Lloyds Bank Head of Data and AI Ethics Paul Dongha is focused on developing AI use cases to generate trustworthy and responsible outcomes for the bank’s customers.
In March, the Edinburgh, U.K.-based bank invested an undisclosed amount into Ocula Technologies, an AI-driven e-commerce company, to help improve customer experience and drive sales.
Meanwhile, the $1.7 trillion bank is also increasing its tech spend to generate revenue while reducing operating costs, according to the bank’s first-half 2023 earnings report published on June 26.
The bank reported operating costs of $5.7 billion, up 6% year over year, partly driven by investments in technology and tech talent, as the bank hired 1,000 people in technology and data roles in the quarter, according to bank’s earning supplements.
Prior to joining Lloyds in 2022, Dongha held technology roles at Credit Suisse and HSBC.
In an interview with Bank Automation News, Dongha discussed the challenges of implementing AI in financial services, how the U.K.’s regulatory approach toward AI could give it an edge over the European Union and what Lloyds has in store for the use of AI. What follows is an edited version of the conversation:
Bank Automation News: What will AI bring to the financial services industry?
Paul Dongha: AI is going to be impactful, but I don’t think it’s going to change the world. One of the reasons it will be impactful, but not absolutely huge, is that AI has limited capabilities. These systems are not capable of explaining how they arrive at results. We have to put in a lot of guardrails to ensure that the behavior is what we want it to be.
There are some use cases where it’s easy to implement the technology. For example, summarizing large corpora of text, searching large corpora of text and surfacing personalized information from large textual documents. We can use this kind of AI to get to results and recommendations, which really could be very beneficial.
There are cases where we can supplement what people do in banks. These technologies enable human resources to do what they already do, but more efficiently, more quickly and sometimes more accurately.
The key thing is that we should always bear in mind that these technologies should augment what employees do. They should be used to help them rather than replace them.
BAN: How will AI use cases expand in financial services once traceability and explainability are improved?
PD: If people can develop techniques that give us confidence in how the system worked and why the system behaved in the way that it did, then we will have far more trust in them. We could have these AI systems having more control, more freedom, and potentially with less human intervention. I must say the way these large language models have developed … they’ve gotten better.
As they’ve gotten bigger, they’ve gotten more complex, and complexity means transparency is harder to achieve. Putting in guardrails on the technology alongside these large language models to make them do the right thing is actually a huge piece of work. And technology companies are working on that and they’re taking steps in the right direction and financial services firms will do the same.
BAN: What is the greatest hurdle for the mass adoption of AI?
PD: One of the biggest obstacles is going to be employees within the firm and people whose jobs are affected by the technology. They’re going to be very vocal. We are always somewhat concerned when a new technology wave hits us.
Secondly, the work that we’re doing demonstrates that AI makes bad decisions and affects people. The government needs to step in and our democratic institutions need to take a stance and I believe they will. Whether they do it quick enough is yet to be seen. And there’s always a tension there between the kind of interference of regulatory powers versus freedom of firms to do exactly what they want.
Financial services are heavily regulated and a lot of firms are very aware of that.
BAN: What edge does the U.K. have over the EU when it comes to AI tech development?
PD: The EU AI Act is going through a process to get put into law; that process is likely to set in in the next 12 to 24 months.
The EU AI Act categorizes AI into four categories, irrespective of industries: prohibited, high-risk, medium-risk and low-risk.
This approach could create innovation hurdles. The U.K. approach is very pro-innovation. Businesses are getting the go-ahead to use the technology, and each industry’s regulators will be responsible for monitoring compliance. That’s going to take time to enact, to enforce, and it’s not clear how various different industry regulators will coordinate to ensure synergy and consistency in approaches.
I think firms will be really glad because they’ll say “OK, my sector regulator knows more about my work than anyone else. So, they understand the nuances of what we do, how we work and how we operate.” I think they will be received quite favorably.
BAN: What do FIs need to keep in mind when implementing AI?
PD: Definitely the impact to their consumers. Are decisions made by AI systems going to discriminate against certain sectors? Are our customers going to think, “Hold on, everything’s being automated here. What exactly is going on? And what’s happening with my data? Are banks able to find things out about me through my spending patterns?”
People’s perception of the intrusion of these technologies, whether or not that intrusion actually happens, is a fear amongst consumers of what it could achieve, and how releasing their data could bring something about that is unexpected. There’s a general nervousness there amongst customers.