Waters Wrap: GenAI and rising tides

Soon after my dad was discharged from the Marines in 1968, he got a job at an insurance company called Equitable Life Assurance Society. (Coincidentally, when I joined WatersTechnology in 2009, our NY office was located on the fourth floor at 120 Broadway—the Equitable Life Building.) He was not a “technologist”, but his best friend, whom he had enlisted with, got him the gig, where he started working with a then-relatively-new technology called mainframes. My dad would spend the next 40-plus years building databases around the globe.

I bring this up because I was on holiday recently down in Raleigh, North Carolina, visiting my family. One night, my pop and I were at a pub and, as fathers and sons so often do, we started talking about AI. I thought it’d be fun to show him ChatGPT, so I grabbed my laptop out of my bag, because even on holiday, I’m always ready to give a tech demonstration. I next asked him what his title was when he retired in 2010. In the ChatGPT prompt, I wrote this very simplistic description and question: Larry M was executive director of global infrastructure at a large, multinational insurance company. Can you tell me what Larry M’s job functions would’ve been?

OpenAI’s chatbot proceeded to quickly spit out 12 job functions with brief descriptions of each. I then went through them, one by one, and asked if the response was accurate to what his actual job entailed. Halfway through, my dad gave this assessment: “Holy fuck! I probably could’ve kept going as a consultant if I could’ve described my job that way—that was exactly the job.”

My dad has been out of the game for more than a decade now, but I feel like many in the industry aren’t quite understanding the tectonic shift that’s underway.

Generative AI is not blockchain, which is to say “not a grand disappointment”; it is a giant evolution in the field of data science that has been decades in the making. The GenAI movement we’re experiencing is not just a leap in machine learning and natural language processing, but it builds on other tried-and-tested technologies like cloud, open-source, and APIs.

This is not to say that GenAI is perfect—it’s still early days, and there are questions about hallucinations, copyright and intellectual property infringements, and the lack of documentation and validation. These are very real and serious concerns, especially for highly regulated entities like banks and asset managers. But firms can’t sit on their hands and wait for the various regulatory bodies to weigh in—the risk of falling behind quickly is too great.

Teamwork makes the dream work

While I was in Raleigh, we published several stories that hit on GenAI.

S&P Global unveiled Spark Assist, its GenAI copilot, which aims to help users reduce the time it takes to complete menial, routine tasks. FactSet built a new portfolio commentary tool, which looks to make the process of writing attribution summaries less time-consuming and challenging. Symphony’s CEO Brad Levy spoke with our Nyela Graham about how an incremental approach to GenAI is necessary. Verafin launched a GenAI copilot for financial crime investigators. And the Intercontinental Exchange’s Chris Edmonds explained how ICE is positioning itself as the leading data provider for firms building their own large language models (LLMs).

This was all over the course of two weeks. And that list doesn’t mention the most interesting story involving GenAI that we wrote over that span.

Last month, my boss, Duncan Wood, who lives in the UK, flew to Charlotte, North Carolina, for an event hosted by our sibling publication Risk.net. While he was there, he interviewed Sathish Muthukrishnan and Jason Schugel. The former is Ally Financial’s chief information, data, and digital officer, while the latter is the bank’s chief risk officer.

Now, you should read this 4,000-word tome, and not just because my boss wrote it, and he’s the greatest writer/editor walking the Earth—the wind beneath my wings, the man I aspire to be (sorry, dad). Seriously, though, it’s a detailed case study as to how Ally has worked with Microsoft to experiment with GenAI internally.

I found this part to be most interesting:

Rather than treating GenAI as a model risk problem, Schugel chose to approach it as a product risk problem—funneling each instance of GenAI through Ally’s existing new product approval process. This had a number of advantages, chief among them the presence at various stages of the review process of business lines and corporate functions, as well as second-line risk managers who collectively cover the 12 forms of risk—and 30 ‘child’ exposures—that Ally identified as its material risks.

The block was removed, no new bureaucratic layers were created—and uses of GenAI were suddenly exposed to scrutiny from a whole range of experts and stakeholders, rather than being the preserve of model risk specialists.

“It was a stroke of genius,” says Muthukrishnan. “The new product committee has constituents thinking about every risk dimension across the company—12 at the top and then multiple sub-levels. It was a very simple idea, but it was also very powerful. Everybody that is part of the committee understands the risks they are assessing. They understand the process through which they assess the risk. They know the questions to ask. And that was what allowed us to accelerate from experiment to execution.”

It wasn’t the only thing. Ally’s risk team also laid down a trio of overarching controls. First, GenAI would be applied to internal-facing use cases only. Second, no information that could identify any individual would be shared with an LLM, and no Ally data would be allowed to feed into a model’s training data. Finally, all use cases would be accompanied by both human intervention as well as additional controls focused on training and oversight.

Again, the story is worth reading because it provides many details, metrics, and thoughts about Ally’s strategy, including a fascinating anecdote about how the AI in its “lab” hallucinated a conversation involving the TV show Breaking Bad. But I think the excerpt above articulates a key point that when experimenting with this new form of information delivery, it’s critical to get a variety of stakeholders involved in the project.

Reading between the lines, I’ll add that it’s not only important that technologists and data professionals are in the room, but that they have an equal voice to that of business leaders. This is clearly happening at Ally, as its head of tech and its head of risk are meeting on a daily basis, even if just for a quick catch-up.

That is not the norm. As I mentioned in the Voice of the CTO series we ran earlier this year, technologists are often told what to do and how to spend budgetary dollars at many large financial institutions.

“If you’re in cost-reduction mode, you want technologists to be in control, right? Tell us how much to cut and have us figure it out. But when you’re in growth mode, technologists aren’t allowed on the business side; they’re told where to direct their attention,” one chief information officer at a global systemically important bank (G-Sib) told me. “So, you go through these natural cycles of cloud, on-prem, distributed, mainframe, and so on. Right now, we’re [being told what to do]—how many years until we’re back in an aligned phase? I don’t know: one year, two years, three years? At least this keeps the consultants happy.”

GenAI offers great promise and great challenges. Teamwork and listening to one another is crucial. It’s not the time for turf wars and quagmires. Those that don’t figure that out will likely be using ChatGPT to update their resumes.

Want to discuss how your firm is approaching genAI or have a contrarian view? Hit me up: anthony.malakian@infopro-digital.com.

The image accompanying this column is “Low Tide, Riverside Yacht Club” by Theodore Robinson, courtesy of The Met’s open-access program.