Text|Guo Wei
The end of the year is coming, review, review, and outlook, and the "report card" of the insurance industry is about to become clear. ChatGPT was born for a whole year, insurance companies and intermediaries competed to enter the game, and the battle of 100 models was vigorous, which was regarded as a breakthrough in the transformation of the insurance industry, and opened the prelude to another deep change in the insurance industry.
At any technological inflection point, a new force will always be born, and thousands of sails will be crossed in the field of research, testing and application of large models. Among them,As an insurance technology platform that crosses borders from the Internet, Yuanbao AI has obvious genes and has always focused its resources on the application of artificial intelligence in the insurance industry, so that after the launch of the insurance industry, the premium scale quickly ranked among the first echelon of the Internet insurance industry, and successively appeared on the world's well-known lists such as KPMG and Hurun.
In October this year, Yuanbao launched the first "application evaluation report" for artificial intelligence large models in the insurance industry, systematically evaluating the application capabilities and performance of 10 mainstream large models from a third-party perspective. In December, Yuanbao became the only enterprise case in the insurance industry to be selected for the "Beijing Artificial Intelligence Industry Large Model Innovative Application in 2023)" by virtue of the application of the industry large model in the insurance field.
Standing on the first anniversary of the large model and the new year of 2024, the engineers of the Yuanbao AI team shared some of their thoughts: when an industry-wide large model competition quietly arrives, what great changes will occur in the competitive landscape of the insurance industry?What are the obstacles to the formation of a large-scale model ecology?The next stop, the opportunity of the industry is again
-insurance today-
The insurance industry is a big model
One of the best areas of application
For a long time, China's insurance industry has been "finding its way". The consensus in the industry is that the extensive style of play that uses the "crowd tactics" of the best people in exchange for market share has reached the time when it must be changed. You must know that at its peak, there were nearly 10 million people in the industry, but only 300 million insurance users served, and there was still a lot of room for improvement in service efficiency and business growthThe serious homogenization of products and services has caused the development of the industry to enter a bottleneck.
Intelligence is seen as the key to breaking the situation in the insurance industry. "There is a consensus among the company that as a manpower-intensive and data-intensive industry, the insurance industry is one of the best application areas for AI models. ”Wang Bo, CTO of Yuanbao, mentioned that from a macro perspective, the AI model will bring deep changes to the insurance industry at the industry level and the user level, and find a way out for the industry to break through the bottleneck.
First of all, the application of AI models will promote more inclusive and popular insurance. The Yuanbao AI team believes that compared with traditional artificial intelligence, the application of AI large models in the insurance industry will be more extensive and deeper, and more complex and refined risk assessment and pricing reference can be achieved through deep learning and analysis of large amounts of data, so as to help insurance companies better respond to risks, improve the cost performance of insurance products, and broaden the scope of insurable risks.
A few days ago, the "Implementation Opinions on Promoting the High-quality Development of Inclusive Finance" was issued, in which "improving the high-quality inclusive insurance system" was proposed for the first time. Wang Bo believes that "the application of AI models in the insurance field in the future will accelerate the improvement of the inclusive insurance system." ”
From the perspective of the industry, the ultimate goal of the large model is to continuously improve efficiency, break the existing operation mode, and bring infinite new possibilities for the development of the industry. Wang Bo said, "At the beginning, the large model was just to provide reference for the first people, improve human efficiency, and double the number of people served. Soon, the robot with the large model as the center, the service ability will not be inferior to the manual, and the efficiency is far beyond the manual, the robot can serve thousands of people at the same time, so the 'manual + large model service' model, can double the number of insurance users who receive the service, or even more. ”
The improvement of the efficiency of the industry by the large model is not only to make the work faster, but also to bring qualitative changes to all aspects of the whole process. Wang Bo explained, "A small idea, a small optimization innovation, can change the entire insurance industry." It's something we can't imagine. "For example, the ability to learn and analyze massive amounts of data through large modelsIn the future, insurance products may also be able to achieve "building block" construction options, and truly achieve "flexible patching" from the product side of insurance companies. Of course, these innovations are built on the basis of data security and user privacy. Since its establishment, Yuanbao has attached great importance to data security and has established a strict control system to ensure that there will be no data security risks in the whole process.
In addition,The deeper impact is that the gradual implementation of the general model will subvert the public's perception of insurance. Insurance will be simpler, more personalized and smarter, and the public's stereotype that insurance is difficult to understand, difficult to invest in, and difficult to pay for will be changed. Wang Bo believes that in the future, "the large model will not only help China's insurance industry enter the era of comprehensive intelligence, but also greatly improve the public's acceptance of insurance, and improve the popularity of insurance, thereby ushering in a new leap forward for China's insurance industry." ”
-insurance today-
Customer service has become the best "training ground" for large models
The possibilities are endless
Looking back on this year, the transformation of the big model for the insurance industry has begun. Lemonade, the U.S. insurtech giant, has developed dozens of prototypes of AIGC-based applications, allowing customers to complete the process of policy application, payment and other processes through communication with the robot without dialing **. Allstate Insurance Company has developed a generative AI application, MyStory, which allows customers to describe the accident to MyStory online and submit the summary document to all parties in the past, eliminating the need to repeatedly tell people from different departments about the accident.
