(Original title: Artificial Intelligence Internet Consumer Finance "pseudo-application" survey) Reporter Chen Zhi reported from Shanghai "Now that artificial intelligence is highly valued, we don't want to vote for it." A person in charge of an internet consumer financial institution said with emotion that they had planned to invest in an institution based on artificial intelligence to improve the efficiency of risk control. A jump was much higher than the valuation of their entire platform. Today, he is more fortunate that he did not invest. The reason for this is that the rapid rise of artificial intelligence will inevitably lead to mixed industries - especially the "pseudo-application" of artificial intelligence in the Internet consumer finance field. For example, a face recognition agency has announced that it can carry out remote signing for consumer financial platforms. Face recognition success rate exceeds 90%, but after testing on many platforms, the lighting angle of the borrower's scene has no small impact on the accuracy of face recognition, resulting in the platform can only use the line to sign the next way to assess the existence of the borrower Fraud. 21st Century Business Herald reporter also noted that in the United States, equity investment institutions have also begun to become increasingly vigilant about the “pseudo-application†of artificial intelligence in consumer finance. In particular, some of the US consumer financial institutions that advertised "artificial intelligence" still have their risk control core based on the judgment of human experience and assess the borrower's loan risk. In addition, many consumer financial platforms that claim to have high accuracy in image recognition and language recognition have not effectively reduced operating costs and improved operational efficiency. "Pseudo-applied" emerged quietly According to the CTO of this large-scale consumer financial institution, the application of artificial intelligence in the Internet consumer finance sector mainly focuses on customer acquisition, customer service, risk control, and collection. For example, in the risk control process, artificial intelligence through big data analysis and deep learning of machines, continuously optimizes risk control efficiency and reduces bad debt rate. At the same time, it can give borrowers more accurate risk pricing; in the collection process, artificial intelligence can target different assets and occupations. The overdue borrower of the age and the age has set up a personalized collection and repayment plan, and sends the dunning information to the borrower at an appropriate time, which can not only take care of the face of the borrower but also remind them to repay the loan as soon as possible to increase collection efficiency; , Face recognition based on artificial intelligence, speech recognition technology can help the platform to complete the remote visa, reduce the cost of offline manual operations. "However, it is indeed a challenge for AI to fulfill these expectations well," he said bluntly. Many heads of Internet consumer financial institutions confided to 21st Century Business Herald reporters that they had experienced artificial intelligence pseudo-applied phenomenon. It is more common for artificial intelligence organizations to seize the platform's "psychological" fear of violent collection, and recommended the collection based on artificial intelligence. The model, claiming to significantly reduce the number of collection teams. "But its actual application effect is not ideal enough." A person in charge of a consumer finance platform told a 21st Century Business Herald reporter. Later, he learned that the agency is copying the artificial intelligence collection technology of the United States, but the consumer finance environment in China and the United States is very different. The United States has a mature personal credit information system, and the borrower’s breach of contract will be blacklisted. Difficulty in daily life and tourism; China's personal credit system is not perfect, leading to an increase in fraudulent borrowers, coupled with lower default costs, the United States artificial intelligence collection model may not be suitable for China's national conditions. "Actually, there are still many black intermediary agencies in the domestic consumer finance field. They face up to artificial intelligence in the face-signing and risk control process, and they will specially produce training materials to teach borrowers how to pass wind-control audits. If the consumer financial platform completely relies on artificial intelligence risk control technology, , it is likely to encounter a lot of bad debts." He further pointed out. In fact, many domestic risk control models that flaunt artificial intelligence, the core risk assessment criteria are still whether the borrower has the records of the People's Bank credit, sesame credit score is more than 660 points, it is difficult to highlight its role in enhancing the efficiency of wind control . According to the CTO of the above large-scale consumer financial institutions, artificial intelligence must play a role in the Internet consumer finance field. Three major conditions are required. One is massive massive data, including borrower socialization, past consumer behavior, occupation, and social payment records. Marital status, age, etc., and the platform can find valuable data from it as an important basis for assessing the borrower's willingness to repay and repayment ability. Second, the platform needs to have a suitable risk control model to match these data. Constantly optimize risk control, customer acquisition, collection and customer service efficiency through big data analysis. Third, the platform must have enough professionals to continuously improve the artificial intelligence technology to ensure that the entire business can catch up with market changes. The great gap between ideals and reality In the eyes of people in the industry, the reason why artificial intelligence institutions serving the consumer finance sector in the Internet can obtain high valuations is another important reason that many consumer financial platforms have high expectation for the deep learning of artificial intelligence machines. Many organizations even believe that although many artificial intelligence technologies may now seem to be pseudo-applications, with the continuous enhancement of deep learning capabilities of machines, it may one day be able to significantly increase the operational efficiency of the platform and become the platform's new core competitiveness. "However, machine deep learning can sometimes also play a double-edged sword." The chief technology officer of the large consumer financial institution pointed out that, for example, in the field of intelligent wind control, the risk control conclusions of machine deep learning are often difficult to explain. . The above-mentioned person in charge of the consumer financial platform also feels deeply concerned about this. Prior to this, they specifically developed a line of consumer credit products for small and micro business owners. At that time, the product R&D team believed that the risk of repayment of small and micro business owners with houses and cars was low. However, the artificial intelligence risk control model has come to the opposite conclusion: the bad debt rate of small and micro enterprises without real estate will be lower than that of real estate. "At first we could not figure out why we would have such a conclusion," he recalled. Later, many people discussed that the logic of artificial intelligence is that many small and micro business owners like to take real estate to different institutions to repeat loans, leading to a sudden increase in the risk of bad debts. However, the artificial intelligence risk control model through which risk factors and borrower data to draw such views, they have not understood so far. “In-depth learning may also have problems with over-completion,†said the chief technology officer of the large-scale consumer financial institution. For example, whether different borrowers have MPF payment records and the consumption records of different borrowers in Jingdong and Tmall’s different e-commerce platforms. All of them will get completely different wind control conclusions through artificial intelligence. Many of the wind control conclusions seem quite reasonable in the experimental demonstration stage, but they cannot withstand the test of practice. In his opinion, whether the deep learning model of the machine can be effective or not depends on whether the technical team can effectively interfere with the artificial intelligence and adjust the risk factor parameters effectively to make the risk control model and conclusion closer to the actual environment. "Actually, many consumer financial institutions also know the bottleneck of deep-learning technology of artificial intelligence machines, but they are deliberately magnified their active role because this helps to increase the overall valuation of consumer finance platforms and obtain more equity financing. The heads of a number of consumer financial institutions bluntly said that this is one of the biggest driving forces for the continued rise in the valuation of artificial intelligence companies in the consumer finance sector of the Internet. (Editor: Yan Yibo)