BoE Paper Studies Impact of Mind-Machine Partnerships on Systemic Risk

While discussing how partnerships between human and artificial intelligence might operate in financial markets to reduce the risks of artificial intelligence applications, the paper focuses on applications in trading, portfolio management, and lending. The paper highlights that machine learning models may cut lending during a crisis, magnifying the downturn, whereas relationship banking protects customers from cutting back lending in response to a crisis and protects banks from runs during a crisis. However, the different strengths of human relationship lending and automated lending suggest that these two approaches should complement each other in a partnership of human and artificial intelligence. Similarly, the fundamental differences between human and artificial intelligence not only help to understand how applications of artificial intelligence affect financial stability risks in the financial market, but also in regulating and supervising the financial sector. A fully automated regulatory process based on machine-readable rules would bear great risks and human involvement is required to interpret legal requirements flexibly and to focus on the overarching goal of a stable financial system. Artificial intelligence tools can help to spot patterns and outliers in regulatory trends but bigger and more timely data does not necessarily make the financial system more stable.

With respect to the supervision in financial sector, the paper notes that artificial intelligence applications can be used for forecasting and systemic risk assessment and can help supervisors to work more efficiently and better discover risks buried in data. The paper, however, notes that trustworthiness is a major challenge for artificial intelligence systems as they do not know what is good and what is not and have no inherent values. This makes them prone to unethical behavior and the fact that artificial intelligence agents collude, or may discriminate, are examples of this. Regardless, several studies have shown that financial distress and failure of banks can be predicted using machine learning methods on the financial information of firms. Though humans cannot be expected to be much better at predicting shocks like COVID-19, their judgment is likely to be superior when responding to these extreme situations. Thus, artificial intelligence is very unlikely to replace regulators, supervisors, or policy makers in the near future, for reasons of robust decision-making in the face of uncertainty; however, these agencies/entities stand to profit from artificial intelligence tools that can assist them in making their decisions. Therefore, in an effective partnership, supervisors, regulators, and policy makers must not be threatened in their autonomy and should always be able to overrule an algorithmic assessment. Artificial intelligence can supplement, but is unlikely, any time soon, to supplant that information set and decision-making capacity. 

Related Links

• Notification
• Working Paper (PDF)

Keywords: Europe, UK, Banking, Artificial Intelligence, Machine Learning, Systemic Risk, Financial Stability, Banking Supervision, Regtech, Fintech, Suptech, Lending, Big Data, BoE