Financial services has had a mixed reaction to Gen AI ever since its launch last year. While some banks like JPMC have actively started exploring Gen AI’s use cases in the industry, other traditional banks are moving slowly and cautiously, if at all.
The launch of tools like ChatGPT and Bard has been disruptive. Suddenly everyone from teachers to bankers has to consider what effect these powerful LLMs have on their daily tasks. Since these tools are open to the public, banks have to decide whether they are fit for discretionary use inside offices or not.
And some like KeyBank have decided it is all too new, things are moving too quickly. “We did end up locking down some things across KeyBank. We worked with our legal team and our Chief Information Security Officer, to put together policies around the use case for AI, ML and Gen AI and blended that together into one policy for now. I think that will mature over time. But we communicated to the entire company that we're not locking down and ignoring Generative AI, we're researching it,” said Dominic Cugini, Chief Transformation Officer at KeyBank, in a roundtable discussion last week.
The discussion overviewed how financial services is approaching the technology, and it seems, for now, it's about putting one foot in front of the other.
Is Gen AI a bit of a fad?
At the moment, Gen AI is riding a high that may or may not be reflective of its actual usefulness to financial services. “We often go into the hype cycle, and kind of come out in a little bit of dismay. The difference here is the number of companies and individuals really focused on getting Gen AI, to where I would say, it's marketable,” added Cugini. Banks still have to do a lot of due diligence before the tool becomes usable across functions, but Cugini expects this maturity to come sooner rather than later.
KeyBank chose use cases that could benefit from Gen AI by analyzing whether its implementation would have significant business impact. One of the areas it looked into was SAR (Suspicious Activity Reporting) generation. “We've already automated the process of collecting all the documentation and putting it in front of a caseworker. Now with Generative AI, we want to see we are able to take all that information and put together the case and write up the case itself. So that the case can just be inspected and reviewed,” he added.
Augmentation not replacement
One concern that has garnered a lot of attention is how Gen AI’s capabilities might replace workers. This kind of deskilling has ramifications beyond the job market, it changes who makes the decisions and how, introducing questions of bias and algorithmic injustice. Which is why Cugini insists that it is important that humans remain in the loop. For KeyBank, this means that caseworkers are still reviewing cases and deciding next steps; making Gen AI an augmentation, not a replacement.
This approach is also echoed by the Chief Technology Officer and quantitative economist of General Reinsurance Corporation (Gen Re), Frank Schmid, who was also part of the panel and sees Gen AI as a decision support tool.
“In insurance, in the space of underwriting but also claims, we do not pursue automated decision making. It's only decision support and decision preparation,” Schmid added. His company is also pursuing a use case where workers can be enabled to have conversations with contracts. “Contracts are complex bodies of knowledge. And we see great opportunity here to enable a conversation with this body of knowledge,” he said.
Governing the use of Gen AI
Governance policies for AI function as guardrails, and also act as enablers of implementation by allowing organizations to articulate the what, when, and how the technology is used.
“We do have governance structures in place. So the question is, what is genuinely new about Gen AI, that's not yet covered by our policies and business code of conduct? When we look at it, it's not a lot frankly,” Schmid added.
Basically, the work that the industry has done in establishing guardrails and usage policies around AI over the past 50 years is standing the test of time. Governance structures for Gen AI won't require an overhaul but an update.
What changes to expect
The key takeaway from the discussion was that while Gen AI will change things in the background, Gen AI’s impact will be closer to agile’s effect on software development and farther from how smartphones impacted banking.
Tools and the quality of products: “Gen AI might make some tools that we use today obsolete, or it might strengthen those tools, depending on how companies that own those tools mature them,” said Cugini. He expects the speed and quality of code to be the most impacted due to Gen AI, since technology makes writing code and detecting bugs faster.
Talent: The biggest change might actually be in the talent that these companies hire. If a bank is leveraging Gen AI in its product development workflows, it needs software engineers that have prompt engineering skills which can leverage the tech on top of having sufficient technical knowledge with which they can perform modifications and detect errors.
This, too, is a change cycle that the industry has experienced before. The proliferation of low-code development tools requires software engineers to shift from code-intensive environments to ones that take a more visual approach to development. While the difference might not have been very palpable to the average customer, for those working behind the scenes it means shorter development cycles.
Gen AI isn't a silver bullet for every use case in banking. There are still some use cases (read: there will always be some use cases) that need a different kind of AI model.
“LLMs in general, are in general stochastic (probabilistic) in nature. And many times we actually need rules. There has to be a confluence of rule based decision making which is non stochastic, and generative AI,” said Schmid.
Gen AI’s application builds on the evolutionary path that the industry has been on for some time. Progress no longer follows a linear trend – we are leapfrogging into the future, but that doesn’t mean the past has to pack up its bags and vacate the premises.