The landing of the large model in the United States has given us hope and direction. However, the domestic insurance industry is different from foreign countries in terms of industry maturity and market characteristics, and the implementation of large-scale models in China's insurance industry still needs more operational innovation.
In China, a shortcut to the application of large models has been found, taking Yuanbao as an example, it has chosen "customer service" as the best application scenario for large models. "The most eye-catching application of the large model is the chatgpt dialogue scene, so it has the ability to process natural language and generate dialogues, which is very suitable for customer service scenarios. It can provide more sophisticated, intelligent, and personalized services than traditional customer service. Wang Bo said.
As early as the beginning of its establishment in 2019, Yuanbao has begun to research large models, from Google's BERT to OpenAI's GPT, from Tsinghua's 10 billion parameter large model GLM to Stanford's 7 billion scale "alpaca" model. At present, based on the open source model, combined with the long-term accumulation of high-quality industry data, Yuanbao continues to optimize.
As one of the earliest teams in the Internet insurance industry to develop its own customer service robots, as early as the beginning of this year, the accuracy of Yuanbao's customer service robots in terms of understanding and recognizing voice and text intent has reached 93%.It can recognize more than 30 dialects and accumulate more than 4,000 common problems in the insurance fieldThe customer bot is on standby 24 hours a dayUser satisfaction with customer service robots has even exceeded about 50% of human customer service in the market.
With the application of the general model, its ability and charm directly overshadow the traditional robot customer service. According to the Yuanbao AI team,The robot with the blessing of the large model has stronger comprehension, expression and professionalism, can better understand and generate natural language, give answers that match user needs, and can also explain complex professional problems in more easy-to-understand language.
It gives people an intuitive feeling that they are like a senior insurance expert who understands me very well and can explain professional issues in the vernacular. Wang Bo commented, and especially emphasized that the customer service robot based on the large model has good applications in professional question answering, product planning, medical knowledge answering and policy management.
For example, when a customer asks, "I have a broken bone and need to be hospitalized, what policies can I cover?"."In the past, most robots could only throw the links of all the user's insurance policies over, allowing users to click on them to view and screen, and the robot customer service combined with the large model can quickly point out which specific policies can be claimed according to the user's insurance situation, and there has been significant progress in professionalism, occupation and dedication.
Next, the challenges faced by the customer service robot mainly lie in the speech recognition of the elderly, dialect recognition and speech recognition in echo scenes, which will also be the direction of the Yuanbao AI team in the future.
Robots can assist people and make services more efficient;And the large model goes one step further and can also train people. According to Yuanbao's practice, the application of large models in customer service training scenarios has been quite effectiveFor example, the intelligent sparring system of the customer service team solves the problems of high training cost, difficult monitoring of training effect, and low training actual combat rate. In addition, the large model is also used to assist the office within Yuanbao Company, which has achieved efficiency improvement.
When talking about the manifestation of the application value of the large model this year, Wang Bo drew a continuous upward parabola with his hand, "The birth of GPT is indeed exciting, but after that, all of us have returned to calm - concentrate on studying and making progress every day, so that we can slowly see the difference, see the effect, and see the direction." ”
-insurance today-
How to break through the two major obstacles to the landing of the large model?
In the future, what resistance needs to be broken through in the extension of the depth and application breadth of the insurance industry model?In the view of the Yuanbao AI team, there are two breakthrough points that are the most critical.
First, we need to solve the data problem. Among the many fields that explore the application of large models, the insurance industry is a field that needs to break through many obstacles, the industry's data foundation is weak, data standards are not uniform, the problem of data islands is prominent, the level of data is low, and there is a lack of big data that can truly domesticate large models and have a certain quality. This also means that it is unrealistic to rely solely on the big model itself to drive efficiency change in the industry. In order for large models to be effective in the insurance field, there must be innovative ways to play.
Secondly, it is necessary to deeply integrate the vertical field of large models and insurance. The application of large models in the industrial field is by no means a simple technical problem, and the key is to create a complete system of "model-industry understanding-scenario application". This requires a dual drive of business and technical thinking, which requires not only understanding technology and recognizing the value of technology, but also having a good business thinking and knowing how technology can be developed around business to form synergy and synergy.
In view of the bottlenecks caused by the above two key points,Yuanbao's strategy is to work together with large models and traditional models in a short period of time, and form a set of intelligent systems from 3,000 models. Before the large model intervenes, the middle office manually identifies user problems. At present, Yuanbao has been realizedThe large model is used as the central control, so as to call the traditional model, so that the real needs of users can be identified more quickly, efficiently and accurately. In such a training process, the Yuanbao AI team will also add counterfactual data to test, improve the learning nature of AI, and improve the ability of large models. And this kind of play also solves the dilemma brought by the data level to a certain extent.
