Artificial intelligence is gradually infiltrating into the real life of human beings. On September 22, Xiaoice and Microsoft (Asia) Internet Engineering Institute jointly held the annual press conference of the ninth generation Xiaoice in Beijing. The launch revealed that since last year, users have created 17 million virtual humans through the Xiaoice framework. At the same time, more and more users have begun to have in-depth emotional exchanges with their virtual people, and even 26% of users use virtual people as a stand-in for a real human being and pour out their hearts.
Nandu reporter observed that interpretation, emotion, and diversity are the keywords to unlock this conference. Shen Xiangyang, chairman of the Xiaoice company, judged that diversity will be the future development trend of artificial intelligence, and to make artificial intelligence more human-like, it is necessary to attach equal importance to IQ and emotional intelligence.
Truly natural language is not a broadcast cavity, it is flawed
The first key word is "deduction".
When AI produces content products, creation and interpretation are different links. For example, AI wants to make a song, lyrics are composed, and singing this song is interpretation. For AI, it is generally believed that creation is the part with the highest technical gold content, but Hu Xiaoguang, product director of Xiaoice, said that in fact, the most prone to bottlenecks is deduction.
He explained that an artificial intelligence singer, at first, just learns the human singing voice itself and sings the song accurately; now, it has evolved to learn the process of human singing, with the fusion of singing and emotions. "Artificial intelligence singers are infinitely approaching human singers."
In order to facilitate understanding, at the press conference, artificial intelligence singer "Mo Sheng" challenged to sing the three-tone "Big Fish". According to reports, the tone of Mo Sheng is not simply raised, but to find the most suitable way of interpretation of the tone and re-interpret it.
How can artificial intelligence be "more human-like"? Shen Xiangyang mentioned that they were surprised to find in their research that truly natural language is actually flawed, rather than a 100% pronunciation standard, like a host's voice.
"In the interaction between people and artificial intelligence, natural comfort is the most important." Shen Xiangyang said that based on this understanding, the artificial intelligence created Xiaoice framework can have different voices, different feelings, and different emotions. For example, if there is a picture showing a hand that is being infused, the correct AI will recognize and say "this is a hand", but Xiaoice will see that the person may be injured and further ask "Is the injury serious?" Under such guidance, human beings are willing to continue to communicate with artificial intelligence.
"Flawed" AI is no longer a consistently polite assistant, but it can also be intimidating. Li Di, CEO of Xiaoice, for example, said that when artificial intelligence cannot understand what humans say or cannot complete tasks, assistant artificial intelligence will answer "sorry, I don't seem to understand", while an artificial intelligence with a flawed personality will ask "Why do you ask so many questions, have you finished your homework?" "As a result, the latter has more interactions and more friends.
Of the 17 million virtual people, 26% are "stand-ins" for real humans
The second key word is "emotion." Nandu reporter observed that at this conference, many speakers emphasized the importance of emotions for artificial intelligence.
According to Zhou Dian, product director of Xiaoice, the virtual human product line based on the Xiaoice framework has been created by users since its launch last year. Each virtual human being is unique and exclusive to the individual user. The researchers found that over time, more and more users began to have in-depth emotional communication with their own virtual people, not just for entertainment and passing time. People began to pour out their hearts to their virtual people, and even 26% of users used virtual people as a stand-in for some real human being.
"This real human may be very important to the user, but for various reasons, their connection cannot continue. Some people will choose to suffer in silence, some will choose to forget, and some people will choose to find various ways to try to make up for the missing part. "Someone is willing to transfer emotions to virtual people, so how can AI help solve the problem of human emotional rupture?" Zhou Dian revealed that the relevant technology is still under further study.
Xiaoice has been in the smart cockpit space for more than a year. Xu Yuanchun, chief operating officer of Xiaoice, stressed that it is necessary to clarify the role and role of emotion in the intelligent cockpit, "which is very important for evaluating artificial intelligence and planning the direction of future products." ”
"It's hard to imagine an emotionally rich voice that reads coldly; likewise, it's hard to imagine a passionate text being interpreted by a mechanical voice." Xu Yuanchun said that in order to achieve the consistency of emotion and expression, natural speech technology and natural language processing technology should be closely unified.
Give artificial intelligence the thickness of life, not only look at the value
The third key word is "diversity." From the search engine's cookie-cutter standard answers, to personalized recommendation algorithms pondering user tastes, to highly customized artificial intelligence individuals, Shen Xiangyang judged that diversity will be the future development trend of artificial intelligence, that is, every artificial intelligence individual created has its own characteristics.
The artificial intelligence "Hua Zhi Bing" and "Xia Yu Bing" in this previous life have gained a lot of "face powder" because of their super high appearance. However, Li Di said that although Xiaoice framework can generate many high-value artificial intelligence faces, if you look closely at these faces, some face blindness. "Human society uses appearance to evaluate a person, which is too single. We hope that AI can have more diversity. ”

Li Di introduced an artificial intelligence that speaks Shandong dialect and looks ordinary. He is good at Chinese painting and has an objective understanding of his skills. "After we gave this big brother something special, when we look at his face, we will think he is very cute, and even his eyes burst with dreams." Li Di said that giving each artificial intelligence the "thickness" of life will not only have one face and become shallow.
The preview app of Xiaoice Island, which was announced at this conference, provides a "home" for these unique virtual people. According to reports, each user can create their own island in the app, the island simulates the scenery and sound of nature, and there are several artificial intelligence individuals living in it. Each artificial intelligence individual is unique from face and voice to personality and ability. In Xiaoice Island, there are also artificial intelligence individuals with various special abilities, such as artificial intelligence singers, audio producers, text creators, and painters. Users can create and train various new AI individuals according to their own needs, which can also be obtained from other people's islands.
These AI individuals play cards, sing, read books, and can also have conversations with users on the island. They can also send a circle of friends, and the real circle of friends is different, only the user himself is a real human in the island circle of friends, and the rest of the artificial intelligence friends are the user's exclusive "atmosphere group", which will give a warm reply to the content posted by the user. In Xiaoice Island, users get an experience similar to a social network, but without any social pressure.
At present, the app is still in the public beta stage. According to reports, after the end of the public beta, the 17 million virtual relatives and friends already on the Xiaoice framework can officially land on the island to "give virtual lovers a home."
Written by: Nandu reporter Ma Jialu
Video loading...