Another agency takes the plunge into artificial intelligence

The U.S. Government Publishing Office (GPO), once a bastion of ink, paper, heavy printing presses, and light-emitting diode or led type, is now fully in the 21s...

The U.S. Government Publishing Office (GPO), once a bastion of ink, paper, heavy printing presses, and light-emitting diode or led type, is now fully in the 21st century. It is looking at ways it can improve operations with artificial intelligence. GPO Director Hugh Halpern recently testified before the Senate Rules and Administration Committee about AI. The Federal Drive with Tom Temin talked to Director Hugh Halpern about AI and much more about the goings-on at GPO.

Interview Transcript:  

Hugh Halpern Sure. The first thing to keep in mind is. GPO is manufacturing operation. We’re a little bit different than most government agencies. We actually produce finished goods for our customers, and we have customers in all three branches of government. And one of the things that we’re really focused on is delivering those customers their products with the highest possible quality at the lowest possible cost. So if we can adopt tools that help us do that, that really accrues to our customers benefit and ultimately the taxpayer benefit. So let me give you one good example. One of the things that we do is we produce the United States passport. Last year we produced 22 million of them. We’ve got another order from the State Department for another 22 million this year. We have equipment that will examine the identity pages, which is made of a new material that we manufacture here at GPO to make sure that those pages are within all of the high quality and security specifications laid out by our customer. And using a very rudimentary form of AI, a very rudimentary form of computer vision. It can look at those pages, decide which ones are within spec and which ones aren’t, reject the ones that aren’t for our quality team to take another look at. Similarly, we are using tools that have some AI features to scan documents as we’re putting them up on gov info, our trusted digital repository, to make sure that those documents don’t contain personally identifiable information. These services can do this far more efficiently than a human being could, and that’s important information we’ve got to keep private. So if we can have tools that do that, and do it quicker and more accurately than we were getting before. We really want to do that. We view these kinds of technologies as a force multiplier for us here at GPO. And to make sure that we’ve got the ability for our folks to deliver for our customers.

Tom Temin One thing that caught my eye was proofreading of documents before they’re published. And clearly that’s a time consuming thing that takes a person. Have you found that the AI tools that can help with this are more advanced than, say, spellcheck or the grammar check that is built in to something like Microsoft Word, which often misses the context of what you’re doing, and suggests things that actually don’t apply. You wouldn’t want those to happen automatically.

Hugh Halpern That’s a great question and we are really at the nascent stages of looking at this technology for that purpose. Proofreading is really key for us. We produce both the Congressional Record and the Federal Register, and we do that overnight, five days a week or any day the Congress is in session. And having tools that can free up our proofreading team, those really talented human beings, to look at those more difficult questions that really require a human being to look at, if they can clear out some of the underbrush, that’s a really important benefit for us. So I can share with you an example I used at the hearing at GPO convention is we always capitalize the word state, when we’re referring to a political subdivision of the United States. But the word state gets used in a lot of different contexts. So while we have scripts that are really blunt instruments, they’re big search and replace programs. They don’t have the ability to discern when we’re referring to the state of Wisconsin or New York state of mind. So if we can get more sophisticated algorithms, more sophisticated tools that can discern some of those things, that means our proofreaders are freed up from having to clean up after the machines and really are able to focus on bigger issues.

Tom Temin We are speaking with Hugh Halpern. He’s director of the Government Publishing Office. So that sounds like in that particular instance, you would use a large language model type of AI. And I haven’t really heard that in all of the discussions. Applying that to legislation, for example, where you have sometimes enormous documents, thousands of pages with very specific language that, frankly, is not found in too many other human endeavors.

Hugh Halpern That’s absolutely true. And whatever you call this, whether you call AI or it’s a large language model or it’s generative. They all have applications to what we’re doing. One key thing to keep in mind for GPO, unlike Congress or unlike, frankly, a lot of other federal agencies. We don’t generate content ourselves. Our job is to deliver content to the American people and really worldwide on behalf of our customers. So we are less looking at some of these generative technologies, although they could have applications, for instance, in our acquisitions team to help write federal procurement contracts, things like that. And we’re looking at some of those in a pilot. But we’re really focused on some of these technologies, much like you get if you’re composing something in gmail or in outlook, where it’s suggesting a better way to phrase things. So as these tools get more abilities, we want to make sure that we can incorporate those into our workflows to make our folks more productive.

Tom Temin And how do you go about deploying them? Because clearly it takes the information technology staff. But what about the crafts people, the readers, the people that have been doing this work? Do you bring them in to really get a deeper understanding of what’s the best way to apply these tools?

Hugh Halpern Oh, absolutely. The key thing to keep in mind is we want to do this very, very carefully. We don’t want to introduce quality issues into our workflow. And the best people who are going to be able to tell us where we can apply a tool, and get the maximum value out of that tool, are the people who are doing the work. You can have the best I.T. folks in the world, but if they’re not the ones actually doing the work, they might design a tool that doesn’t work in our workflow. So having the folks who are actually doing the work as part of the development process is really key for us. And again, this is not to replace any of these really talented individuals. It’s to, again, act as a force multiplier to allow them to keep up with the increasing volume that we’re getting for federal documents and let them continue doing the work that they’re doing just more productively.

Tom Temin Do you think that I could maybe help with the logistics, because you do have printing operations, even though they’re much more customized and the duplicator is in effect, blown up as opposed to printing presses where you can have small quantities, large quantities. So there’s a lot of scheduling, paper delivery, etc., of variables that go into the daily operation. Can it help maybe speed those up or make things more efficient?

Hugh Halpern Absolutely. And that can help with our production planning. But it can also help at the end of the press. So I was talking about some of the computer vision tools that we currently use. As we invest in new printing equipment, new presses, most of those are going to come with some version of a computer vision tool, the check quality on the back end as well. And with newer technologies, like digital inkjet and similar technologies. It’s much more robust in the ability to find an error, stop, correct that error and pick up than older offset technologies. Although frankly, some of the offset presses we’re looking at incorporate these same kinds of computer vision technologies as well.

Tom Temin Yeah. So compatibility through the supply chain then would be an issue in deploying AI so that it won’t fill up what’s built into the machines for example.

Hugh Halpern No, absolutely. Although most of the major manufacturers are working to incorporate similar kinds of technologies into their products, and some of that we can add to existing equipment, and some will just be introduced into the workplace as we go through our normal cycle of replacing equipment.

Tom Temin And while this is going on, we should comment on the fact that the GPO had a billion retrievals in 2023 through the gov info portal. Tell us about that one, that’s a milestone.

Hugh Halpern Well it is, and actually this week marks is Gov Info’s eighth anniversary. So we’re very proud of that. Gov info is the world’s only ISO certified trusted digital repository. That’s really the highest level of certification you can get for that kind of repository for digital documents. So we have found that folks like getting data digitally, more and more, and we’ve seen that as a steady increase in the volume of documents folks are getting from gov info. We’re also supplying that data to other sites as well. So our partners at the Library of Congress use our gov info data when they’re populating their So this is data, that is really useful to anybody trying to follow the legislative process or figure out how a bill became a law. And we worked really hard to make that available, as do our partners.


Copyright © 2024 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.

Related Stories