Central banks already use AI and CBDC, why not together?

The newest version of ChatGPT makes an unconventional upgrade that explains how AI could boost the adoption of Central Bank Digital Currencies (CBDC).

The relevant change is that ChatGPT 4o hesitates. This isn’t a performance enhancement. Rather it makes the program act more like a human. This was done to appeal to users who are more likely to have confidence in a technology that feels familiar.

This is the same kind of balance that CBDC designers are attempting to achieve. CBDC could make central banking very different from what it looks like today. But such a shift would require fundamental operational revisions. Could AI make this shift less dramatic?

This question is of particular relevance since many central banks are already using AI and experimenting with CBDC. AI could find an additional use in mitigating two economic tradeoffs introduced by CBDC deployment.

1. The rules vs discretion tradeoff

Policy-making frameworks are described as either rules-based (an activity in the economy trips a pre-determined response) or discretion-based (an activity in the economy results in a bespoke human-designed response based on circumstances).

When a CBDC is added to the mix of instruments, its programmability characteristic favors a rules-based approach.

Programmability makes it possible to automate more central bank operations. For example, if the outflows of CBDC from a single bank are happening too quickly, the instrument itself could be programmed to follow the rule that transfers of CBDC will halt above a certain outflow velocity.

The impact of this approach depends on the central bank. The US Fed and Bank of Japan rely more on discretion, while the European Central Bank (ECB) historically follows a more rules-based approach. This has restricted their vision of what a CBDC might do. For example, the ECB has said that it is not interested in programmability, which limits some of the use cases of a CBDC. For example, food stamps and fragmented ownership use cases both create the most value when the token itself is programmed to execute rules.

Applying AI to a bank’s existing decision structure could make CBDC’s rules bias more palatable to discretion-minded central banks. For example, AI could suggest policy responses based on the data provided by the CBDC. This is similar to the role AI already plays for retail users in the economy today. By supporting individuals in their current tasks, AI can make them more comfortable with the potential activities that the technology allows.

2. The fraud prevention vs adoption tradeoff

A second tradeoff that central banks face with CBDC implementation is how to collect data in a way that balances fraud prevention vs consumer adoption. This tradeoff surfaces in the effort to make CBDC cash-like.

While cash transactions can be anonymous, the Fed points out that such a characteristic is impossible in an electronic payment system. This is problematic because behavioral science shows that anonymity increases the appeal of currencies. Unfortunately, this is true for both users and scammers. Central banks are then faced with the problem of how much privacy to apply to CBDC data.

Our interest in this very real problem was sparked by our calculation that CBDC adoption remains below 0.2% in every issuing country. This suggests that consumers are unwilling to transact using CBDC. One among many reasons for this is the suspicion that the issuing authority can see their transaction data at a time when public trust in government is at historically low levels.

CBDC-issuing countries are applying different strategies to balance consumer demand for privacy with the need to analyze transaction data to combat fraud. Jamaica for example allows anonymous transactions below a threshold amount, while others, like the PRC, collect all transaction data regardless of wallet tier. Figure 1 shows some examples of wallet anonymity.

Fig 1. Country-level privacy decisions for CBDC

Source: R3 research, Kansas Fed, IMF

This raises the question of whether AI – which central banks already use for projections and forecasting – can be used to combat money laundering and fraud without reducing adoption even further. BIS Project Neo has already begun exploring this and it appears very likely.

There is little to suggest that consumers who mistrust their government will change their minds if AI is applied. But the amount of activity around using AI for fraud prevention suggests that its use for CBDC analysis is just a short step away. We’re already seeing a shift in the way that central banks approach the privacy conversation, particularly PII. This is part of the global trend to treat personal data as a customer asset, and to which AI can contribute.

“Through working with our extensive global ecosystem, we have achieved a major milestone in both our DFMI initiative and the evolution of securities trading as a whole.” said Antonio Queiroz, Chief Digital Officer at Euroclear. “As a financial market infrastructure with a track record of providing high quality settlement services to the world’s leading financial institutions, we are excited to continue building on our partnerships  to drive digital transformation in the industry and to improve transparency and efficiency across capital markets as a whole.”

“This issuance is not a sandbox or proof of concept project – it is a live market issuance and it demonstrates the role DLT can play in transforming capital markets,” said David E. Rutter, CEO and Founder of R3. “FMIs such as Euroclear play a critical role in the financial ecosystem. We’re working with them and central banks in areas such as securities settlement to improve how we communicate and exchange value, ultimately to deliver a better, more efficient, open, trusted and enduring digital economy. This is about bringing participants and ecosystems together to modernize mission-critical processes in the capital markets industry. Corda is emerging as the technology of choice because it was designed for regulated markets from the ground-up, ensuring resiliency, governance, scalability and security.”

Name Surname
Job title/decription

3. The future of digital central banking

This discussion aimed to provide a glimpse into a future of digital central banking that integrates both of today’s major new technologies. Others have looked more deeply into how AI could be applied in central banking, or how AI might work with crypto. Our goal was to examine both of these elements together to understand how the CBDC design decisions we see in the market could be changed with the introduction of AI as a part of the process. 

The message is that CBDCs and AI don’t need to work together. However, AI’s data analysis capabilities can be used to analyze CBDC data in ways that could change the preferences of both consumers and institutions.