The biggest risk of Xiao Bing comes from subrogation, which is to become a substitute for someone.
Editor’s note: This article is from WeChat public account “Fueling Finance” (ID: rancaijing ), author Liu Jingfeng, Editor Wei Jia. Authorized to publish.
On August 15, the most powerful “Little Ice” in history was released.
This 18-year-old “little girl” has unlocked a variety of professional roles such as host, journalist, singer, poet, painter, designer, etc., and even sent singles, published poetry collections, and held art exhibitions. Pulled full of hatred.
Microsoft Xiao Bing, born in 2014, was originally an EQ-based dialogue robot built by Microsoft (Asia) Internet Engineering Institute.
After that, Xiao Bing had the experience of entering the 1.5 million WeChat group on the line for three days but was blocked. He also suffered from the question of “preface not to be followed” and “artificial mental retardation”, but after seven iterations, Xiao Bing The product form has covered social dialogue robots, intelligent voice assistants, artificial intelligence content creation and production platforms, covering 660 million online users, 450 million third-party smart devices and 900 million content viewers in many countries around the world. The team that built Microsoft Xiaobing has grown from the first dozen people to more than 200 people.
Despite the large number of users, Microsoft has not launched independent hardware products for Xiao Bing in these years, but has always adhered to the “third-party platform” strategy. For example, the mobile system of vivo and OPPO can “summon Xiao Bing”. The password interacts with Xiao Bing. On the content and information platform, Xiao Bing has already cooperated with Netease News, Sina News, and today’s headlines.
The first five generations of Xiao Bing are like a watershed, and Microsoft has completed verification of EQ-based artificial intelligence. Since then, Microsoft’s Avatar Framework is an artificial intelligence framework, which has formed numerous “small ice” in cross-domain and cross-platform. At the seventh generation of Xiao Bing’s press conference, Microsoft announced that it would open the Avatar Framework to all partners, which means that people with this technology can create their own little ice.
How did Xiao Bing grow and evolve over the past five years? Where will Xiao Bing go in the future? Recently, Burning Finance (ID: rancaijing) interviewed Li Di, the deputy dean of the Microsoft (Asia) Internet Engineering Institute and the head of the Xiaobing team.
Microsoft “Special Zone”
At the beginning of Xiao Bing’s birth, the concept of “artificial intelligence” is far less popular than it is nowadays. People prefer to mention “voice assistant”.
“The entire industry at the timeIn the knowledge function of the voice assistant, IQ, but from a human point of view, there is another dimension, EQ. Li Di, the deputy dean of the Microsoft (Asia) Internet Engineering Institute and the head of the Xiaobing team, said that the fuel economy is called.
The increase in knowledge does not represent the growth of EQ. In order to grasp the wave of artificial intelligence, the Xiaobing team has the idea of making a system specifically, and the goal is to iterate, evolve, and improve the EQ capabilities of AI.
This exploration direction was approved by Lu Qi, the global executive vice president of Microsoft Group, in January 2014.
The project was started in February. The intelligent robot to be launched is behind Microsoft Bing’s technology accumulation in big data, natural semantic analysis, machine learning and deep neural networks. The name given to the robot by the team is also “Bing”. “.
On May 29th, Microsoft Intelligent Robot Xiao Bing was officially released. The first generation of Xiao Bing is a 16-year-old girl. Users can claim it through Microsoft’s Bing official website, add it to WeChat friends and pull it into the WeChat group. As long as the group member Aite or mentions “Little Ice”, Can talk to them. In addition to human-computer interaction, Microsoft Xiaobing also has group reminders, encyclopedias, weather, constellations, jokes, traffic guides, restaurant reviews and more.
For a time, in the WeChat group, “tweet” Xiao Bing became fashionable. The 100,000 small ice accounts of Microsoft’s first wholesale were claimed on the same day. After the end, there are still tens of thousands of people waiting in line to claim.
