VA wants to be a leader in using artificial intelligence
March 2, 202012:23 pm
9 min read
Best listening experience is on Chrome, Firefox or Safari. Subscribe to Federal Drive’s daily audio interviews on Apple Podcasts or PodcastOne.
Veterans Affairs last fall launched a new office with the goal of establishing the department as a leader in the use of artificial intelligence. It named Dr. Gil Alterovitz, formerly a health informatics researcher at the Harvard-MIT Health Sciences and Technology Division, as its director. For a progress report, Federal Drive with Tom Temin caught up with Alterovitz at an ACT-IAC artificial intelligence shared interest group meeting. He explained how the National Artificial Intelligence Institute is organized.
Dr. Gil Alterovitz: It is a joint effort between the VA Office of Research and Development, along with Secretary Center for Strategic Partnerships. And so in doing so, it is able to, work across different parts of the organization and has resources allocated to it, including within the central office but also in the research building near Union Station, 1100 1st Street. It’s kind of that new futuristic building if you’ve seen out there. We have a few offices there just labeled artificial intelligence. We’re working on that frosting, but we’ll let you know when that happens.
Tom Temin: Is it the National Artificial Intelligence Institute for the country or for the Veterans Affairs Department.
Dr. Gil Alterovitz: Yes, so the National Artificial Intelligence Institute is within the Department of Veterans Affairs. Its mission is to help veterans in the area of healthcare and well being. And so to do that mission, it’s important to work within the VA. But it’s also important to work across and to learn what other agencies and other parts of government are doing with veterans. So we’ve interacted with, for example, the Department of Defense. The JAIC, the Joint AI Center, to learn about how active serviceman transition to become veterans, which you know is an important time. We interacted with the National Science Foundation on funding approaches toward grants around AI that can benefit the veterans. We feel that the data is very unique that we have. We have the largest integrated healthcare system in the country. We have the largest genomic database linked to healthcare data in the world, and so that is a national resource. A resource that can help our veterans. A resource that other departments certainly we can work together on. So I see it as a starting point that we’ve been able to recognize, leverage this resource to make sure that we’re addressing our mission.
Tom Temin: What are the main activities of the institute? Are you testing algorithms? Are you developing use cases? Are you looking at how data can be used safely or all of the above?
Dr. Gil Alterovitz: Well, there are a number of different aspects that we’re focusing on right now, and there are a number of ongoing activities. So one of the first activities you alluded to is developing the use cases. We’re developing both criteria for the use cases, as well as the organizations and the types of feedback that we’d like to be able to prioritize the use cases that are developed. So that’s kind of one piece,
Tom Temin: I would think that not much in AI can proceed until you have a really solid use case, a requirement.
Dr. Gil Alterovitz: We’re working kind of two use cases. One the type of use case where there’s
an existing use case and adding AI essentially super charges that use case. And then another type of use case where maybe there may not be any Ai there now, but having AI can kind of make it into a unique or different type of use case. So one were adding AI can increase efficiency results. Another one where AI can really change kind of the picture of what’s being done.
Tom Temin: Beyond use cases, we’re going to say there’s some other activities?
Dr. Gil Alterovitz: Yes exactly right. So the use cases are the beginning, as you said, but they’re not the only on ongoing activity. As we’re doing developing the use cases, we’ve already been running things called the AI tech sprints, artificial intelligence tech sprints, which are a ways of engaging industry. They also help to feed into the future for use cases, but usually in these we’ll have industry, academia, nonprofits, different groups working together around a small data set to showcase and build new AI tools that then they’ll demonstrate and show how they can potentially be useful for the veterans. They may then end up using that for non veteran uses, well right, commercial use or other uses. And then at the end, there’s a demo day as part of this. This year we also had them, the ones who wanted to, to engage in a national competition which anyone could apply to from across the country who participate in a process, a particular type of sprint process, which is the opportunity project. So the AI tech sprint include that process, plus added and five other pillars and features. But they basically submitted and the winner of the creating the future of health of all the different teams was actually selected from this AI tech sprint. There was another team also recognized around clinical trial matching, and they’re now working with the Office of Information Technology within the VA around how to integrate that into system. So there’s work that’s ongoing beyond even the text print itself, and then the winner actually got a cash prize that was announced by the executive office of the president’s federal CEO.
