Inside the SEC’s AI approach: Spotting risks, building on machine-learning past

When Dave Bottom needs directions, 99.9% of the time he’ll rely on whatever the navigation system he’s using tells him. 

But the Securities and Exchange Commission’s chief artificial intelligence and chief information officer didn’t always have that level of confidence in the technology — especially given his prior experience leading the National Geospatial-Intelligence Agency’s IT services directorate, where he had a front-row seat to work on state-of-the-art mapping tech.

“When Google Maps first came out, the base coverage just wasn’t there, right? Whether it [said] to go left or go right, you really questioned it,” Bottom said. “Over time, they fixed the base coverage problem. … [Now] you don’t really … question it anymore.”

For Bottom, the parallels between artificial intelligence now and older geospatial issues of that kind are obvious — and instructive as he settles in as the top AI official at the nation’s securities regulator.

“How I’m thinking about the work at the SEC is really informed by the work that I’ve done on the geospatial side,” said Bottom, who joined the agency as its CIO in January 2020 and took on the AI portfolio in March of this year. “Questions that had to be answered [by] … the geospatial community … 10 years ago now are the same types of questions other organizations are facing.”

Dealing with those questions at the NGA — not to mention all sorts of data and IT issues that popped up during his time with the CIA, IBM and the Department of Homeland Security’s Office of Intelligence & Analysis — has positioned Bottom uniquely well to take on the AI-related challenges facing the SEC.

As an independent federal agency, the SEC had some leeway in how it approached various Office of Management and Budget AI guidance and requirements. Still, for Bottom, the choice to operate as if it were a Chief Financial Officers Act agency was simple.

“As I like to say, the SEC is both a regula-tor, but we are also regula-ted, right?” he said. “In the context of federal legislation, OMB guidance, those types of things, we want to make sure that we’re setting the example.”

Part of that rationale comes from Bottom’s personal philosophy, but it’s also a bit of a nod to the SEC’s history with AI. The commission has a long track record of using machine learning, he noted, and as it shifts into more generative AI work, it only makes sense that the agency is “meeting our obligation.”

With that in mind, Bottom in September published the SEC’s AI compliance plan in accordance with OMB’s M-24-10 memo, offering fresh details into how the agency is strengthening AI governance, advancing innovation, developing talent and managing risks. 

Bottom doesn’t personally believe that there’s anything especially noteworthy in that plan, but the document does set the table for the SEC’s next big AI initiative: the release of its use case inventory.

The agency is on track to make its inventory public this month, meeting a federal deadline to do so. While the inventory promises to include some yet-to-be-revealed generative AI use cases, Bottom said the agency has primarily used the technology so far to handle the massive amounts of data it receives.

Take the SEC’s Electronic Data Gathering, Analysis, and Retrieval system, for example. EDGAR is the agency’s searchable database that houses the millions of documents filed to the regulator by publicly traded companies. Thanks to machine-learning technology, SEC staffers are better able to manage these documents, conducting risk and trend analyses and more efficiently assessing disclosures.

Identifying “latent risk issues” via AI is one element of success that the SEC has seen with the technology, as well as “being able to tease out what the right nuggets are in the content we take in,” Bottom said. The agency has experienced “a lot of success in what I would call the ‘data-processing standpoint’ — looking and finding issues that might represent risks to the market in general.”

What AI can potentially do to improve fairness and efficiency in the markets or guard against misconduct is outside Bottom’s purview as CAIO, a position that has him concentrated on internal uses of the technology. The current focus is a handful of generative AI pilots, some of which are targeted specifically at helping the SEC’s army of attorneys.

Those legal staffers are “well schooled” in qualitative analysis, Bottom said, but AI’s content summarization capabilities can provide a major assist. “We’re looking for the key nuggets and patterns that machines should be able to generate pretty well,” he said.

There will also be improvements in queries thanks to the technology; Bottom mentioned a looming shift in search, from getting “a list of links back” now to “probably getting a narrative response back” in the future.

Bottom feels confident in overseeing these AI pilots thanks in large part to the pandemic-era work the SEC did with its cloud environment. Joining the agency two months before the onset of COVID-19, Bottom was forced to use the situation “as an accelerator” to move to managed services as quickly as possible. The SEC’s enterprise data warehouse and analytic capabilities, he added, were part of that rush to the cloud. 

Nearly five years removed from that transition, Bottom said he believes the cloud embrace has left the agency “in a good position to be able to adopt the generative AI capabilities that … the cloud service providers are providing.” 

“We’ve made significant investments and improvements in our ability to access managed services,” Bottom added. “Our ability to access technology that’s being delivered by … providers has gotten much easier. … Gone are the days where we’re trying to bring something [on-premises or] we’re trying to build it ourselves.” 

As the SEC awaits word on what, if anything, from President Joe Biden’s AI executive order survives in a second Trump administration, Bottom and fellow members on the agency’s senior steering committee are moving full steam ahead, providing guidance to internal stakeholders on the technology, identifying policy issues that could be addressed with AI and coordinating across the commission on those issues.

There’s also the business of making sure AI is leveraged against cybersecurity threats facing the SEC, and figuring out the best ways the technology can be used to assist in the regulator’s financial management, Bottom said. All it will take for the agency to realize AI’s potential is the right navigation.

“I think AI can make a difference,” he said, “basically in every facet of the commission.”