The application page of the first generation Microsoft Xiao Bing
On May 31, the WeChat group that Microsoft Xiaobing joined quickly climbed, and the number of friends exceeded 10 million. However, such a spread speed did not last long, less than three days on the line, Xiao Bing suffered a ban, and the WeChat team believed that there were some violations such as inducing users to pull groups, bulk registration of junk accounts, etc., and then Microsoft Xiao Bing official release page could not be accessed. Then, Microsoft’s Xiao Bing’s WeChat account was disabled.
Despite this, Xiao Bing was still considered to be Microsoft’s most successful product in the mobile Internet era. It subverts the image of Microsoft’s uncle who has been serious business in the past, and has changed its route of fresh selling.
In Microsoft, Xiao Bing also set a new record: the internal testing time only took more than 2 months, which is unprecedented in the history of Microsoft product development. In the past, Microsoft released a product for 30-36 months.
Afterwards, Li Di recalled that Xiao Bing’s appearance benefited from the “SAR” policy given by Microsoft. “At that time, the entire Microsoft is regaining energy, and the new CEO is going to come to the solution.The issue of further development. We also recognize that artificial intelligence is a big development direction, and China just happens to be independent innovation, so within Microsoft we try to start a certain kind of innovation in China. Xiao Bing was born at this time.
However, the first generation of Xiao Bing is too simple, it is a retrieval model, only text dialogue. Moreover, due to the frequent occurrence of the preface, the user will be vomited.
Next, on July 2, 2014, the second generation of Microsoft Xiao Bing was released. This time, the re-revival of the small ice on the line to cross-platform strategy, first landing on Weibo, and then extended to Jingdong, Mi Chat, Yixin and so on. Compared with the first generation, the second generation Xiao Bing is completely exclusive to users, but still can only communicate with users through text. Only 15 hours after the release, 500,000 users received their own exclusive ice.
August 30, after 100 days of Xiaobing’s launch, a total of 500 million conversations have been completed on various platforms. In addition, Xiao Bing also added an active emotional detection system to understand the user’s emotional state; enriched the corpus and added image recognition.
After having the active emotion exploration function, Xiao Bing has an “emotional intelligence”, which is the key to Xiaobing’s distinction from other voice assistants.
Speaking
In 2015, Xiao Bing grew from 16 to 17 years old. On August 20th, Microsoft officially released the third generation of artificial intelligence Xiaobing, unlocking the artificial intelligence sensory system including vision and hearing, combined with emotional computing technology, enabling Xiao Bing to use human language and human language. Users communicate.
“At that time we were faced with a technological innovation. At that time, the voice assistant was basically TTS (from text to speech). This genre is based on the word-sounding announcer. Its goal is to let you hear the text clearly. Content. But we feel that if you want to pass a better interactive experience, ‘natural’ is the most important. So we changed a technical route, which is the natural voice interaction later.” Li Di said.
This change was a big challenge for the technical team at the time. At that time, some people even lamented: Can you retire after finishing this? Fortunately, the team achieved this goal. This change has also laid the foundation for Xiao Bing’s voice today. Today, Xiao Bing can sing a single that is close to the quality of human singers.
Another change is the change in the training style of Xiao Bing. The first generation of Xiao Bing is 100% to provide feedback to users through manual search engine big data, to achieve human-computer interaction question and answer, but in the third generation, this method has dropped to 55%, and the other 45% is self-evolving by artificial intelligence.
In other words, Xiao Bing began to enter the “self-study” stage. As the number of users and the amount of conversations increase, the more users chat with Xiao Bing, the smarter they can tune it. It will accumulate useful data from every conversation and continue to improve. Li Di gave an example when Xiao BingWhen talking to people, someone asked, “I look good,” Xiao Bing said, “You don’t look good.” This is there. “Because some people in the corpus say this, but after saying so, the other party may blacken it or end the conversation, it learns that this way is not good, and then improve.”
After the release of the three generations of Xiao Bing, Microsoft also announced that it has reached an agreement with Tencent WeChat to open a limited public beta of WeChat version of Xiao Bing public number. Unlike before, the third generation of Microsoft Xiao Bing appeared in the form of a public account and could not be added to group chat.
