Technology Application | Research on the application of artificial intelligence technology in anti-money laundering work

Text / China Anti-Money Laundering Monitoring and Analysis Center Ye Gang Xiang Lu Qin Wei

With the continuous development of computer technology, artificial intelligence technology is gradually becoming a strategic technology driving a new round of scientific and technological revolution. General Secretary Xi Jinping has repeatedly emphasized that "it is necessary to deeply grasp the characteristics of the development of a new generation of artificial intelligence, strengthen the integration of artificial intelligence and industrial development, and provide new momentum for high-quality development". In recent years, artificial intelligence has been widely used in banking, insurance, securities and other financial fields, constantly giving birth to new financial products and gradually forming a new financial format. This trend represents both an opportunity and a challenge for anti-money laundering efforts. This paper summarizes the development status of artificial intelligence technology and its main application scenarios in the field of anti-money laundering, analyzes the opportunities and challenges brought by artificial intelligence to anti-money laundering work, and puts forward the basic process and work suggestions for building an anti-money laundering information system based on artificial intelligence technology.

The formation and development of artificial intelligence technology

Artificial intelligence is a cutting-edge discipline in the research and development of theories, methods, technologies and applications that simulate the thinking of the human brain. It attempts to build intelligent machines that are similar to the way humans think on the basis of studying the nature of human brain intelligence. Research in this area includes machine learning, natural language processing, evolutionary computing, and intelligent robotics, among others.

Since the concept and theory of artificial intelligence were first proposed, the development of the technology and industry has gone through the following three main stages.

The first stage was from the 50s to the 70s of the 20th century. This stage is the stage where the concept of artificial intelligence is proposed and formed. It is generally believed that the Dartmouth Conference held in 1956 promoted artificial intelligence as a widely studied discipline. Since then, there has been a boom in research on related technologies. However, due to the limitations of processing power, data volume and other factors, computers at that time could not effectively complete the computing tasks of artificial intelligence algorithms. Therefore, the research of artificial intelligence at this stage is more at the theoretical level.

The second stage was the 80s and 90s of the 20th century. At this stage, with the improvement of computing and storage capabilities, scientists began to explore how to use artificial intelligence technology to build expert systems to assist humans in solving practical problems. This type of system learns the solution of existing problems and forms knowledge and stores it in the knowledge base. When encountering similar problems, it can automatically analyze, make decisions, and solve them. However, the use scenarios of expert systems are only limited to solving problems that have been solved by other means, and due to the high hardware cost and limited application scenarios, artificial intelligence has not been able to achieve industrial-grade applications at this stage.

The third phase runs from 2015 to the present. The landmark event of this stage was in March 2016, when Google's AlphaGo defeated Lee Sedol, a Korean professional nine-dan Go player, in a human-machine battle of Go. Subsequently, in November 2022, OpenAI released the chatbot program ChatGPT, which can imitate the way humans think and talk, chat with users, and even complete tasks such as email writing, copywriting translation, and essay writing, once again setting off a global AI boom. With the explosive growth of data volume, the rapid improvement of computing power, and the continuous emergence of new algorithms, the research field of artificial intelligence continues to expand. In particular, the development of deep learning technology has brought artificial intelligence into a new stage of research, and gradually formed a complete industrial chain division of labor and collaboration system.

The main application scenarios of artificial intelligence in the field of anti-money laundering

Money laundering refers to the act of criminals concealing and transforming illegal gains through various means to make them formal. Anti-money laundering is carried out based on artificial intelligence technology, and illegal activities related to money laundering are monitored and prevented by artificial intelligence technology. The main application scenarios of artificial intelligence in the field of anti-money laundering include the following:

Automatic monitoring of trading behavior: Artificial intelligence technology can analyze massive financial transaction data in real time and find abnormal trading behaviors in it. By continuously analyzing and learning from historical data, AI technology can assist analysts in discovering known and unknown patterns of money laundering, identifying suspicious behaviors that are difficult to detect with traditional rule-based methods, and improving the efficiency of monitoring and analysis. In anti-money laundering work, the use of artificial intelligence technology to implement transaction monitoring can automatically process transaction data, quickly detect anomalies, and warn of new money laundering risks, so that analysts can focus on the results screened by artificial intelligence technology and improve work efficiency.

