Terry Gerton We have seen recent reporting and we’ve certainly talked to officials at the Department of Veterans Affairs who confirm that the VA is expanding its use of AI and AI adjacent systems across its operations, both on the benefits side and on the healthcare providing side. You’ve expressed some concerns around things that veterans should understand as that moves forward. What are the concerns at the top of your list?

Benjamin Krause I think the biggest concern that I have with the way VA is rolling out AI is the lack of transparency and informed consent from veterans where AI is being used and potentially influencing the quality of the healthcare or quality of benefits adjudication that is happening for them on behalf or being conducted by the Department of Veterans Affairs.

Terry Gerton What would that transparency look like to you?

Benjamin Krause Well, so if you’re a veteran, good example would be a veteran getting a decision back. VA claims that they’re only using AI up until the final decision is being issued. And therefore, because there’s somebody checking the very final decision for adverse decisions, at least, we are doing the right thing and we’re using AI appropriately. As a practitioner who’s seen a lot of decisions that appear to be coming from a computer where the language in the doesn’t make sense, it seems clear that whoever’s doing the checking is not doing a great job in a lot of these instances. And so to your question about, you know, what does transparency look like? I think that every decision where AI touches it, where it influences a particular direction of that decision, whether it’s the final decision or not, needs to get some kind of disclosure. A good example is like how AI is interpreting data that is being taken in, whether claims are getting referrals for disability benefits exams, how the evidence is being weighed, things like that. We really need to know how AI is being used. We also need to know what the algorithms are from the contractors and how they’re weighing evidence in a variety of like, I guess I would call it sub-decisions. So we’re not just worried about the final decision, which is like a thumbs up or thumbs down, either you’re entitled or not. We’re worried about how they’re weighing the evidence, where the information is coming from, whether the claims examiner truly has an entire copy of the file, which sometimes they don’t, how the VA is arriving and what information they’re giving to doctors to review. That’s just a list of a small number of things where AI does have an impact or automation depending on whatever term the Department of Veterans Affairs is more comfortable using on the benefit side. So going back with the notice, I think every decision where AI has a meaningful impact, should have a disclosure at the bottom. So even if the final decision maker was a human, we still should know whether AI played a role in how the data was synthesized, typed into the decision letter itself. I mean, the VAOIG recently found when it came to death benefits that there were, I think, of the 8100 they looked at, 8,000 had some kind of significant error, or at least the error was significant enough, for OIG to notice. So that’s, what, 98% if I’m doing back of the envelope math, or 99%. So there’s some kind of error. So VA’s have always focused on the final decision. Well, we’re not just worried about that. We’re worried about all the subordinate decisions that lead up to the final decision. And we want to know precisely how AI is being used because that can influence whether a subordinate sub decision had an influence on the outcome and a veteran can more easily see that or practitioner can more easy focus on that particular issue. So that’s, you know, one example, full disclosure where AI is used or automation and go on a step back. We kind of want to what VA considers artificial intelligence to be. Versus automation. It’s a term that if you were to look at, for example, VAOIG and the different reports that they’re looking at, where it does appear that they are talking about artificial intelligence when they’re using the term automation anyway. The VAO IG has not yet issued, to my knowledge, a substantive decision. Using the term artificial intelligence as it relates to veteran benefits administration. So to my knowledge, it hasn’t really been expressly addressed. They’re always talking about automation. And if you dig through the records or you dig to the contracts, you eventually get to the bottom of what’s called a privacy impact assessment or something like that. And then you realize like, oh, they’re talking about robotic automation process or something like that, right? And oh, that’s AI. Okay. Well, let’s be forthcoming with that information because people want to know where the machine maybe is having more input than simply just spell check. And I think veterans should be entitled to that information. The Department of Veterans Affairs on the veteran benefits side is required to provide substantive notice to Congress when they’re using artificial intelligence and how. And I don’t think that they’ve been really living up to their end of the bargain when it comes to full disclosure. Healthcare, same type of issue. So one of the things that we’re seeing is not just AI being used to create notes. I mean, everybody who’s used Zoom recently probably appreciates how AI summarizes meetings and how AI maybe even does a transcript. But even for people like myself, and I’m very, you know, relatively familiar with AI at this point, I know that it makes errors and you have to really look hard to make sure that it synthesized the information the right way, depending on the perspective. There’s a lot of nuances to how AI weighs the information. That’s part of what going back to Veterans Benefits Administration, we’re not totally clear about how is, how is Veteran Benefits Administration weighting the data. And we know that that’s the kind of the secret sauce when it comes to ChatGPT, Claude, Grok, some of these other tools that are not open source. We don’t know exactly how they’re weighing the assessment of data. It’s kind of a trade secret, right? Or at least secret enough to where they can come after you if you get that data because it’s private or confidential. And so when it comes to veteran health administration, where you’re dealing with healthcare and decision-making that can lead to life or death, I think veterans also should be aware of the extent to which AI is being relied on for note-taking, for potential healthcare decision-making and getting like true transparency and how these tools are being rolled out, especially with generative AI. And when they put that data into a veteran’s record, a lot of people don’t realize that when it comes to OIG, or rather when it becomes the VHA, and the health information, that information can also affect your benefits on the other side of the house, depending on how it’s written. And so it’s not just healthcare about a cold sore or something like that, or some very serious condition, like a colonoscopy where they detect cancer. Like any of these types of medical records or notes can be used for a formal federal benefits adjudication on the other side of the house. So the VA really has kind of a double duty to make sure that the data they’re putting into the record is accurate and they need to be transparent with exactly how they’re doing it with these different generative AI models.

