Quantum Machine Learning: The Revolution of the Financial World through AI and Quantum Computing

In addition to companies, legislators in major regions around the world are also preparing for quantum computing. In most cases, the focus is on the dangers posed by quantum computers and their sheer computing power. Although no quantum computer is yet ready for series production, the coming IT age is already casting its shadow: In particular, “Harvest now, decrypt later” (capturing encrypted data today in order to decrypt it in a few years using quantum computing) poses enormous challenges for the economy. As my colleague Elisabeth Wolter writes in her blog post, banks and insurance companies need a roadmap to prepare for quantum computing.

Despite all the dangers, however, it is also important to bear in mind the potential advantages associated with quantum computing. One of the great strengths of quantum computers is their ability to recognize connections in a fraction of a second while also considering a multitude of possible variables. This makes them the ideal sparring partner for another technology that has caused a sensation recently and still almost single-handedly determines stock market prices: artificial intelligence. Together, they are converging into quantum machine learning. In the future, quantum computers will help train more advanced AI models. Possible areas of application for this are already emerging today.

Nouhaila Innan, Muhammad Al-Zafar-Khan, and Mohamed Bennai have already published a study in the International Journal of Quantum Information in which the authors claim to have proven that certain QML models are capable of distinguishing between genuine and fraudulent financial data with a confidence level of 0.98. Results like these show the potential of QML and how banks can benefit from it in areas such as payment transactions:

• Complexity: Payment flows generate large amounts of data with many individual characteristics, such as time, amount, origin, and destination, as well as a certain history that is more accessible to quantum computers than to conventional IT systems. 
• Plausibility: Quantum computers find it easier to identify cases where essential information is hidden behind the complexity of mass data. In tandem with AI, such situations are easier to see through.

Put simply, AI could be trained to detect known patterns in the data, while quantum computers search for unexpected correlations and thus uncover previously unknown patterns of fraud, for example. Once these patterns are known, AI models can then be trained to reliably detect them and effectively prevent fraud. Quantum computers train AI models.

Of course, such projects are still in the pilot stage. Nevertheless, it pays to start early with initial projects to maintain the upper hand in the arms race with criminals as soon as the first quantum computers become widely available. A certain head start is also important when it comes to finding application scenarios in which a quantum computer promises competitive advantages. Three areas are particularly relevant in this regard:

• Portfolio & risk management: Jiawei Zhou has published a paper in Springer Nature Link in which he explains how quantum algorithms can help optimize an existing portfolio or price derivatives. The two key figures CVaR (conditional value at risk) and VaR (value at risk) play a particularly significant role in this context. This is relevant for both banks and insurance companies because they are regularly required by regulators to quantify risks and cover them with sufficient capital. Quantum computing allows more and more complex scenarios to be mapped, which AI then analyses. 
• Payment processing & liquidity optimization: According to McKinsey, QML can help to improve payment processing and liquidity management in companies. For example, AI takes over the routing of payments and monitors or forecasts payment defaults, while a quantum computer determines the optimal combination of payment methods and correspondent banks. This would also make it easier to assess how payment flows affect a bank's liquidity in its central bank account. Alternative routing methods are also likely to further increase complexity in this area.
• Cryptography & security: On the one hand, quantum computers will undermine existing cryptography and encryption methods. On the other hand, AI can be used to identify and prevent potentially dangerous patterns in data access at an early stage. To this end, QML models analyse which data is considered critical, how it should be encrypted, and which attack vectors exist. Given the risks already being reported today about hacked crypto exchanges, it should be clear that data security will become one of the key decision-making criteria in modern banking in the coming years.

As things stand today, the main priority is to prepare methodically and organizationally for the quantum age. This includes understanding exactly what quantum computing is and what it is not.
The simplest approach for banks and insurance companies is to view quantum computing as a standalone resource, similar to a processor (CPU), graphics card (GPU), or AI model (Google's TPU). Companies therefore need to consider how to gain access to quantum computing as a resource on the one hand, and how to use and integrate this resource within the company on the other. This requires a certain degree of control over the core IT infrastructure because quantum computing cannot be integrated into existing architectures via an API, gateway, plugin, or middleware.

Quantum computing therefore does not replace banking interfaces. It is not a new type of core banking integration, nor does it communicate with payment systems.

All of this makes the upcoming integration into existing systems challenging. Hybrid architectures are likely to emerge that combine classic IT systems, AI systems, and QML on a single platform. This also includes data: both the quality and format of the data must be prepared for QML (preprocessing). At the same time, the data sets must remain auditable and traceable to meet the applicable regulatory requirements.

Because many use cases currently exist only as research projects, the economic benefits remain difficult to grasp, at least for now. Both banks and insurance companies are therefore advised to monitor the market closely and initiate initial projects where they have an impact on their ability to make the right decisions in the future. PPI AG is happy to advise financial institutions on the planning and implementation of these projects.

Combining quantum computing and AI with QML offers numerous opportunities for banks and insurance companies to make their processes more efficient, secure, and smarter. At the same time, the field remains largely untapped, particularly because the necessary infrastructure is not yet available on a large scale.

Financial companies should keep an eye on the research landscape surrounding quantum computing, QML, and post-quantum cryptography to remain capable of acting and to have as many options as possible clarified in advance as soon as the first quantum computers become available.