HAWK:AI on Explainable AI – a panel discussion at the Global RegTech Summit 2022

Our Chief Technology Officer, Wolfgang Berner, shared some key insights during a roundtable discussion on the topic of "What is the current state of AI, ML and Blockchain in RegTech?"

The discussion, which took place during the Global RegTech Summit 2022, saw Wolfgang joined by Nassos Economopoulos, Global Head of Cloud and Emerging Technologies Risk at HSBC, Stacey English, Director of Market Intelligence at Theta Lake, and Sholthana Begum, Head of Innovation and Regulatory Technology at the Bank of England, as well as moderator Georgios Samakovitis, Associate Professor (Enterprise) at the School of Computing & Mathematical Sciences, University of Greenwich. 

Below, we've shared some highlights from the discussion: 

Which benefits does AI contribute to AML aside from obvious ones like cost reduction, efficiency, and similar gains?  

Wolfgang explained that one of the best things about AI is how it can help identify odd behavior without even looking for anything particular in the first place. "To give you an example," he explained," we recently identified some fraudulent activity like fake invoicing at a client of ours by applying anomaly detection paired with entity resolution.  

In Wolfgang's example, anomaly detection worked by identifying a cluster of odd transactions for certain customer types. Substantial amounts of fan-out activity also provided warning signs, which were subsequently underpinned by finding correlations between counterparties. 

"These correlations weren't particularly obvious, but AI could pick up on them," Wolfgang continued. "A combination of minor, easily missed things, like the use of different email addresses containing similar names or overlaps in phone numbers, eventually led to us finding the fraudulent activity without even searching for it. That's what AI is extremely good at." 

What are the main challenges in the implementation and management of AI in the finance world? 

Despite the fact that AI and machine learning have been in use for some time, there remains a belief that they represent a "black box" – that they are dangerous and untrustworthy. "We have the technology to deliver transparency around AI, but we still need to create processes, paired with technology and governance, to actually solve the technology's trust issue. It's about trust from the regulator, the auditor, but most importantly, from the people that are using the systems." 

Explainability is key – particularly around individual decisions. What criteria were used? How did the model arrive at its conclusion? How will the model evolve over time? You need to be able to answer these questions using relatable, everyday language people can understand –especially compliance teams.  

"If you can't explain something, it probably won't be used," Sholthana added. "Not everything needs to be explained on a technical level – you don't need to be a data scientist to use AI – but the output needs to be explainable." 

"It's important to remember that not all AI is the same," Stacey said. "Firms sometimes find themselves using tools that are built using very general-purpose AI, and this means their compliance team is then completely tied up having to check the results. However, with more targeted data sets, it is possible to explain results to compliance teams." 

Even when more complex AI models are used, a lack of transparency should never be tolerated. "Hawk AI is very strict on this," Wolfgang commented. "If we can't explain a feature, we won't use it. There is no room for opaqueness in AI." 

How can regulators, financial services, fintechs, and consumers work together on this?

Bringing the session to a close, the panelists discussed how closer collaboration could foster the implementation and management of AI in the finance world. Although work is underway in this area, more could be done. 

"We can strengthen formats of open communication to encourage an exchange of information between all the parties involved within organizational finance," Wolfgang said. "We should facilitate regular exchanges so individuals in the industry that are concerned - especially regulators – can be reassured that AI is the way to go." 

Events like tech sprints and informal get-togethers where regulators can learn more about the available technology are essential. "A common regulatory framework is also key," Nassos added. "It makes things easier from a cost perspective, especially if you operate across multiple jurisdictions."  

"A good example of the whole industry working together is around data collection," Sholthana said. This is the part of the regulatory journey where we can all apply technology – the bank itself, regulated entities, and third-party providers. In general, regulators support collaboration, competition, and innovation – because, ultimately, the outcomes benefit everyone."