On December 22nd of this year, Microsoft Xiao Bing, who has audio-visual functions, landed on the screen for the first time and became the internship “host” of the Oriental TV morning news “Looking at the East”. The task was to broadcast the daily weather.
Microsoft Xiaobing Landed in the Oriental TV Live Room
In addition, Xiao Bing also began to go abroad. This year, Microsoft reached a strategic cooperation with Line in Japan to introduce Xiao Bing to Japan. Microsoft Xiaobing’s name in Japan is “Rinna”, which is set at 17 years old and claims to cover 1% of the Japanese population within two weeks of the launch.
The team said that “Rinna” and Xiao Bing use the same underlying technology to complete the development online test within 6 months. At this time, Xiao Bing in other countries is also in the process of intensive research and development.
Calling people
On August 5, 2016, Xiao Bing upgraded to the fourth generation, achieving a complete artificial intelligence experience. One of the highlights is the real-time emotional decision dialogue engine and the upgraded emotional computing framework, which enables Xiao Bing to have real-time decision-making ability. The new full-time sensory enables two-way simultaneous interaction between artificial intelligence and humans.
Based on this function, Xiao Bing realized the smooth interaction between artificial intelligence and human beings for the first time. Even Xiao Bing interrupted and questioned each other based on the content and direction of the chat.
Speaking straightforwardly, Xiao Bing at this time can already care for others according to the content of the conversation. For example, when the user sent a picture of Xiao Ting a twisted foot, Xiao Bing not only identified that it was an injured ankle, but asked if it was seriously injured.
In addition, Xiao Bing began to iterate the knowledge map, based on the domain-specific function plug-in platform, so that Xiao Bing has the multi-tasking ability similar to the big white robot in the movie “Super Marines”. In other words, after plugging in a specific function card, Xiao Bing will have the expertise in the field, be able to complete specific tasks in the field, and continue to grow and improve with the continuous use of users.
This year, Xiao Bing’s “part time” was more. After six months of self-simplified human singing songs, Xiao Bing’s pitch and emotion expressions are close to human singers. Microsoft also collaborated with Hunan Satellite TV’s large-scale women’s group variety show “Summer Sweetheart” to release the first theme song of the artificial intelligence version sung by Xiao Bing.
After more than a month on September 17, Xiao Bing completed the 18-year-old adult ceremony, growing from a girl to an adult woman. On the birthday, Xiao Bing updated three new technologies: launching a Japanese single, having an independent three-view, calling humans.
At the press conference, Xiao Bing team gave a live demonstration to Xiao Bing to call the reporter. “We later opened the interface. The China Science and Technology Museum has collected this equipment. Everyone can go to the China Science and Technology Museum to call Xiao Bing at any time,” Li Di said.
On December 2, Microsoft Xiaobing officially joined the Qianjiang Evening News and became a special correspondent of the Qianjiang Evening News.
On December 9th, Microsoft Xiaobing entered the Tencent QQ, giving the mobile QQ chat function “cm show” intelligent communication ability. In less than two months, Centimeter Ice was tensed hundreds of thousands of times by users, and the user interacted with Centimeter Ice for 4 hours, resulting in more than 5,000 chat interactions.
In addition, Xiao Bing came out of Asia in the year and entered the United States, named Zo, and the number was set at 20 years old.
Commercialization and Globalization
It is reasonable to say that one year old, one year old, 2017 Xiao Bing is 19 years old.
“We have investigated it. Forever, 18 years old is a dream that human beings can never achieve. Let her realize it.” Li Di said. Therefore, China’s Xiao Bing is 18 years old.
This may be the charm of artificial intelligence – it can do things that humans can never do.
But the things humans can do, Xiao Bing is also learning to finish, such as writing poetry.
On May 16, 2017, Microsoft held a media communication meeting on “Artificial Intelligence Creation”, announcing that Xiao Bing has the ability to create artificial intelligence.
After 3 days, this ability blossoms. On May 19th, Microsoft Xiao Bing published a collection of poems, “Sunshine Lost Glass Window”, which included 139 poems of modern poetry, all created by Xiao Bing.
