Sima Huapeng, founder of silicon-based intelligence: All people who use AI should focus on things that are unique to human beings

Reporter: Zhu Chengxiang Editor: Dong Xingsheng

"Hello everyone, I'm Liu Qiangdong, doesn't it look a little different today? I haven't broadcast live for a long time, and I'm still a little nervous...... Not much to say, Lao Liu's digital live streaming starts now. On April 16, Liu Qiangdong, the founder of JD.com, started a digital human live broadcast on the JD platform.

The live broadcast of "Dongge" digital human quickly attracted widespread attention, has the era of digital human replacement of live broadcast come?

On April 18, Sima Huapeng, founder of silicon-based intelligence, the head manufacturer of digital humans, accepted an exclusive interview with a reporter from "Daily Economic News". Sima Huapeng said: "Liu Qiangdong's digital human live broadcast is a very big brand promotion for our industry. His use (of digital human live broadcast) shows that this industry has received a very important recognition. ”

Regarding the use of AI, Sima Huapeng suggested: "In the future, all people who use AI should focus on things unique to human beings, such as creativity, such as things with temperature." ”

Dismantling digital human technology

Does "digital human Liu Qiangdong" have the support of silicon-based intelligent technology? Sima Huapeng said: "It is not convenient for us to disclose it to the public, but the earliest pioneers of these technologies must be us. We have more than 40 invention patents on the entire (digital human) live broadcast, and dozens of applications are being applied. There are a lot of people in the industry who are licensing our technology, and there are also a lot of APIs (Application Programming Interfaces) that are using silicon-based intelligence. ”

Sima Huapeng introduced: "It (digital human) is rendered with AI. If you want to do the same style as Dongge now, you need to provide a few minutes of video. Then we will train it into a digital human model. Behind this digital human model, it will be driven by a large model. Regularly going to some products, replying to some customer needs regularly, and making some records on a regular basis are all scripts that have been made in advance, and these scripts are essentially driven by large models, which are relatively close to human behavior and feedback. ”

So, in addition to the above-mentioned AI rendering technology, what technical support is needed for digital human live broadcast? Sima Huapeng said: "Our large model is multimodal, which is a combination of text generation, sound generation and digital human generation. In fact, many people are reporting that his (Liu Qiangdong's) voice and sense of rhythm are not the same as his original speech. This is likely to clone the usual speaking speed, but during the live broadcast (process), the length of a user's stay is relatively short, and you must disclose a large amount of information in a short period of time, so you must play your voice faster. ”

Regarding speech synthesis technology, Sima Huapeng said: "All our technologies are our own. In terms of end-to-end generation, [speech synthesis] is already a very mature technology. ”

Digital human image and speech synthesis are all external forms of presentation, and the live broadcast content of digital humans is still inseparable from text. Regarding the text, Sima Huapeng said: "In terms of text models, we are basically original technologies, and recently the 'Yandi' model has just passed the filing of the Cyberspace Administration of China. ”

Can it be an alternative to live streaming?

At present, the live broadcast you see is likely to be a digital person. Sima Huapeng introduced: "We will create the general direction of AIGC live broadcast in 2021, (Liu Qiangdong digital person) is basically the effect that our products can present in 2023, and we have sold tens of thousands of such live broadcast rooms, JD.com, Taobao, Douyin, Kuaishou and video number platforms." ”

Sima Huapeng said: "When we watch Dongge's live broadcast, will we treat him (digital human) as the real Dongge, which is an important sign to determine the degree of intelligence of digital humans." From the start of the broadcast to the present, there have been a lot of evaluations in the industry, and everyone thinks that some traces of machinery can still be seen. ”

"The core of the development of this (digital human) industry is 'invisible'. Sima Huapeng emphasized.

He believes: "We now also have a large number of live broadcast rooms, and the effect is much better than this (Liu Qiangdong digital person)." If his live broadcast is not on his own platform, but on other platforms, it will be easily recognized by the machine, and he will soon be restricted. ”

Therefore, in Sima Huapeng's view, the core of the digital human industry is to make the entire live broadcast room conform to the Turing test. As for the development of the digital human industry, it believes: "(In this field) the Turing Test 1.0 is indistinguishable (whether it is a human or AI), the Turing Test 2.0 is a two-way emotional interaction, and the Turing Test 3.0 is a 'life and death match', and in the future, the relationship between us and AI may be Soul Mate (soul mate), which is your very important assistant, friend, and may also be your close partner." ”

At present, many question-and-answer large models mainly provide professional assistance to the audience, but if you appeal to emotion, is it more advantageous to have a multimodal large model with images, sounds and texts? "More than 90% of the design of the human brain is for graphics, so we say that there are pictures and truths. The ability to read texts is really important, but for most people, it is very tiring to generate pictures and then imagine a space and restore it. Therefore, multimodal or video-based interaction capabilities can greatly reduce the mental energy consumption corresponding to information communication between us. ”

For the future form of AI e-commerce, Sima Huapeng believes that the greater value in the future is professionalism. For example, let's say we have a good AI expert in the food field, who can provide us with a lot of professional content from the perspective of data and algorithms. Such an expert will bring more influence than the existing human expert, which will lead to real and better e-commerce transactions.

It further explains: "Human beings do a lot of things with a certain bias or bias, or with a certain commercial purpose. AI experts driven by numbers and algorithms will become our friends and partners, and it is likely to become the mainstream of AI e-commerce in the future. Now these costs are still relatively high, and they are only appearing in the luxury sector. I hope that in the future, there will be such an expert consultant on every type of small product to provide us with a very good way of communication. This is a very important part of the future of AI e-commerce. ”

National Business Daily