Automated processing of transaction reports: Artificial intelligence (AI) technology enables computers to understand, process, and generate linguistic information just like humans. Therefore, it can be used to analyze and process human language carriers such as documents and voices, and to process the obtained information such as information mining, subject classification, and quality evaluation. In anti-money laundering work, artificial intelligence technology can be used to pre-process suspicious transaction reports such as report quality evaluation, key content extraction, and classification of crime types, so as to prompt analysts of information such as report quality and key content of reports, and effectively improve the processing capacity of suspicious transaction reports.

Automatic analysis of transaction networks: Artificial intelligence technology can draw and analyze complex networks of entities, mine and analyze potential relationships or hidden connections between transaction subjects, and then identify high-risk individuals or entities. In anti-money laundering work, entity and network analysis technology can be used in the fields of penetrating fund analysis, money laundering criminal gang analysis, subject portrait and other fields, providing analysts with networked analysis methods.

Automatic assessment of money laundering risk: At the transaction level, AI technology can automatically assess the amount of money laundering risk related to entities, accounts or transactions by analyzing various factors such as transaction history, subject lists, geographic information and relationship networks, predict future transaction trends and risks, and take early warning measures in advance. At the institutional level, AI technology can combine the evaluation index system and reference scoring points to qualitatively or quantitatively evaluate the overall money laundering risk control ability of financial institutions, which is conducive to promoting financial institutions' anti-money laundering performance and maintaining economic and financial security and social stability.

In short, AI has great potential in combating money laundering crimes and improving the efficiency of risk management and control, which is conducive to improving the comprehensiveness, accuracy and effectiveness of anti-money laundering work.

Figure: Anti-Money Laundering AI Architecture Diagram

The basic process of building an anti-money laundering system based on artificial intelligence technology

With the further development of artificial intelligence technology, the construction of an anti-money laundering system based on this technology will become an effective means to improve the overall anti-money laundering monitoring and analysis capabilities. The basic process of building such a system can be broken down into the following main steps.

Clear goals: It is important to have clear goals for system building. Only by clarifying the goal can we formulate an effective system construction plan for travel, and ensure that the system development and application do not deviate from the original intention of the construction. This involves determining what problems the system is designed to solve, what results to achieve, and so on. In general, the goal of building an anti-money laundering system based on artificial intelligence technology is to comprehensively analyze all kinds of financial data, automatically monitor suspicious transaction behaviors, identify relevant money laundering patterns, and automatically warn scenarios that may have money laundering risks, so as to improve the efficiency of anti-money laundering monitoring and analysis by means of informatization.

Data collection and preparation: Developing, training, and testing AI-related technologies requires high-quality, diverse, and massive amounts of data. Therefore, to build an anti-money laundering system based on artificial intelligence technology, it is necessary to first collect and process basic data. At present, there are two main sources of information for anti-money laundering: one is transactions and related information, which mainly comes from the elements directly related to transactions in large and suspicious transaction reports. For example, the transaction amount, the subject of the transaction, etc. The other category is other supplementary information, including data in the form of natural language texts, multimedia, etc., as well as various types of data from third parties, such as "suspicious point analysis" and "description of funds and behaviors". After the above-mentioned raw data is collected, it is necessary to shave, complete, disambiguate, annotate and other processing processes according to certain rules, so as to prepare the data for the development of artificial intelligence systems.

System development: System construction needs to comprehensively use artificial intelligence technologies such as machine learning, deep learning, and natural language processing to carry out feature extraction, system development, and system deployment in order to achieve the established goals. Among them, feature extraction extracts the data features related to anti-money laundering by analyzing the prepared data, such as transaction frequency, transaction behavior characteristics, subject age characteristics, etc., so as to convert the original data into eigenvalues that are more suitable for machine calculation. The calculated eigenvalues will be used for the training and evaluation of the AI algorithm, and the structure or parameters of the AI algorithm will be continuously adjusted until the best prediction effect is achieved. After the algorithm matures, the system can be deployed in the production environment for anti-money laundering monitoring and analysis.

