The Heart of the Machine is original
Author: Zhang Qian
A few years ago, there was a kind of topic that was very hot: "In the xx industry, AI is replacing humans" But as people's understanding of the current development of AI gradually becomes more rational, such topics are becoming less and less common. Instead, "What aspects of ai can enhance or extend people's abilities?" "What problems can AI and people solve together?"
This downward revision of expectations is inevitable, after all, with the current weak artificial intelligence, it is unlikely to create the highly intelligent "receptionist" in Westworld. Terence Terence, author of Deep Learning and known as the "Father of THE World's AI", Terry Sejnowski also said back in 2019 that "in the future, humans and machines will be cooperative rather than competitive."
The image of the receptionist (robot) in the American drama "Westworld".
Today, this view is being accepted by more and more people in the industry, and the concept of "Collaborative Intelligence" is beginning to receive more attention. Cécile Paris, chief research scientist at CSIRO, Australia's largest national research institution, even points out that "[collaborative intelligence] will be the next scientific frontier of digital transformation".
Cécile Paris and others argue that to better develop "collaborative intelligence," we must rethink workflows and processes to ensure that humans and machines complement each other. In addition, exploring how machines can help people develop new skills that may be useful in various areas of the workforce is also an important direction.
In China, this kind of thinking has already begun, especially in finance, insurance and other industries that directly serve people. These industries have taken some detours in the application of AI. For example, externally, intelligent customer service and robo-advisors were once all the rage, but whether from the perspective of customer satisfaction or business depth, these tools could not meet the needs of enterprises; internally, each enterprise may have a dedicated data analyst, who directly deals with the data center (machine) of the enterprise. But as soon as there are more lines of business, the pipeline of "machine-data analyst-business person" will be blocked by insufficient data analysts. This demonstrates the importance of optimizing human-machine collaboration processes and processes and helping ordinary business people develop new skills. And this thing, in fact, there are already many companies doing.
At a press conference some time ago, ZhongAn Insurance, the first Internet insurance company in China, demonstrated two new products aimed at solving such problems. Behind these two products is a condensation of the company's years of thinking about changing the way human-machine cooperation is done and empowering ordinary business people.
Rethink the platooning of man and machine
If the allocation of resources in an enterprise is likened to a platoon, the man and the machine may be two completely different branches of the army. Cavalry can be invincible in the vast steppes, but once in the jungle it is not as agile as infantry. Similarly, machines can work 24 hours a day, processing massive amounts of data, but are not good at communicating directly with people. Therefore, before introducing machines, it is imperative to figure out which links are most suitable for the machine.
In its own customer service, outbound calls and other scenarios, ZhongAn sorted out this problem: the reason why many intelligent customer service agents were not successful before was because they exaggerated the role of the machine and ignored the flexibility and creativity of people. So, this time, they rearranged the position of the person and the machine: the machine was responsible for "standing guard" for 24 hours, continuously mining customer intentions, emotions and other information from text and voice interaction data, and then handing over the problems found to people to solve.
This is the core idea of the user interaction mining platform launched by ZhongAn this time - Whale Detective.
Unlike many previous interactive machines, whale scouts are primarily tasked with "discovering," just as whales search for prey in the ocean. It will try to figure out several questions, such as who is the customer? What business do you want to do? Has the problem been resolved? Are there any negative emotions that arise? Will there be a complaint? Do I need to introduce them to new products? These are condensed into an analytical framework called CIREO. In order to figure out these problems, Whale Probe integrates various popular technologies in the field of AI such as deep semantic understanding, multimodal machine learning, large-scale pre-training, knowledge distillation, graph optimization, and model compression.
After figuring out these problems, Whale Detective did not directly solve them, but labeled them (insurance, consultation, surrender, claims, renewal, etc.) and generated charts (intention rankings, customer group charts, emotional barometers, public opinion cloud maps, etc.) for reference by various departments (such as sales, customer complaints). As for what to do next, it is up to people to decide. This new approach to human-machine collaboration has helped ZhongAn reduce the complaint rate by 30% and transform the business from list to customer.
This platoon deployment reveals the consideration of ZhongAn as an AI application company for the input-output ratio. Duan Chaoyang, the company's chief data officer, said, "In the choice of AI application scenarios, there are some more extreme situations in the market, one situation is that everyone is concentrating on very difficult and complex scenarios, such as completely driverless cars, or completely replacing people with machines to interact with customers, which requires very large investment." In his view, "it is very important to distinguish the boundaries of technology, the boundaries of theory and the boundaries of practical applications", which makes ZhongAn choose "the right and high point of production to cut in". Whale Scouting is a good example.