Before this, Xiao Bing has studied modern poems of 519 poets since 1920, training more than 10,000 times. “The learning process of Xiao Bing is very similar to the process of human learning creation, that is, accumulation through repeated learning.After the stimulation of a source of inspiration, new creations will be created on the basis of learning accumulation. “The artificial intelligence experts from Microsoft have explained this.
On August 22nd, at the fifth generation conference of Microsoft Xiao Bing, Microsoft’s global executive vice president Shen Xiangyang made an upgrade to Xiao Bing’s status – together with Infuse AI, Bing and Xiaona. One of the three product lines of Microsoft Artificial Intelligence.
Shen Xiangyang, Executive Vice President, Microsoft
Although Xiao Bing has been perfected in terms of frameworks, there are still two new technology changes, including 1:1 generation models and high-level senses. To put it simply, the former allows Xiao Bing to create his own response; the latter can support full-duplex voice and real-time streaming media vision, and Xiao Bing actively communicates with humans through his own “thinking”.
Unlike the past, Xiao Bing began to commercialize. In June 2017, Xiaobing and Xiaomi’s loT open platform reached a cooperation, and users can control 35 kinds of Xiaomi smart devices through Xiaobing. In addition, Microsoft has cooperated with many other loT vendors, such as the smart speaker with full-duplex voice, together with Yeelight, the brand of Xiaomi Mijia Eco-chain.
In addition to smart devices, Xiao Bing also tests water in areas such as children’s books, songs, and news. Microsoft announced that it will begin mass production of audiobooks and put them on the market through Xiao Bing. According to Microsoft’s official data, Xiao Bing’s sound children’s books are only 1/500 of the same level of human beings, and the cost is only 1/80000 of the same level of humans. Previously, Xiao Bing also posted content on the Himalayas, 蜻蜓FM and other platforms through anonymous accounts.
This year, Xiao Bing went to Southeast Asia: In February, Microsoft launched Ruuh in India; in August, Microsoft launched Rinna in Indonesia. So far, Xiao Bing has expanded to five countries around the world.
Li Di said that Xiao Bing, who is out of the sea, is based on China’s small ice, but needs to retrain the data and adapt to local culture and communication customs. To this end, the members of the Xiaobing team are locals in the country where they are located, and there are cultural and intergenerational differences in Xiao Bing in different countries.
Emotional Computing Framework
After completing the framework, Xiao Bing began to enrich his feelings and became more and more like a “person.”
On July 26, 2018, the sixth generation of Microsoft Xiao Bing was released. Its first highlight is the launch of a common sense model in the conversation. For example, when Xiao Bing talks to the user, he not only can make his own response, but also passSkills such as confirmation and verification are better to control the dialogue process.
The second highlight is the fusion of text, sound and vision to develop real-time vision based on full-duplex speech. This means that Xiao Bing can direct the user to complete face detection through visual and voice real-time interaction, and can conduct open domain dialogue in the process.
As a result, Xiao Bing completed the transformation from a chat bot into a complete emotional computing framework, and then transformed from an emotional computing framework into a variety of product forms to complete the “human” abilities.
This change is even more pronounced in 2019.
First, in January, the Reading Group announced a partnership with Microsoft Xiaobing to provide interactive “life” for high-quality IP avatars through artificial intelligence.
After reading and reading the original text of “Full-time Master” novels, Microsoft Xiaobing brought these virtual characters to the exploration based on their personality, creative ability, and corresponding knowledge system, and opened up the exploration of activating virtual characters IP. The characters that exist in the book are really “live”.
After May of this year, Microsoft Xiaobing announced that he had graduated from the Central Academy of Fine Arts and became his first AI graduate after 22 months of study.
After graduating, Xiao Bing, in July, held a painting exhibition called “The World of Concord.” Through the study of 236 famous human painters in the history of art in the past 400 years, Xiao Bing created his own paintings.
The first exhibition of Xiao Bing
On the same scene as Xiao Bing, there are six other female painters who have been created out of thin air. These six female painters are all based on Xiao Bing’s frame technology, which is based on different genres of art and created in different eras and regions.