Tom Temin: You’re an AI expert. How do you make sure that the work that the institute is doing is new? That is to say that you don’t reinvent work that may already have been done by the thousands of other organizations, like Mitre and so forth, that are also pursuing artificial intelligence?
Dr. Gil Alterovitz: That’s a great point now, especially in this area with artificial intelligence. There’s a number of different players, there are players within VA medical centers or players outside other agencies, there are players within industry, within academia. So that’s why basically, this first phase of evaluating the use cases by getting input from the different players and running Ai tech sprints where you see that different types of tools, whether or not they’ll end up being a tool that is used in a certain case. But you get to see the different technologies by going to the different conferences, leveraging the academic background that I and some of the others have in industry, in the military, and as a veteran on the team. We have a number of veterans on the team as well. That is a way that we can capture that. There’s gonna be this committee which will bring together different parts of the VA, and they will bring the expertise that they’ve had in interacting with different vendors as well in learning about what are the potential options for us.
Tom Temin: You’ve talked in terms of the military domain, Veterans Affairs continuum reaching back to people that are service members. What about the disease cross cutting types of domains such as TBI (traumatic brain injuries), which has to do with professional football and traffic accidents or things like kidney related issues and diabetes, which are really national scourges?
Dr. Gil Alterovitz: Yes. So those are definitely areas that we’ve seen people thinking about AI in those areas and they will be a matter of prioritizing there isn’t seeing which one’s AI can either best supercharge or be able to have a use case that AI shows kind of a completely new wave of helping. But just as an example, you mentioned kidney injury, just this summer an acute kidney injury study was published where using deep learning in collaboration with a commercial entity and leveraging their technology, it was possible to predict acute kidney injury 40 hours ahead of time before the kidney damage was done, and in doing so, you can then take preventative measures to prevent acute kidney injury from happening. So you can imagine a future where using the technology such as this if it continues through could potentially prevent AKI from occurring on making that disease that basically, come like smallpox, isn’t one that you see on a regular basis anymore.
Tom Temin: You’ve only been operating four months or so now. Any early promising results you’ve seen in any of the work being done there at the institute?
Dr. Gil Alterovitz: We’ve seen a lot of work being able to leverage some of the existing work already. For example, being able to kind of put together a couple of different pieces of work. We’ve just begun so we’ve seen a few areas like I mentioned the acute kidney injury, where we’ve tried to help to contribute, to make that successful as much as we can. There’s work that’s been going on on imaging over in a Tampa site. That’s quite exciting to note, and that’s pretty recent/ There’s a number of other imaging areas that are promising as well as natural language processing that we will see in the coming months.
Dr. Gil Alterovitz: So imaging is definitely one of the areas that I think we see quite a bit because there are studies that are coming out comparing imaging analysis by AI to a consensus of actual physicians doing that in many times the AI is able to do the same or better than the consensus. That suggests that while it won’t be AI that makes a decision, it could make a recommendation that then the physicians can see also why it predicted that, it’ll show a region of like, oh this was an area of why I as the AI, for example, was thinking that you should look at this particular lesion and the physician may decide otherwise. But that is the kind of thing that I seem in the near future. It won’t be an automated decision, but more of a recommendation system.
Tom Temin: As you look at AI as an industry, I’m gonna make the analogy of cybersecurity. Every government agency, every industry partner in this whole endeavor says not enough people, not enough talent. What’s the story with AI?
Dr. Gil Alterovitz: Well, I think you could say some of the same things about AI. We have decided on a few challenges, and we kind of update that list of challenges. I was just presented here one of the three challenges, the top one was actually a talent retention, attraction. Those are issues for a number reasons, competing against salary levels of industry and retaining, having a ways in terms of promotion process is that allow people to stay. One of the advantages I think we have is the mission. I think a lot of people are attracted to the mission, and that is something that we have that. And also, as I mentioned that the data. I think literally no place in the world has the type of data and then that deal linked to a very special mission like we do, so that is really helping in the tunnel instruction retention.