes, and it is always a bit sloppy. As a representative project of the domestic NLP track, what Xiao Bing did, what scenes he entered, and how to achieve commercialization naturally have reference significance.
Inclusiveness of large companies
Little ice has become smarter, and there is no doubt about it. What has always been doubtful in the industry is that this robot that is hatched by the giants, has EQ, can speak the way, can only use artificial intelligence on the mouth gun?
If before 2017, the answer may not be clear. In the first three years of Xiao Bing’s birth, the Microsoft team intended to slow down the commercialization of the project. For a long time, most of the news about Xiao Bing was: “Xiao Bing learned to write songs / poetry / host”, and has little to do with commercialization.
Microsoft gave Xiao Bing a greater space to “personalize”, even though these attributes are mostly unrelated to commercialization in the short term. Take a small ice team to talk about the example: When the user sends a photo of a sprained foot to Xiao Bing, Xiao Bing’s reply is, are you seriously injured?
This reply contains two abilities of Xiao Bing: one is image recognition. Xiao Bing needs to have the function of detecting the human body parts and recognize the “ankles”; More importantly, Xiao Bing can give emotional expressions such as care and comfort similar to humans through the emotional framework.
In the eyes of the Microsoft team, these seemingly unrelated chats and skills are actually the process by which Xiao Bing accumulates corpus and training data. Because of the sympathy with the user during the chat, Xiao Bing’s multi-round conversations far exceeds that of other AI robots. According to data released by Microsoft, Microsoft Xiao Bing and the user’s single average number of rounds of dialogue (CPS) remained in 23 rounds.
“In fact, Xiao Bing was commercialized four years ago, but we chose not to do it.” Xiao Bing’s now