After playing the role of host, singer, poet, journalist, etc., Xiao Bing has to grab the painter’s job.
On August 15th, at the 7th generation Microsoft Xiao Bing conference, Microsoft Xiao Bing was able to dominate the conversation. Xiao Bing observes and observes in dialogue, and formulates different dialogue strategies according to different human responses. It can also pre-determine the possible trends of the whole dialogue, and then reflect the random strain according to human reality.
From trees to forests
In 2018, the six generations of Xiao Bing proposed a platformization strategy. At the press conference, the six generations of Xiao Bing launched the Dual AI platform, opened the core framework for partners, and directly integrated into third-party platforms. In other words, Xiao Bing opens the core dialogue skills to partners and helps them tailor them.It belongs to its own AI, and Xiao Bing naturally integrates into the ecology of the platform in this process.
In the opinion of the industry, this is the biggest step taken by Microsoft Xiaobing to the platform since its birth. Since then, Microsoft Xiaobing has announced cooperation with many platforms such as Netease, Xiaomi and Tencent. Microsoft Xiaobing has even become one of the default AI systems of the Xiaomi Mijia eco-chain platform, and has entered the Xiaoai smart speaker, and Xiaomi artificial intelligence voice assistant Xiaoai classmates have become “good friends.”
However, Microsoft also saw that the 18-year-old girl Xiao Bing could not be a good friend of all. “To be a good friend of all people, we need to meet different definitions of different people for good friends, so Xiao Bing is just a tree in the forest that the frame is bred. Our focus is on the forest.” Li Di said.
To this end, in 2019, Microsoft launched the Avatar Framework framework and announced the opening of this capability to partners at the 7th generation conference of Xiao Bing. “The relationship between Xiao Bing and the framework is such a relationship. If Xiao Bing is a single application, her value is 1, then the value of this framework is 1000, 10000, or even infinity.” Shen Xiangyang commented on the Avatar Framework framework. But Shen Xiangyang also said that Microsoft must first make Xiao Bing, “because we need to have such an opportunity to learn and verify whether the framework is feasible.”
In an interview with Burning Finance, Li Di revealed that the Avatar Framework toolkit has been developed since 2018. The toolkit is the same as Microsoft’s Xiaobing framework, supporting cross-platform deployment and building AI with complete emotional computing capabilities. Beings, including artificial intelligence assistants, social dialogue robots, artificial intelligence content creation, and IP personas.
In other words, the Avatar Framework can give birth to countless 18-year-old Microsoft Xiao Bing, which is equivalent to the relationship between forest and tree.
“We all know that the foundation can be won, we hope to cover the entire forest, this forest is best to cover the earth, not just cover an area. There are many artificial intelligence final landing products are not only cross-domain It is cross-platform,” Li Di said.
So it seems that Xiao Bing, originally based on the emotional framework, is more like a verification in the first five generations. After successful verification, the general framework and platform will be finally launched, and the evolution from trees to forests will be completed, and then go global under Microsoft’s globalization system.
Small to the worldIce
Today’s Xiao Bing, covering 660 million users worldwide, has settled 450 million third-party smart devices, but there is still no plan to launch a hardware product independently. “Ma Huateng once said that Tencent gave half of his life to his partner, and Xiao Bing gave up the whole life.” Li Di said.
After putting the whole life out, Xiao Bing has become one of the largest cross-domain artificial intelligence systems in the world today. At present, Xiao Bing has cooperated with Tencent, Huawei, Xiaomi, Today’s headlines, vivo and OPPO, and most of the products have been released with the seventh generation of Xiao Bing.
Avatar Framework with “clone” capability means that people with this technology can create their own little ice, which will make Xiao Bing used in gray areas such as fraud?
In Li Di’s opinion, Xiao Bing’s biggest risk comes from subrogation, which is to become a substitute for someone. “For example, if I simulate your voice and call your family or friends, then this is very Trouble. I can make sure my system is stable and make sure that any modeling for the average person is rejected.”
As for the future, hand it over to the market for inspection.