System application and continuous improvement: After the system is deployed and launched, analysts can use the system to carry out real-time monitoring and analysis services. The system can not only assist analysts in data collection and collation, but also assist analysts in case analysis and report writing. At the same time, in the interaction with analysts, AI technology continuously improves itself by learning the analytical ideas and work habits of analysts. By analyzing performance indicators, user feedback and other information, system development and maintenance personnel determine the areas that need to be enhanced or expanded, introduce new intelligent algorithms in a timely manner, improve existing algorithms, and realize continuous optimization and upgrading of the system.

Risks and challenges posed by artificial intelligence

While AI technology can enhance anti-money laundering efforts, analyst engagement and expertise remain crucial. Without the intervention of human decision-making, the system may miss important clues or even lead to the wrong conclusions. In general, the main risks and challenges faced by building an anti-money laundering system based on artificial intelligence technology include the following.

Data quality-induced bias: The results produced by AI technology are highly dependent on the data used. Arguably, data is at the heart of AI systems. High-quality data is a prerequisite for developing AI systems that meet expectations. As the saying goes, flaws or errors in the data can lead to biased decision-making in the system, and may even produce completely wrong results. For example, it has been shown that ChatGPT will give illogical or incorrect feedback or even make up facts when relevant data is not available or when it is disturbed by disinformation. Therefore, when using AI technology, it is necessary to screen and validate data to ensure the accuracy and reliability of the data to reduce the risk caused by data quality issues.

Explainability of conclusions: The computational process of AI technology is often complex, making the decision-making process difficult to understand and explain. In the training process, in order to find the relationship between the input data and the output conclusion, many artificial intelligence algorithms often fuse multiple input data items to construct a complex mathematical model, so that the input data can better conform to the expected conclusion after the calculation of the model. For example, when an AI algorithm is trained on a batch of transaction information, the algorithm may end up with a computational model that is difficult to interpret as a method for identifying transactions with a higher risk of money laundering. However, people often cannot intuitively understand the logic behind such mathematical models.

The continuous evolution of money laundering methods: With the continuous development of the global economy and finance, the upstream crimes involved in money laundering activities have extended from the fields of economy, drug trafficking, and smuggling to the fields of public involvement and duty crimes. In particular, with the emergence of trading carriers and trading platforms such as cryptocurrencies, virtual assets, and the dark web, they are increasingly becoming a hotbed of money laundering crimes due to their strong concealment and difficult tracking. In the face of this situation, AI systems may not be able to effectively respond to new types of money laundering unless they are constantly adapting and keeping up with these changes.

In conclusion, AI-based information systems are not one-and-done and need to be updated and upgraded regularly to minimize the risk of misjudgments and misstatements.

Job proposals

1. Strengthen research on artificial intelligence technology. In view of the opportunities and challenges brought about by AI technology, we should further strengthen research in related fields and the tracking of cutting-edge technologies, so as to thoroughly understand the technology, so as to control the various risks that may be encountered in the application process within an acceptable range. To this end, the following methods can be adopted to promote the research work: first, formulate a practical strategic plan for the development of artificial intelligence technology, clarify the technical route and key tasks, strengthen the coordination and cooperation between business and technical departments, and further promote the research of artificial intelligence technology. The second is to strengthen cooperation with relevant enterprises, universities and scientific research institutions to jointly study the latest developments and development trends of artificial intelligence technology, and jointly build and share research results. Third, according to its own business needs and characteristics, select typical scenarios to carry out application research on artificial intelligence technology, and continuously explore the application practice of this technology in anti-money laundering monitoring and analysis.

2. Strengthen the construction of talent team. Artificial intelligence is a comprehensive discipline that integrates professional domain knowledge and computer technology, involving many fields, such as computer science, social science, engineering, etc., which puts forward high requirements for both builders and users. Only compound talents with rich business experience and proficiency in related technologies can make the greatest value of artificial intelligence technology. To this end, the following ways can be taken to strengthen the talent reserve: first, according to their own business needs, formulate a talent training plan, and improve the technical literacy of the talent team by increasing internal training and introducing external resources. The second is to strengthen cooperation and exchanges with external relevant units, broaden technical horizons and work ideas, and improve research level and application capabilities through cooperative research, exchanges and visits.

(This article was published in the first half of April 2024)