Duan Chaoyang, Chief Data Officer of ZhongAn Insurance.
Break the dimensional wall between man and machine
After completing the platooning of people and machines, the company has one more thing to do, that is, to break the dimensional wall between the two.
At the beginning of the birth of computers, although ordinary people can also enjoy the transformational results brought about by technological progress, it is difficult to interact directly with them and participate in the creation of value. Ai also faced this challenge in the early stages of landing.
In order to break this dimensional wall, people have made a lot of efforts, and low code is one of the more popular ones.
Low code is not a new concept, and a similar pattern was already in place 20 years ago. Entering the era of "digital intelligence", the internal needs of enterprises are becoming more and more abundant and changeable, and this concept has gained new attention. However, due to the large number and complexity of industry scenarios, it is difficult to have a common platform that can cover all industries and meet the needs of all users. This is especially true of the Internet insurance industry where ZhongAn is located.
Within ZhongAn, the most urgent scenario for low code requirements is data analysis. Because a lot of times, their business people are demanding to know not only what happened, but also why it happened (attribution) and what the future holds (prejudgment). This "from seeing to foresight" analysis needs are very heavy: each link in each business line may have different analysis needs, and different management levels focus on different data dimensions. However, the number of analysts who can afford to take on these needs is limited. If this ability to communicate with machines can be extended to everyone in the business, the efficiency of the analysis will be greatly improved.
So they built their own low-code platform called Jizhi. It allows business people to create their own data analysis frameworks in a drag-and-drop manner, reducing the strong reliance on data analysts. In order to improve its ease of use, the developers have also added light deployment, multi-terminal usable features in the just-released Jizhi 2.0 version, and preset templates for ten major scenarios, so that business personnel only need to go through a simple four-step process to generate data models and visual dashboards. With this platform, ZhongAn's internal data analysis efficiency has increased by 50% and labor costs have been reduced by 40%.
In ZhongAn, there are many similar applications that redefine the relationship between man and machine, and they and the algorithms, platforms, and guarantees that accompany them together constitute a large framework - ZhongAn Digital Intelligent Empowerment Methodology 4633. In this methodology, the algorithm layer encompasses atomic power one by one, unencapsulated machines. The platform packages these machines into modules for the business layer to call in the form of building blocks, so as to realize the application value of increment (customer acquisition, repurchase, and user value), loss reduction compensation (risk control), excellent operation (improving the efficiency of operational decision-making), and market expansion (creating new products for the business, such as new insurance types). More importantly, this set of methodologies and its products (including Whale Detective and Jizhi) are not only applied to ZhongAn internally, but also oriented to the industry output through three dimensions: light consulting services, platform construction, and model application.
In this direction, foreign exploration is also taking place, such as the team of Cécile Paris mentioned above. On November 30, she and other scientists formed CINTEL (short for Collaborative Intelligence), a future science platform with a budget of $12 million ( about $54.42 million). Through this platform, they will explore how humans and machines work and learn together, and how this collaborative approach can improve human work.
The core of industrial digital intelligence is human-machine collaboration
Whether it is rethinking the platooning of people and machines, or breaking the dimensional wall between man and machine, it is essentially through the innovation of processes and processes to find a better way of human-machine collaboration.
In the view of Dr. Zhou Bowen, an expert member of the National New Generation Artificial Intelligence Development Research Center, the in-depth collaboration between people and AI will become the core of the digital intelligence of the industry. He divides the collaboration between humans and AI into three levels, one is that AI helps humans get rid of simple, repetitive, and boring work; the other is that AI and people have an enhanced effect on collaboration, relying on AI capabilities to help humans improve work efficiency; and third, on the basis of enough data and observations, based on AI to reshape workflows and complete innovative collaboration. Judging from the efforts of zhongan and other enterprises, domestic enterprises have achieved some results at these levels, but there are still many things to clarify on the road to comprehensive digital intelligence.
Reference Links:
https://techxplore.com/news/2021-11-secret-ai-doesnt-job.html
http://www.news.cn/english/2021-11/30/c_1310342090.htm
https://baike.baidu.com/item/%E6%95%B0%E6%99%BA%E5%8C%96/56172035?fr=aladdin
https://www.jiemian.com/article/5965239.html