Chris Hemsley, the PSR's Managing Director, reflects on the latest PSR breakfast roundtable on regulation and AI.
The PSR held its third industry roundtable on 17 October, this session was to discuss the role of AI in payments and implications for the regulation of the sector.
The discussion benefitted from an introduction from Stefan Hunt from Keystone .
We were joined by payments colleagues from banking, legal, trade and consumer associations. The thematic discussion provided much thought and consideration around the potential opportunities and risks associated with the use of AI in payments.
Fraud detection and prevention is a key use case, but there is less certainty around other use cases which could include improving customer service (especially using Generative AI), improving efficiency of payment processing, greater personalisation and increasing ways to safely authorise payments.
There were several interesting themes and questions that emerged for the discussion; from which I have pulled out a number that struck me as particularly important. This is certainly not an attempt at a complete record of the wide-ranging discussion.
The first thing that struck me was a useful reminder that AI and especially machine learning has been used in payments for many years: in particular, delivering benefits by identifying and preventing fraud. This has been happening in the background in a way that most people are not aware. There are clear benefits from these tools, in terms of protecting people from harm and identifying illegal activity.
This type of AI is quite different to that which has captured the public consciousness through Generative AI, and the likes of ChatGPT.
It is also striking how quickly AI models are increasing in their complexity and sophistication. Behind the scenes, there has been an exponential growth in the power of models. They are now much more able to generate text answers to natural language questions and generate images from text prompts. With this pace of change, Generative AI will start to feature in many more markets – payments is unlikely to be an exception.
Although we have been using AI for some time, there are several reasons to think we may be at an inflection point on their use in payments. As the complexity and capability of AI models increases, within payments we are seeing separate changes that are increasing the amount of data associated with an individual payment (e.g., as ISO20022 is introduced into more payment systems), coupled with increasing commercial and regulatory pressures (e.g., to tackle economic crime).
Against this background, it is important to consider the ‘black box’ nature of many AI models (formally the “emergent property” of AI models). This means that it is likely to become impossible to explain why a particular AI made a particular decision. There was some discussion on the extent to which this matters and how it could affect the way in which regulators oversee these decision-making processes.
This potential for greater use – and the black-box nature of many AI models – raises questions about wider consumer and public acceptance. This is not an issue limited to payments and can be seen in the wider debates about the role of AI in the economy and its impacts on society.
This raises an interesting question about the ethics of data usage (a topic we noted rather than dwelled on in detail) and whether there was a need for a positive case for innovation in this space to support understanding and acceptance by consumers and the public.
Finally, we also discussed the potential impacts of AI on the competitive dynamics in payments. Holders of data could see the importance of their role increase. We noted here that data was held both by individual payment firms and by the central payment system operators (and those providing the technical infrastructure). This raises questions around the benefits of data sharing, pooling data sources and the risks and issues associated with doing this.
Overall, the discussion identified a number of linkages between the wider debate on AI and the issues in payments. It also felt like the start of a series of important discussions, as we all grapple with the implications of the growing – and perhaps accelerating – application of AI within payments and the economy.