„The OSTHAVEN view of the present & future of Artificial Intelligence in Banking & Payment“

The OSTHAVEN view of the present
& future of Artificial Intelligence in Banking & Payment

POSTED ON 16. APRIL 2019 BY TIM DANKER

 

Are we really abolishing ourselves?

 

The topic of artificial intelligence (AI) has been discussed so often in recent days. The German government plans to spend several billion Euros to make Germany fit and to keep up the pace in the dramatically increasing competition of R&D within the field of AI. Companies from Silicon Valley compete with transfer fees for demanded researchers. One might think that AI is about to seize world domination. The myths and stories picture various horror scenarios and outdo each other in drama. Celebrities like Elon Musk and the recently deceased Stephen Hawking warned of nothing less than the downfall of mankind.

High time to take a look at this topic on our OSTHAVEN blog – factually, differentiated and specifically related to our industry – Banking & Payment.

 

Artificial? Intelligence?

Let’s start by taking a close look at what we are actually talking about. Before we deal with the “artificial” part of intelligence, let us first focus on the topic of “intelligence”. To this day, the field of “intelligence” is largely unexplored and lacks a clear definition of what exactly intelligence is. The existing definitions blur between biology, physics and philosophy. Up until today we do not understand exactly how we humans actually function. Expressed in lay terms, one can say that our body is streaked by nerve tracts and muscles, whereby in their control centre, the brain, all threads run together. Neurons flow back and forth in our brain and nerve tracts, switching and acting in a similar way to modern computer processors in our thoughts and actions. From birth, and actually before that, we humans gradually learn all our abilities. During our first weeks of life, for example, our brain learns that what we see in front of our own eyes are our own arms, hands, legs and feet and that we can control them with targeted muscle contractions. So the cute clumsy baby, who unconsciously hits itself in the face with the hand, eventually becomes the cute toddler, who consciously takes his shaky first steps. Behind this process lies a complex interaction of nerves, brain and muscles that we have not yet fully understood. You could say it’s a miracle. But the more we explore the field, the more we see that what looks rather complex at first glance is on a small scale simply the interaction of biological and physical processes. It gets even more complicated when you think about what exactly our soul is, where our free will comes from and in very simple terms what makes us get out of bed every morning and what makes us do everything we do all day long.

The vernacular would declare us humans as intelligent beings. Some certainly more intelligent than others. Humans do things using their mind. Humans are weighing a given situation and decide based on their assessement of the situation. Humans plan ahead and anticipate. Humans act based on experience and what has been learned. Humans try out. But humans also do make mistakes. Why is that? This is a question that we are unable to answer completely based on the current state of knowledge. But we are certain – we are intelligent.

Now the human race has set itself the ambitious goal of artificially recreating “intelligence”, which it neither understands nor is able to define precisely. So let us note that we cannot exactly understand or even define the target we are pursuing.

We now have arrived at the interesting aspects of this dilemma, which makes this topic incredibly attractive, exciting and complex. Research on AI is entering a new field, driving and chasing other fields of research forward and has even helped us humans to better understand ourselves.

Broadly speaking, we have begun to reproduce the researched areas and functions of the human brain with computers, which are ultimately very complex electronic circuits.

 

Man versus machine – two simple examples

For the following thought experiment imagine a photo of any cat. Who will recognize faster whether the animal on the photo is a cat – man or machine? The answer is usually man. For another thought experiment, please briefly consider who will find it easier to calculate the following formula: (2342 * 2345) / 234444 + 23445 * 12499584 – man or machine? In this discipline we would most likely be beaten and use a calculator instead.

A researcher at Stanford University, together with Google, has taught an AI to recognise cats in images. Any cats, in any colors, positions and clippings. For this, AI had to be trained manually in a cumbersome process with over 1 million images. But if you present this artificial intelligence with a picture of a dog, it is at a loss.

What we can learn from these two thought experiments is that computers can work insanely well with structured, formalised content, such as mathematical formulas. There are clearly defined numbers, operators and calculation rules that have to be adhered to in order to reach a result. Computers can do this at speeds unthinkable for the human brain.

We can also see that human brains, together with its various senses, can work impressively well with unstructured and incomplete data sets. As an infant we once learn what a cat is and can recognise cats until the end of our lives; and that goes also for cats in other colours and completely different forms. We humans are able to abstract the format “cat” accordingly and apply it in completely new circumstances. The easiest way to make this clear is that every human who knows what a cat looks like can usually draw a cat. These drawings usually have at least one thing in common – pointed cat ears. Most of the time we also add a cat tail.

We owe it to Stanford University’s impressive research that we are able to train these skills on computers as well. At the same time, this research also highlights the enormous challenges that such a goal entails and, alongside, illustrates the very different strengths and weaknesses that we humans and computers have.