In the coming 2024, there will inevitably be new layouts and strategic priorities in the industry around large models. There are three main directions of Yuanbao:
The first is to continue to promote the application evaluation report of mainstream large models in the insurance industry. At present, the industry does need to have a more objective and clear understanding of the effect of large models at the application level, so as to smooth out the information gap and make decisions.
The second is to dig deeper into more application scenarios in the insurance industry and increase the depth of application.
Third, for a long time, Yuanbao has been continuously optimizing and integrating on the basis of open source large models, and next, more new models will be introduced, and continuous attempts will be made to extend the breadth of applications.
According to Wang Bo,With the optimization effect of open source large models getting better and better, the construction of professional vertical large models has been in the planning of Yuanbao. It is reported that this vertical large model is a model trained for specific tasks and data in the insurance field, and is more widely applicable to the entire insurance industry. "If it is successful, it will provide high-quality solutions for the industry and do something for the construction of a large model ecology in the industry. ”
-insurance today-
The large model will reconstruct the industry landscape
Small and medium-sized enterprises are expected to overtake in corners
Every breakthrough technological revolution in history is ultimately manifested in the liberation of people by technology, the replacement of people, the elimination of capabilities that are no longer applicable, and the emergence of new capabilities and types of work - the AI model will also go through the same process for the insurance industry. The huge impact of the large model on the insurance industry is not only reflected in the efficiency improvement and reform of the vertical industrial chain, but also in the horizontal reconstruction of the industry competition pattern.
In the foreseeable era of comprehensive insurance intelligence, the competitive advantage accumulated in the traditional human model in the past will no longer apply, and one model may be able to replace 100 or 1,000 people. Whoever can do more with fewer people wins. At this point, Wang Bo and the Yuanbao AI team believe ——The big model brings the insurance industry back to the same starting line, and small and medium-sized enterprises can even overtake in corners because of their more flexible organizational structure.
OpenAI was only a small company with dozens of people before, and it also made ChatGPT and many powerful products. Therefore, as long as the landing direction is found, the opportunity for small companies to overtake in corners is very large. In Wang Bo's view, "the general large model may be the Matthew effect, but the application of the large model in the vertical field is more innovative than the application, data and talent." It doesn't matter who does it first, and it doesn't matter who does it fast. In the end, it is who can cut AI into the application scenario and solve the problem for users is the only one who matters. ”
In the competition of large models in the future, what characteristics do the winning companies need to have?Wang Bo listed 4 aspects:
First, in terms of business philosophy, it is necessary to have both business thinking and technical thinking, so that technology revolves around the business
Second, in terms of talent strategy, people who can talk to AI and ask questions to AI are needed to be AI commanders rather than executors
Third, in terms of data reserves, we ensure that the data is diversified and the coverage scenarios are abundant, and only with high-quality and diversified data can AI have room to play
Fourth, adhere to long-termism in the strategic layout.
"We should neither overestimate the short-term effects of technology nor underestimate its long-term value. ”Wang Bo believes that "the 'disruptiveness' of any so-called 'disruptive' technology depends on the change to the real world after application, and this change requires time and patience." ”
Wang Bo also emphasized that the speed of change of AI models to the industry and society will be faster from the perspective of the effect of this year, but it will still take time. In 2024, the commercialization of large AI models will accelerate, and in 2026, it may enter the period of large-scale application.
Quantitative changes in technology will eventually form a qualitative change. Wang Bo pointed out that the large model is expected to greatly improve the public's acceptance of insurance and the popularity of insurance, thereby opening up a broader market space and helping China's insurance industry usher in a new leap. "It's still the eve of dawn, and we need to look at the power and future of AI with a long-term mindset. ”
Postscript
The first financial work conference held in October this year pointed out that high-quality development is the primary task of building a modern socialist country in an all-round way, and finance should provide high-quality services for economic and social development, and do a good job in science and technology finance, green finance, inclusive finance, pension finance, and digital finance.
With the top-level deployment of the financial industry, a new era full of hope has also opened. For the insurance industry, the exploration of insurance technology has become a key direction of high-quality transformation in recent years. Nowadays, with the acceleration of policy traction, it is foreseeable that the further expectations of the insurance industry for technology and finance will be interpreted to a considerable extent by the large model representing the peak productivity of the moment.
And this expectation is gradually becoming a reality through the practice of "Yuan Bao" one after another in the current industry. This kind of thinking and exploration of technological innovation and industry breakthrough, and the focus on technology and the market, should not only be the persistence of Yuanbao, but also the background color of the industry, and will also be the most powerful interpretation of the spirit of the Internet and the value of insurance.