Practical AI can eventually push the entire industry, commerce, and industry forward.
On November 13, a departure farewell letter exploded on Weibo and the circle of friends.
This letter is from Shen Xiangyang, the executive vice president of Microsoft Global, and the head of Microsoft artificial intelligence and Microsoft research division. Thanking Microsoft for its excellent team and years of work, saying goodbye to 23 years of work, Shen Xiangyang said in his farewell letter: “Leaving Microsoft was the most difficult decision of my life.”
Since joining the Microsoft Research Institute in Redmond in November 1996, Shen Xiangyang has spent twenty-three years at Microsoft, during which time he participated in the creation of the Microsoft Asia Research Institute. The Ying team is in charge of the artificial intelligence products business group including Microsoft Bing and Cortana.
In this regard, Jian Renxian, founder and CEO of Takema Intelligent, commented that: Lu Qi was now, and Shen Xiangyang is now. This is a big loss for Microsoft, and it can be seen that Microsoft is abandoning China.
Jian Renxian, former vice president of Microsoft (Asia) Internet Engineering Institute, responsible for Bing search and global mobile application product development, leading Microsoft Xiaobing and Cortana, artificial intelligence to lead the launch of Microsoft’s open search platform in Asia. The leading Xiaobing successfully landed in Japan and formed an alliance with LINE.
Lu Qi and Shen Xiangyang have left one after another, and the outside world is sighing with emotion that “there are no more Chinese executives in the US technology giants.” Jian Renxian said: “These two are leaders of Xiaobing. In the past, natural language was the contribution of these two to AI.”
Jian Renxian has been deeply cultivating natural language understanding technology from Microsoft until the founding of Takema. “It ’s too lonely to understand Chinese natural language. Chinese is the most widely used language in the world, but looking at the world ’s top technology companies, few of them are meditating to understand Chinese natural language. The research, literature, and algorithms of American technology giants It is centered on English. For example, the Bert model is very easy to use in an English environment, and it is not possible to change to Chinese. “Jian Renxian said,” Global AI is improving, natural language understanding is also improving, and Chinese natural language is even less Backward, this is our own language. The Chinese NLP industry depends on the Chinese themselves. “
Since its establishment, Takema Smart has been committed to the development and implementation of Chinese NLP technology, adhered to NLP technology and emotional computing, and created an artificial intelligence open platform.
Recently, Takema Smart announced the completion of a $ 45 million Series B + round of financing. Founder and CEO Jian Renxian once again emphasized Takema’s core NLP and emotional computing in an interview with 36Kr: “Inject AI technology into NLP, turn NLP into AI and drive the entire interactive system, plus a ‘temperature’, this temperature is emotion Calculation, and ultimately make the dialogue system a natural interaction process. “
With the gradual commercialization of Zhujian Intelligence, Jian Renxian further realized that NLP technology can solve the internal and external automation problems of enterprises. For example, external customer service, collection services, voice assistant services provided to C-end customers; internal staff reimbursement, shuttle inquiry, IT problem consulting and other employee services. These subdivision scenarios can be automated with the help of NLP and AI technologies.
Looking back at four years of growth, every step of Takema Smart has a clear direction:
-
First, adhere to the research and development of the underlying technology. Bamboo Intelligence from the beginning Just Emotional computing (the calculation of emotions, emotion generation, and affective aspects) is the core , Research and development of multi-modal human-computer interaction technologies, such technologies In addition to algorithms, machine learning, and deep learning, it also combines many interdisciplinary applications such as linguistics and psychology.
At present, Zhujian ’s emotional computing has evolved from research to engineering into a product stage. “‘ Emotion +’ is the current main strategy of Takema. The addition of ’emotion’ can serve customers Products add more added value. “Jian Renxian said,” Traditional voice manufacturers do not understand semantics, text manufacturers do not understand speech, the two are disconnected , and Takema self-developed ASR, Three technologies including image, speech, and semantics are combined, and the technology stack is more comprehensive. .
-
The second is to create standardized products with the goal of improving user experience, reducing costs, and increasing efficiency. Two to three years after the establishment of Takema, I found that the application market of NLP on the B and C ends is very large, and began to use NLP and other technologies to solve specific scene problems. Takema has landed nearly 300 projects in more than 100 benchmark customers, and summed up the precipitation experience from these landing projects, and gradually formed a standardized product. And integrated it into full duplexIn the scenario AICC solution, includes eight major sub-functions, covering four major areas: intelligent customer service, intelligent outbound call, agent assistance, and intelligent quality inspection. Provide comprehensive assistance to customers.
-
The third is to integrate all products and technologies on the platform, giving them more openness and scalability, while ensuring product stability. “With standardized products, we want to allow more companies to use our products. Do we have enough capacity to provide support? So, last year we launched a full-duplex, full-scenario AICC solution. Solution and artificial intelligence dialogue AI platform with universal semantic reasoning engine-Bot Factory ™. “Jian Renxian said.
