23 Sep

The new heyday of artificial intelligence: AI in Banking 2019.

Modern learning algorithms have the potential to turn the industry completely upside down.

The various industries, in particular the financial industry, cannot agree on a common tenor as to where the AI journey should lead. There is, however, consensus that modern learning algorithms have the potential to turn the industry completely upside down.

When Alibaba founder Jack Ma and Silicon Valley legend Elon Musk presented their ideas of artificial intelligence at the World Artificial Intelligence Conference in Shanghai last week, their views could not have been more different.


The current trend traces back to the first conceptual approaches in the 1950s, but the developments of the last decade in particular have led to numerous new attempts and breakthroughs in the implementation of artificial intelligence. The main drivers are the exceptional growth of data volumes data volumes as well as groundbreaking developments in IT hardware.

While the first big steps in face and image recognition have put AI back into the focus of public interest, it is now big data clusters such as the ones of Google, etc. which define the market standard.

The financial industry is particularly interested in the AI potential. On the one hand, this is due to the expectations of its customers and regulatory requirements and, on the other hand, even more due to the available data volumes in the banking companies – precisely these volumes turn out to be the most essential advantage compared to new FinTechs.

Nowadays, the concept of AI is a very broad one and constitutes an umbrella term for the topic of intelligent procedures. Machine learning (ML) is already seen as a more narrow term of artificial intelligence and comprises, for instance, clustering or regression procedures. The most elaborate, sophisticated and data-intensive level is referred to as deep learning (DL), which uses complex neural networks. AI is not able to act intuitively, but cognitive performance is enabled by training models.

It is striking that the term robotics is repeatedly used in connection with machine learning, even though it actually rather corresponds to a sequence of instructions and should thus be placed in the nomenclature of AI at best. A chatbot, however, can very well be regarded as an implementation of ML considering corresponding technological aspects.

Nowadays, more than 2.5 trillion bytes of new data are generated on a daily basis. If one were to print out this amount of data, the distance between the earth and the moon could be covered multiple times. The computing power available today makes it possible to process these data volumes. Banks do not primarily see their duty in basic research – instead, the focus is on the effort to integrate intelligent additions into existing and new applications. Required frameworks are obtained from commercial but also from free sources.

By now, there are numerous abstract examples for artificial intelligence, in particular complex procedures such as idiomatic translations as well as autonomous vehicles or sensor-aided learning. In the financial world, intelligent fraud prevention measures, e.g. for preventing credit card fraud, money laundering or the identification of digital identities (using similarities of names for fraud purposes) have prevailed. Text processing as part of document scanning or chatbots is particularly popular in the industry.

Current general AI market trends:

  • Knowledge graphs – connecting different objects from a complex, unstructured data volume in a systematic way. Data is not saved in simple sequences, but in individual, partially connected nodes.
  • Hypermind – summarises sensor-aided AI. Primarily, projections/augmented reality (glasses similar to “Google Glass”) in combination with image recognition support people in their activities.
  • Multimedia opinion mining – is the umbrella term for multimedia segmentation, which clusters media according to personal opinions, preferences or feelings. For instance, negative news can be mitigated or defused in combination with a positive image. The spreading of fake news can be considered a negative example for this type of AI.

Currently, the financial industry still focusses on more tangible approaches such as:

  • The intelligent automation of credit ratings
  • AI processing of text documents
  • Automated payment transactions according to default classification

Apart from this variety of chances, financial institutions are increasingly facing a large number of obstacles. In particular, the results are often not traceable, which primarily poses a problem to the supervisory and data protection authorities. Although autonomous learning processes ideally lead to seemingly correct results, they cannot be consistently checked for their causality and the corresponding decision-making basis.

Some first promising attempts at a solution to this problem are approaches regarding the textual output of important decision nodes, i.e. a kind of logging, which is intended to allow for a certain degree of traceability. This approach, however, is still in the early stages of research. The visualisation of (interim) results is a process which is significantly easier, although it will not always lead to plausible explanations.

The ethical component of artificial intelligence involves another substantial obstacle. A well-known and controversial example of this can be found in the recognition software of autonomous vehicles. Demonstrably, such systems had trouble detecting pedestrians if they did not match the typical white person. Particularly, darker skin tones were detected far less easily. This malfunction can be traced back to the selective choice of test data. The problem of such biased data will also affect the financial industry in one way or another, for instance in AI-supported lending, and the industry will have to deal with it accordingly.


It is a fact that financial institutions have made very little use of their AI potential so far, which is hard to understand given the enormous data volumes at hand. Artificial intelligence is in another heyday which – as recent progress has led to believe – will persist (unlike in the past).

Read more about this topic and the digression to PSD2 and GDPR in connection with artificial intelligence in our SDS Report. Download the report on the subject AI in Banking now.