Advances in modern technology have made financial crime more sophisticated. In turn, the development of protective mechanisms and new approaches to economic and financial security is also not standing still. Finance and economics are some of the most attractive targets for various types of cybercrime. In this article, we will explore how artificial intelligence can help prevent cybercrime in the financial sphere.
The Current Situation with Artificial Intelligence – Cure the Disease, Kill the Patient
The Guardian has reported that the development of artificial intelligence technologies can lead to the emergence of new forms of cybercrime, political unrest and even physical violence over five years.
If developers do not follow security recommendations, criminals can use AI for nefarious purposes. For example, AI can automate the detection of critical software weaknesses or select potential victims for computer crime and cyber fraud.
Refinitiv has released the Financial Crime Report 2019 on financial crimes and technological capabilities to counter them. Over the past 12 months, 72% of organizations in the world have faced financial transactions such as fraud, money laundering, corruption or tax evasion.
According to this cybercrime report:
- 67% of respondents believe that the best way to combat cyber fraud is the use of cloud data and appropriate information storage and processing technologies.
- 56% called artificial intelligence and machine learning tools worthy rivals of bank fraud.
- 47% talked about API technologies that allow some programs to interact with others.
- 40% consider blockchain to be an effective measure to address “what is cybercrime” in the nearest future.
Thus, we can observe a dual situation. On the one hand, vulnerabilities and flaws in the algorithms of artificial intelligence and machine learning give rise to the development of new types of risks in banking and crime online. But on the other hand, it is these systems that are capable of fulfilling ambitious hopes in terms of ensuring the security and transparency of the financial sphere.
What Are the Main Vulnerabilities of AI-Based Systems Used in the Financial Sector?
Modern artificial intelligence is able to work with any amount of data, recognize voices and visual objects, help clients, draw its own conclusions and even develop business strategies.
However, until now the banking industry has not been in a hurry to test new solutions in the field of AI. They fear the so-called "black box effect." A black box is a device that is not transparent to the end user. Non-transparent AI is built on neural networks, deep learning algorithms, genetic algorithms, model ensembles, etc.
A common feature of such methods is that the “logic” of their predictions and the decisions they make is difficult to explain in simple words, and in some cases, it is simply impossible. Such a device does not disclose the relationship between the data entered into it and the output data. The user simply cannot see or understand it and therefore refuses to trust the results of the device.
Based on this, banking systems based on artificial intelligence can be viewed as the following two types of attacks:
- Adversarial attacks. In this case, the danger is that after analyzing information, the algorithm makes extremely erroneous conclusions. For example, in the case of visual recognition, if the program makes an incorrect conclusion about an object (for example, it claims that it is an image of a dog, when it is an image of a cat), the entire subsequent chain of conclusions and recommendations will also be erroneous.
- Data poisoning attack. In the event of such an attack, artificial intelligence systems are influenced by data that human intelligence considers immoral and unacceptable. One of these cybercrime examples was the case on Twitter in 2016, when a chatbot learned to advocate Nazi views.
Thus, if an opaque system gives an erroneous conclusion, for example, it gives a false positive signal and cannot explain it, communication with the regulator will become impossible. A transparent system will explain that the basis of its conclusion is a criterion, according to which, for example, the date of birth of the client being checked should not fit into the plug +/- 5 years. If the operator then considers that this criterion cannot be used, he can correct it (depending on how much he is willing to risk). So, AI in the field of AML banking must ensure full clarity of its decisions to bring cybercrime numbers and risk reduction to acceptable levels.
Blockchain Possibilities for the Financial Sector in Conjunction with AI
- Blockchain is a strict registry, and impossible to deceive (although the practice has faced attempts in the case of Bitcoin). In any case, the registry is unified. It is chronological, straightforward and practically not subject to any adjustment. The entire registry is in peer-to-peer networks. If a branch starts to change the registry, the system responds harshly to the change and deletes the branch. It is impossible to enter the server and fix something. Each computer is part of a unified system.
- Another important point is that the hierarchy disappears. Each element is part of a whole. Peer-to-peer networks are a substitute for the hierarchy of networks where there are server-machines and client machines. In the absence of a hierarchy, each element must follow clearly established rules, otherwise, it will be excluded from the system. The banking system runs exactly according to such laws. Everything else is considered unfair competition.
- The most important benefit is the consensus mechanism. It is based on the principle that every change within one link of a chain can occur only in agreement with other links.
Why AI and Blockchain Can Help in the Fight Against Financial Crimes
The most important prerequisite regarding the benefits of AI for dealing with financial crimes is that AI-based algorithms must be built on a matrix that prevents the entire system from malfunctioning. In the case of AI on the blockchain, this matrix should be based on four main principles of additional control:
- S (Severity). According to this principle, the basic concepts that were laid in the algorithm at the beginning of its training cannot be replaced by any other. This principle minimizes the risk of a data poisoning attack.
- O (Open to Deep) means that the network should be able to build long chains of judgments based on the data received.
- I (Irrational Ability). This principle suggests that the algorithm should be able to discard a more probable or advantageous conclusion if it contradicts the initial data with which the program was trained. However, this data can be used to make further judgments.
- D (Definability). The neural network during the course of reasoning can outgrow the boundaries of the matrix in understanding the application area to achieve the goal.
Blockchain-based artificial intelligence systems can make the banking system transparent and virtually invulnerable to external penetrations. However, it is necessary to lay the right learning mechanisms and implement learning on the correct data.
If you want to learn more about this topic or discuss your AI/Blockchain project, Archer Software team is waiting to hear from you! Let’s get in touch!