Terry Gerton Benjamin Krause is the founder of Krause Law. So Mr. Krause, you’ve expressed a number of concerns there in a set of examples about where it could be to the veterans, perhaps, detrimental if AI actually makes a mistake. But let’s take a step back and maybe assume that VA does go the path that you’ve suggested and a veteran’s notification does have disclosure that AI was used and the algorithm is clear. If VA gets it right, what are the benefits to veterans here?

Benjamin Krause Well, so if VA gets it right, it’s going to cut down on human cost in their personnel. So that’s one of the, probably the bigger moves. So if artificial intelligence is suddenly accurate, and 100% real, you know, accurate, and we can 100% trust it, then it’s gonna reduce the number of jobs that the VA has or requires within its existing workforce. I think that’s kind of the first thing. And so what that also means is AFGE, one of the largest union shops involved with the federal government, that they’re going to be probably losing headcount. So that’s kind of a tertiary effect of this, right? So if AI does replace jobs, which is the promise, we’re still not totally sure if that’s going to happen, then it will reduce the headcount, which in theory would reduce the cost of personnel compared to the benefit outcome for veterans. So it may reduce the overall burden on the taxpayer of VA. It may also result in quicker decision-making in claims adjudications. Once it interprets information and that’s accurate, in theory it’s going to reduce the amount of time that a veteran has to wait for benefits. And again in theory, I don’t put that in air quotes here, In theory, it’s going to help potentially also speed up the backlog for claims adjudications at the Board of Veterans Appeals. So that may also be a benefit. In the healthcare arena, it allows for potentially more comprehensive note taking. And I think in that instance, though, we still want to make sure with veterans that they’re granting informed consent to being recorded in this way. One thing that VA very frequently forgets is that veterans are humans and that veterans are Americans and that veterans are patients. And there are certain rights that Americans have and patients have to being recorded by a different party that’s not party of the communication. So like in California, it might be illegal because California has certain recording laws. Maybe states like Illinois, there are quite a few states that have recording laws and so when it comes to healthcare VA really does have a duty to inform the veteran, hey, you’re probably being recorded. This is going to be going into your, your medical file. That also may result in eradicating the patient-doctor confidentiality relationship that does exist because it’s being recorded. So there’s certain effects that have ethics implications that have other implications regarding healthcare that need to be addressed. However, if the information is being conveyed correctly and it’s been recorded correctly and everyone’s thumbs up in agreement with this process, then it would result in hopefully quicker scheduling for appointments, quicker appointments and a loss of reduction in risk of error of physicians and nurses when they’re taking vitals and taking other like history of a veteran, things like that are necessary for a good healthcare appointment.

Terry Gerton It doesn’t seem like just yelling stop is going to slow down the VA’s deployment of artificial intelligence. They’re moving forward on a number of fronts. What would you most want them to do to engage the stakeholder population to build trust in their deployment of AI as they move forward?

Benjamin Krause The biggest way to earn trust is transparency when it comes to the federal government, understanding clearly how these systems and tools are being used using the correct terms so that Congress and the American public understand how it’s affecting them potentially. Also what the potential ramifications may be. I saw in March there was a congressional hearing where people were concerned that government contractors and other actors may be taking this data that’s being gleaned from these databases, as well as potentially AI and used for nefarious purposes. So making sure the Department of Veterans Affairs is taking appropriate steps to prevent extraction of vital veteran data like PII and whatnot, in the marketplace, making sure VA is doing the right thing to protect vets and not being not transparent. And so the lack of transparency, I think, is the biggest key to VA overcoming the lack of trust that’s been built up over really the past few decades.