During a Joint Economic Committee hearing prior to the congressional recess, Rep. Nicole Malliotakis (R-NY) questioned Sterling Thomas, Chief Scientist at the Government Accountability Office, about utilizing artificial intelligence for government efficiency efforts.
Category
🗞
NewsTranscript
00:00Thank you, Mr. Chairman. I want to thank you all for being here today. I wanted to point out that
00:05in my district, we're experiencing a real problem with treasury checks being stolen.
00:12And obviously, this is not something that's unique to my district, although it's been heavily
00:16impacted. As you've mentioned, the GAO estimates a total of $162 billion in improper payments
00:23in fiscal year 2025 and more than $2.8 trillion since 2003. So you have the issue of
00:29improper payments, and then you have those of the fraudsters and the folks that are actually
00:35physically stealing these checks, washing them, reissuing them, and cashing them. And in my
00:43district alone, we've had roughly 400 constituent cases of stolen federal checks that total $5.4
00:51million. So this is a big issue, right, if we multiply that by all the members and the senators
00:58in this body, you're talking about a substantial amount of money. And I would love to hear from
01:03any of you on what you think could be a viable solution. Obviously, moving people to direct
01:09deposit is one thing. We're looking at the president did issue an executive order. We recently passed
01:14legislation out of the House to require IRS to reissue payments if the consumer wants or the taxpayer
01:22wants via direct deposit. Maybe tracking checks so we know what point these checks are being stolen. And
01:30then AI, as you're discussing today, can come in handy as well. So we'd love to hear from any of you on any
01:35ideas or solutions you may have or what you've seen work potentially in the private sector as it relates to
01:40financial crimes that could be applied here.
01:49Yeah, so I don't have an answer for the direct question of what do you do about washing checks,
01:54congresswoman, but I appreciate the question. You know, within how do you, you know, so we're talking
02:01about really fraud, because fraud is where you get into the point of someone's purposely trying to defraud the
02:07government, which is what this is about. You know, analytics certainly and AI could certainly help
02:13this by tracking the actual payment beyond just the check, making sure that the individual is
02:17actually receiving it, providing some sort of methodology to check back. Those are types of
02:21things that the private sector uses that can then be repurposed potentially for the government.
02:26It is not an area that we have looked into though at GAO.
02:28Okay. Mr. Chilson?
02:30So obviously taking the middleman, I mean, obviously taking the middleman out, which is the
02:37mailbox, I guess. And doing direct deposit makes a ton of sense. It's not really an AI approach.
02:45It is a obviously an established technology, but there you're relying on the commercial sector,
02:52which does have a lot of capabilities to identify, you know, when electronic payments are working and
02:58when they're not. Electronic payments are not fraud free, but the companies have spent a lot of money to
03:04identify how to make sure that money moves properly between accounts. And so I do think that the more
03:09you can move people to that type of approach, the better it will be.
03:12Dr. Thomas, in your testimony, you highlighted the need to extend the Social Security Administration's
03:19authority to share its full death data master file with the Department of Treasury, which is set to
03:24expire next year. And I would like to know your thoughts on how much this has recovered, how
03:31successful it's been. It's my understanding it's about $31 million in payments. And can you highlight any
03:36other multi-agency data sharing that needs to take place? I would love to, but I want to make sure I
03:42give you exact information. So if you don't mind, I'll get back to you on that. Okay, no problem. And
03:46then one more question for you. What is the most significant barrier stopping the federal government
03:50from implementing these advanced AI technologies to eliminate waste, fraud, and abuse? Yeah, so I think
03:56there's two areas that I want to particularly highlight. And, you know, this is one of the biggest
04:02questions out there. And, you know, the first one, I think we've already talked about quite a bit, which
04:06is the, just the partitioning of data, trying to make sure that data is available to the anti-fraud
04:11community so that we can track people across the different parts of the government and track the
04:16information so that it can actually be used for large types of analytic models. The second is actually
04:21just as important and probably needs to happen for the first, which is a foundational data that is
04:26validated. It shows these are actual indicators of fraud or improper payments. We can't get into a
04:33habit of marking something that is not fraud and making a guess at it. We do actually need to go
04:37through the entire process with the IGs to make sure that something has actually been validated and put
04:42that into kind of a gold standard data set that is going to produce the best outcome for our analytical tools.