li di, CEO of Xiaoice: Today's AI is still in a barbaric era, and it is not right to quickly achieve future commercialization

Economic Observer Network intern reporter Zheng Linna "Commercially, what we often say is that we must exercise restraint. On October 13, Li Di, who had just finished showing the latest voice achievements and virtual portraits of Xiaoice at a forum, expressed his views on the commercialization of AI to the Economic Observation Network.

Baidu's Xiaodu, Xiaomi's little love classmates, OPPO also announced the voice assistant Xiaoyi in September this year, Li Di joked at the interface REAL technology conference, artificial intelligence voice assistant is currently in a "small" era. However, at present, the commercial layout of several Internet companies in related fields is not small. According to this year's Baidu Q1 financial report, Xiaodu's smart device revenue achieved double-digit growth, and laid out the education track and launched the Xiaodu intelligent learning tablet. Xiao Ai also shuttles through Xiaomi's various smart devices, in 2021, the monthly active users exceeded 100 million, known by users, and in the "small" generation is considered to be an old Xiaoice but it has always been relatively low-key.

The first product was launched Xiaoice 2014, and now Xiaoice has nine iterations. From the "AI girl" wearing a student uniform and wearing a pink hairpin on her head, focusing on the upgrading of naturalness MOS and average comfort duration, to this year's ninth generation Xiaoice annual conference, the latest technical indicator: diversity, is working to present a variety of AI Being (artificial intelligence individuals) under the "Xiaoice framework".

In September of this year, a young girl with a ponytail and a guitar playing and singing on the roof suddenly became popular, except for her body movements that can be traced in reality, her voice and face are all generated by computers. This is the AI Being "Huazhibing" launched by Xiaoice Company in cooperation with Tsinghua University. Subsequently, the "Xia YuBing" launched by Xiaoice Company in cooperation with the Central Academy of Fine Arts was also launched. In addition to the high-value "high-quality students", under the framework of Xiaoice, there is also a Shandong big brother who is good at Chinese painting but speaks a dialect.

Unlike pervasive AI assistants, Xiaoice companies are currently pursuing a variety and customization. Li Di said, "So far, one of the tasks that has not been widely valued is the problem of artificial intelligence diversity, how to use a better basic framework to support a variety of AI Beings, rather than pouring all efforts into making an AI Being." ”

Why is there such a huge amount of AI Being needed? Li Di believes that there is a lower, more basic relationship layer between people and artificial intelligence, "Everyone has his own needs, and then he needs a variety of personalities, he needs different abilities, we will know a good friend like Einstein, but we don't just need a friend." So last year we started to improve the Xiaoice framework, to make it more general, so that with Xiaoice this framework is not a Xiaoice, but a variety of 'people'. ”

These "people" will appear in the AI social software "Xiaoice Island" APP. This APP has gained many expectations as "the first AI social APP for mankind". But in Li Di's view, this is not a commercial software that needs to be monetized, but an experimental site for recovering data. "All products become a piece of data that can be iterated on in turn, forming a loop, which is a big difference between our type of products (and other products). That is to say, technology is not used to produce products, but products are used to promote technology. Li Di said.

In Li Di's view, today's AI is still in the barbaric era, and it is very early to put commercialization in 5 years, and it is not right to quickly realize future commercialization. However, it said that the restraint of commercialization does not mean that it is afraid of the end, but that commercialization is limited to several specific fields, including finance, sports, etc., such as education and medical care. Li Di revealed that Xiaoice the current revenue of the B-side is already a scale of 100 million yuan.

"China's market depth is particularly large, from first-tier cities to sixth-tier cities, and the general audience is more educated. So frankly, as long as I have a technology and then quickly expand the technology, then I may have a short-term good return, but whether this income can be sustained should be considered. Li Di said.

Taking the latest Xiaoice as an example, its communication ability is basically the same as that of people, and its voice is not mechanical, the tone is undulating, and it is basically the same as that of real people. Xiaoice can make a sound that is exactly consistent with the user's voice based on the 30-minute to 60-minute audio submitted by the user.

Li Di told the Economic Observer Network: "Many parents come to us and say if they can make my voice, and then I let this voice tell stories to my children every day. This demand is good, but can not be done, why? Because this voice can tell a story to your child, and you can also make a phone call to your child one day, and the child says if Dad can go to the door today, and there is an uncle standing in a white car to pick you up. Because there's no guarantee that your system is secure, the only way to do that is not to model the average person. ”

In terms of voice imitation, Li Di said that the Xiaoice framework is currently not modeled for ordinary people, and the B-side API is not open.