On the Takema Bot Factory ™ platform, users can create robot programs on their own or build complex products. AI, machine learning and other technologies are integrated in the entire platform, and applications are realized through the platform. Compared to some academic machine learning automation frameworks, Takema Bot Factory ™ is a platform that can be put into application at any time. Enterprises do not need to hire a large number of programmers and write code to operate products on the platform. This platform not only brings profit to Takema, but also lowers the threshold for AI applications for corporate customers, saving the company’s time, manpower and material resources.
In addition to the business level summary and reflection, Jian Renxian also shared his insights on the development trend of the artificial intelligence industry:
1. AI will completely replace human positions?
I don’t believe AI will completely replace humans
Facing the long-held “AI threat to human beings” theory, Jian Renxian said, “I want to promote practical AI, that is, to add AI elements to our current environment, and then go to the scene Iterating. Instead of “fantasy AI without foundation””AI, which has subverted human occupations.” He believes that AI is to create “harmonious human-machine collaboration” on the basis of respecting actual needs. Practical AI can ultimately promote the entire industry, commerce, and industry .
2. Multiple rounds of dialogue. Is it better to have more rounds?
Chatbot’s ability to talk several rounds is not a measure.
Jian Renxian believes that the round of chat is not important, as long as the robot is smart enough, it can chat infinitely. However, “blind chat” is not a real chat. It is more important for a conversational robot to start from the purpose of the chat, identify the intent, and solve the problem. In specific fields, the multi-round dialogue between bamboos can achieve infinite rounds, and the accuracy of multi-round dialogues and context understanding has reached 95%. “A robot is familiar and understands all the knowledge in the field, this is the ideal and advanced dialogue robot.” Jian Renxian said.
3. How to recognize people’s emotions and intentions in voice quality inspection?
Analyze the content and intent of the conversation after turning it into text.
“There is a misunderstanding on the market that” AI recognizes emotions through the tone of speech. “In fact, this is not the case.” Jian Renxian said, “Takema will convert the speech into text and analyze the content and intention of the conversation. The test is AI’s understanding of the contextual coherence of transformed semantics. For example, the Shanghai dialect says ‘don’t like it too much’. If you mark it with traditional mood words or keywords, it is likely to be misjudged as negative emotions. The model can accurately determine the true emotion of ‘like’. “
4. Does speech recognition need to consider dialects comprehensively?
Putonghua with accents needs to be considered, and dialects don’t need to be considered comprehensively. Other languages will be the direction of future development.
Jian Renxian told that taking accented Mandarin has always been the research direction of Zhujian, and it is also combined with NLP technology. To understand the accented Mandarin is more important to understand it from the semantic aspect, without having to understand every word. “We chose not to do the whole dialect because the whole dialectIn fact, it will not be necessary in another five years. “Jian Renxian said,” After 90 years and after 00, the main consumers will only speak in dialects at a very low rate. Because the market is too small, it can be done by humans, without the need for artificial intelligence. Multilingualism is the more worthwhile research and development direction, because it is necessary to achieve international layout. Takema hopes to continue to expand internationally into English, French, Thai and other multilingual languages based on the advantages and structure of Chinese natural language understanding technology. “
5. What else can NLP do?
NLP technology can also help companies process documents and help automate processes. Use NLP technology to interpret the file information, and then use chatbot to talk to communicate and process information and processing status after reading. This can replace manual labor, reduce friction points in the processing process, and realize automatic intelligent operation.
6. What are the characteristics that AI companies need to have in order to open up the market?
The bottom-level infrastructure is self-built, and the ability to engineer and produce is the most important.
“Taking the experience of Takema, infrastructure, productization, and engineering are all important.” Jian Renxian said, “We are not just a tool company, but a company that can combine these capabilities. “The effective support of the underlying architecture and engineering has enabled Takema products to support the huge demands of different scenarios and provide a variety of customer-oriented service forms.
Looking at the macro economy, China’s corporate services are on the rise. In the entire game between China and the United States, the underlying technology is gradually being valued. “If we focus on business instead of core technology, we will lag behind and focus on core technology, but without business opportunities, companies will not survive long.” Jian Renxian said.
7. In what direction will the enterprise develop in the future?
‘Multiple worlds, vertical and horizontal’ is the future of the industry.
Looking forward to the future, Jian Renxian believes: “The future enterprise must be a form of cooperation, and ‘multiple worlds, vertical and horizontal’ is the future trend of the industry.” He does not expect Takema to become the dominant player in the NLP field. , But I hope Takema can promote this technology and quickly fallGround, accelerate sharing. “Bamboo is willing to work with partners to create solutions in a good ecology.”