Artificial Intelligence

A look at the Pentagon buying strategy for artificial intelligence

The Defense Department has been steadily developing artificial intelligence capabilities. But how should it go about purchasing AI tools? To get some ideas,...

Tom Temin@tteminWFED

July 12, 2023 12:32 pm

8 min read

The Defense Department has been steadily developing artificial intelligence capabilities. But how should it go about purchasing AI tools? To get some ideas, the Government Accountability Office recently looked into how a handful of companies handle it. For what they found, Federal Drive with Tom Temin spoke to the director of GAO’s Contracting and National Security Acquisition, Jon Ludwigson.

Interview Transcript:

Jon Ludwigson From our perspective, AI is a broad category of capabilities that provide the opportunity to have a machine do something that humans do. So one of the good examples that I think can give a good picture is the government collects a lot of data, and that data needs to be sifted by a person into items of interest and items, not of interest. Machine can do that conceivably faster than a person, eventually. And so some of the efforts have been focused on trying to deploy AI in that kind of environment. There’s a variety of different ways that AI functions, but essentially it is that notion of putting a machine in place of a human to do things.

Eric White And what are the concerns or what were you all trying to look at when it came to defense entities purchasing AI, software or components?

Join us May 7 at 2 p.m. ET where government and industry leaders will discuss the benefits of digital-centric engineering vs. document-engineering approaches, best strategies and lessons learned, sponsored by Arcfield. | CPE eligible

Jon Ludwigson So GAO looked at AI a few different times over the years, going back to 21, we did a broad look at sort of AI and sort of the application of AI across the government in what we refer to as our AI framework. And then in 22, we took a first look at examining DoD’s acquisition of AI and the things that it was trying to do. Really kind of a lay of the land piece provided an opportunity to sort of describe the kinds of tasks that DoD had set out for itself in terms of developing different types of AI and sort of a picture of how that was going across the department. And then most recently, the report that we released talks about the process of sort of more of the acquisition side of things. So when you think of AI, it can be either organically developed so somebody in-house developed something or you can hire a company to do it or you can buy a product off the shelf. And so what we did in the most recent report was to look at how private companies who also have considerable interest in putting machines in jobs to free humans to do other things. They are in the process of acquiring AI. We’ve reached out to a number of companies to understand how it is that those companies look at applying AI within their businesses and how they acquire it. And then we took that information and compared it to the way the department is doing it and made some suggestions about how we think the department could improve its operations, trying to leverage the things that we outlined in the discussions with private companies.

Eric White That’s interesting. So you were able to analyze the way that these companies are using AI and apply it, even though probably different missions apply the same kind of ideas and techniques to what the Defense Department does. Is it that all-encompassing?

Jon Ludwigson Private companies obviously have different interests than the Department of Defense. So there are banking entities that have interest in this. There are logistics companies that have interest in applying AI. There are ways to have machines do things that humans currently do where machines maybe can do it faster or more efficiently than having a person. Or, as you may know, it’s difficult to get people for some tasks. And so a machine can replace the need to go acquire the machine for the person to use and hire a person to use that machine. You can put AI in to examine data, as I said, is a good example. And so what we tried to do was to take a step back from the particulars of what these companies were doing, but look at the general way in which they are trying to understand how to apply AI within their businesses. So we identified a number of things, kind of five categories from understanding the mission need. And the second category was making a business case for using AI and tailoring the contracting approach and testing and evaluating the proposed solutions and finally planning future efforts to use AI. So really, those five categories, when you take a step back from the particulars, the companies are looking at deploying AI within their businesses. This is the way that they’re looking at that process from a 30,000-foot level.

Eric White All right. And so based on what you heard from those companies, what kind of recommendations did you issue to DoD and the armed services?

Jon Ludwigson We identified that the department has a lot of efforts underway. There are a whole panoply of things that AI offers some opportunity to use within DoD’s mission areas. And what we found was that there was no DoD-wide guidance on acquiring AI, and there was also no service level guidance. And so what we did is we identified that we thought the Department of Defense should step in to establish some initial department wide guidance and that the services should develop their own service specific guidance on how to acquire AI. And and both of those pathways should be informed by the information that we had identified. And the department agreed with both of those categories of recommendations, actually four recommendations, one for DoD and then one for each of the three services.

Eric White And you mentioned the use of AI to analyze data. Did DoD mention obviously what they were what you’re allowed to tell us? Sure. Any other ways that they are utilizing AI and may actually contract out to a company to whether it’s have them create their own AI entity or use their software?

Jon Ludwigson Well, you can. And really, there’s a variety of things that that I can’t I won’t talk about. But yeah, when you when you think about AI and the application of it, you can look at the private sector and some companies have talked about self-driving cars or automated delivery mechanisms. And I think some of those ideas, this idea of autonomy is applicable to DoD. When you think about it in terms of whether it’s conveyance of materials, so delivery truck for Company X, Y, Z that the Department of Defense needs to deliver materials in a variety of situations. Some of those very dangerous where having a human in the vehicle, delivering the material is very risky and dangerous to the person. If you had an AI capability for that, you would be able to keep the soldiers safer and use those soldiers for the other tasks. Right? So, I mean, that’s the basic idea. And I think that when you think about AI, this autonomy function, examining data there just in a considerable number of instances where AI offers the opportunity to improve DoD’s capability. So taking an example of looking at a battlefield, you have lots of things that you need to look at. You have lots of streams of data that give you an opportunity to have a sense of what’s out there, so to speak, and identifying targets and those kinds of things. AI can be a pretty good way to sift through the data to identify objects of interest and data that is not of interest. So it doesn’t contain an object of interest.

Eric White I’m curious on what you — it was mentioned a little bit in the report — of what you all heard from companies when it came to the topic of intellectual property and data rights, just because you know you can, how far back can you base the data that’s being used for an AI tool? What did you hear from the companies when you asked them about that?

Read more: Artificial Intelligence

Jon Ludwigson Data rights is one of the critical concepts that companies really try to think through and from from our perspective, it’s something the DoD really needs to think about as well. Because when you think about tailoring this contracting approach, you need to understand what you’re going to own. So if you’re going to organically develop it, develop it. So if you’re a company and you have a cadre of people who are skilled in developing AI tools and training the AI tool with the data to produce outcomes that are what you’re looking for, that then the company owns the intellectual property. If you are hiring a company to develop things, it gets a little more dicey and you need to make sure that you’re clearly defining who owns what. And in some cases, a company might have some vestigial capability to start with, and they’re going to be hired to use that capability with new data or or make a tweak to make it more suitable for whatever the purpose is. That’s where it gets a little bit more complicated to figure out who’s going to own the intellectual property of the resulting product for DoD. What matters, and really for private companies as well, is when you engage in these contracts, you get a capability. But if that capability is is really owned by somebody else, if the intellectual property is owned by somebody else, then you may be sort of locked in with that vendor. And that vendor then can charge you maybe a higher price than what the open market would would provide if the intellectual property was yours. However, acquiring that intellectual property in the course of the contract can be more expensive. So these are complicated situations that you really need to think through at the front end, because once you’ve locked yourself in, if you haven’t made clear who owns it, then you’re going to end up probably in court figuring it out with the judge.