Bringing artificial intelligence to banking supervision
Artificial intelligence – or AI for short – is technology that enables a computer to think or act in a more human way. Big tech companies brought AI to our everyday lives through digital assistants, smartphones, online ads and search engines with features like automatic suggestions and targeted results. But is there a place for AI in banking supervision, a field that relies extensively on the professional experience and expert judgement of supervisors?
The answer is yes. The key prerequisite for successful AI is the availability of data, and supervisors have access to large amounts of data in different forms. These data can be structured, such as quantitative data, or unstructured, such as text in reports and assessments. With AI, supervisors can draw deeper insights from any type of data and make more informed, data-driven decisions. AI can identify patterns that humans fail to spot and thereby enhance the quality of supervision. It can also make supervision more agile by flagging anomalies to supervisors in real time.
One major area of AI is machine learning, which can be supervised or unsupervised. Both forms could be applied in banking supervision. In supervised machine learning, the computer uses data that are pre-labelled by a human to classify information or make predictions. For instance, the computer could predict the level of non-performing loans in the euro area based on economic indicators, bank-specific information and historical data. In unsupervised machine learning, the computer learns from the data without any guidance or pre-labelling by a human. For example, it could learn from the main features of supervisory data to identify liquidity issues in banks by itself.
Another area of AI is natural language processing, which could help supervisors to analyse unstructured data. Computers using natural language processing would be able to understand and assess information provided in text form, such as annual reports and audit reports or capital and liquidity assessments. This tool could be used to extract and summarise information and identify sentiment, i.e. whether the content is positive, negative or neutral.
To explore the potential of AI and other pioneering supervisory technologies (SupTech) in the context of banking supervision, the ECB has established a SupTech Hub. This hub connects internal and external stakeholders, helps supervisors learn more about the latest technologies, and supports colleagues in using advanced analytics and developing new tools. The ECB has already reached out to a number of peer authorities and academic institutions around the world to exchange AI expertise and practices. It has also started to develop new capabilities in the European supervisory community through training and workshops. Moreover, the ECB has launched various AI-related projects, including one on using natural language processing to improve the search function for unstructured information and another to introduce machine learning to certain work-intensive processes. For advanced analytics purposes, the ECB is building network maps based on different databases containing supervisory data.
However, there are two sides to every coin: the more supervision relies on IT solutions, the more exposed it becomes to the risks associated with complex technologies. Early adopters of AI have found that complex computations are sometimes hard to comprehend. Supervisors should therefore approach AI with caution. Regardless of whether or not the algorithms used in machine learning have been constructed in a transparent way, supervisors are still responsible for their work, even if part of it is performed by a computer.
New opportunities offered by technological progress always entail challenges. However, risks and challenges should not deter supervisors from using new supervisory technologies. The ECB will continue, with caution, to explore new technologies and introduce AI into its supervisory processes in order to deliver high-quality supervision.