Can AI Cross Borders?

One of the most important applications of AI in payment systems is the automation of processes that lead to faster and more reliable transactions. For example, AI can help automate data validation to instantly detect and correct errors in transaction data, eliminating delays. This is particularly important in cross-border transactions, where data integrity is critical. In addition, AI models can predict and adjust exchange rates in real time, significantly increasing the efficiency of currency conversion.

Another key area where AI is playing a transformative role is in fraud detection. Deep learning algorithms continuously analyze transaction patterns to detect fraudulent activity. They can also identify subtle anomalies – for example, an unusually high number of small transactions in a brief period of time from an account that typically makes only a few large payments. This provides an additional layer of security. Unsupervised learning models are particularly effective because they can detect such transactions that were not previously flagged as fraudulent. This enables effective fraud prevention and improves risk management.

AI also offers significant benefits when it comes to regulatory compliance. AI systems continuously monitor transactions for illegal activity and ensure that they meet all legal requirements. This is particularly important in international payments, where different countries have different regulatory requirements. Natural language processing algorithms can analyze regulatory documents and help ensure compliance with changing regulations. However, their use in a cross-border context is not yet fully developed.

Traceability and transparency of models are critical when using AI. Algorithms, especially those based on deep learning architectures, are often black boxes, meaning that their decisions are opaque. The issue of liability is a key challenge, particularly in cross-border payments. If AI algorithms trigger erroneous and untraceable transactions, banks face dire consequences. To ensure the traceability of complex AI-driven decisions, regulators are increasingly emphasizing the need for transparent AI systems. Techniques such as local interpretable model-agnostic explanations (LIME) and Shapley additive explanations (SHAP) aim to break down the black-box nature of AI models and make their decisions comprehensible to users and regulators. Research is currently focused on fraud detection in credit card transactions, while its use in production payment systems remains to be seen.

Gabriel Juri
SIX