 

Fields of application of AI in banking & payment

In the area of payment, the topic of AI, especially machine learning and pattern recognition, has long been a fixed component. The most prominent and widespread example is risk monitoring at payment service providers and credit card companies. Today, systems based on artificial intelligence generally support the real-time detection of credit card fraud. Companies such as Risk Ident, Fraugster or Feedzai have established themselves in this area, but companies such as Adyen also count on these technologies. The Otto Group company collectAI is currently establishing itself in the field of receivables management, which implements a customer-centric approach in receivables management that is optimised, automated and individualised through AI, thereby reducing effort and costs and at the same time achieving higher success rates in the receivables process.

The use of new technologies in banking is somewhat more differentiated and not widely used yet. First market participants established themselves in the area of Robo Advisor, such as Scalable Capital, which is pursuing a value-at-risk investment approach and uses advanced risk management and simulation algorithms based on AI technologies. Interestingly enough, final investment decisions are still controlled by humans. More and more applications can be found in the field of bank account analysis, which automatically allocate expenses to categories and detect patterns in account movements in order to plan or control budgets and expenses. Examples in this field are the App numbrs, Kontowecker of the savings banks or N26. Programs that automatically trigger account movements, such as savedroid, go even further. In addition, there are applications that provide financial recommendations on the basis of existing knowledge and more and more chatbots and self-service offerings. If you combine all these things and think them a little bit further, a fully automated, individualised and AI-supported  holistic banking advice will be possible in the future.

The banking business seems predestined for the use of these technologies for the simple reason that the data basis of banks is highly standardised and the processes in banks are strongly regulated and formalised. To put it simply, one can assume that processes in banks can be replaced or extended by AI-supported technologies, especially if they follow clear and formal rules.

 

What will become important in the future with AI in Banking & Payment

In particular, with a view to the payment and banking industry, we have identified some aspects that we consider to be important dimensions with regards to the progress of AI as important topics of the future.

These aspects are:

  • Auditing of algorithms and automated decision processes
    When algorithms and AIs make important decisions, they must follow clear rules and be comprehensible. Traceability is difficult to achieve in most AI systems. Therefore, a solution must be found so that audit requirements can be met nevertheless.
  • Regulation for algorithmic decision-making processes
    Not only internal company audit provisions will have to be taken into account, but also requirements from regulators will be given and these must be able to be verified. Rigid regulatory requirements and dynamically growing AI systems seem to be not really compatible. This requires new approaches on both sides.
  • Data quality
    AI-systems are generally based on large amounts of data, which is the only way to achieve good results and efficiency gains. Very simply and accurate one can summarise this with the well-known IT saying: “Garbage in. Garbage out.” If you want high-quality decisions, you need high-quality data.
  • Room for error
    As explained at the beginning of this article, AI-systems learn analogously to humans. People make mistakes and so do AI-systems, mostly due to poor data quality or simply wrong training. In a world with 0% fault tolerance, to which the banking and payment world tends to belong, this means a new challenge. This is particularly important, as we are usually talking about mass transaction business in which real money is moved. Here it is necessary to design solid test frameworks, which are developed especially with regard to the peculiarities of self-learning and self-changing systems.
  • Risks of data monopolies
    As Spiderman had to learn on the cinema screen: “with great power comes great responsibility”. The larger and more significant data monopolies become, the greater the risk of abuse. Politicians, public authorities and society will have to find ways to avoid data monopolies and ensure fair and healthy competition in order to contain the risks of increasing concentrations of power.
  • Regulatory decision-making processes
    How do you give regulatory approvals for self-learning and self-changing systems? Licensing and control processes have to evolve with technical developments and find new ways.

 

A plea for cooperation and the eternal dilemma of AI

The world will change. Computers will take over activities that humans do today. But until further notice, people will not be displaced by computers. Many tasks will change and completely new ones will emerge. For the time being, computers in particular will need humans to teach them the things we want to automate. A computer by itself can do less than a new born child when it comes into this world. Like with the baby, the computer needs a human person who shows him the world, explains it to him and teaches him how to do his tasks.

People should engage with AI and use the respective strengths of humans and computers to maximise overall benefits. We will thus free up precious life time that we will be able to use for new, creative and important things, while boring, recurring tasks and activities will be more and more automated.

The great dilemma of AI is at the same time the most important factor for human beings. Computers are dumb. Computers only work when  a human has given them a clear goal. This is a great chance for humans to not become redundant for the time being. At the same time, there is a great risk in this particular detail. Because computers pursue their goals with all means. Even if it’s the wrong target or an immoral target. Finally, a very simple example to think about: Imagine a fully automated banking advice. The AI takes over all of your banking transactions fully automated, including your investment decisions. The computer behind it can now be given two similar but fundamentally different objectives. On the one hand, it can be given the goal of maximising the investment – that is, getting the best out of it for you in a risk-optimised, cost-optimised and return-optimised manner. On the other hand, it could also be given the objective of maximising the investment while at the same time optimising the bank’s earnings by, for example, exclusively acquiring its own bank’s ETF products, which are perhaps not the most cost-effective for the client but the most profitable for the bank. You think you would notice the difference? And who monitors these algorithms? Your bank?

We are only at the very beginning of this story…