Best listening experience is on Chrome, Firefox or Safari. Subscribe to Federal Drive’s daily audio interviews on Apple Podcasts or PodcastOne.
The Defense Advanced Research Projects Agency (DARPA) has joined the effort to prevent suicides among active duty service members. The numbers are grim: 30,000 in the last 20 years. The DARPA approach seeks to prevent suicide by working what you might call upstream from conscious thoughts. It’s called the Neural Evidence Aggregation Tool or NEAT. Here with...
The Defense Advanced Research Projects Agency (DARPA) has joined the effort to prevent suicides among active duty service members. The numbers are grim: 30,000 in the last 20 years. The DARPA approach seeks to prevent suicide by working what you might call upstream from conscious thoughts. It’s called the Neural Evidence Aggregation Tool or NEAT. Here with details, the program manager in DARPA’s Defense Sciences Group, Dr. Greg Witkop.
Tom Temin: Dr. Witkop, good to have you on.
Insight by Sonatype: Agencies must consider security, user experience, culture and overall integration to create a successful software development process. Find out how three agencies overcame the challenges and moved to DevSecOps.
Greg Witkop: Tom, thank you so much for the opportunity to be here.
Tom Temin: Tell us about this program because typically, the way to find people that might be at danger of suicide is through talk therapy, when they’re answering questions consciously. You’re trying to look at the unconscious and how can that happen?
Greg Witkop: Precisely, Tom, what makes behavioral health screening so challenging, particularly in military populations, where there’s such an enormous commitment to accomplishing the mission, as well as even more fundamentally to not, never letting your unit down, is you seldom ever see or hear people complain, or seek out the help that they need. But if you think about it, this is really not that different than the vast majority of health issues. We don’t know if indigestion or heart attack is a cause of the pain until we get changes in EKG or see cardiac enzymes elevated. We don’t know if a swollen thumb is from just a soft tissue injury or a scaphoid fracture until we get an X-ray. What we’re trying to do with NEAT is to actually create a new way of seeing these invisible wounds of war. My proximal goal is to create that system so that clinicians have access to make earlier diagnosis, prevent more serious damage, and provide objective endpoints of treatment. The intermediate goal is to provide soldiers with that objective evidence, much like an X-ray, so they don’t feel like they’re letting their units down. And finally, frankly Tom, my ultimate goal in this is to provide more evidence to intentionally objectify mental health, to remove that stigma that’s caused by this lingering Cartesian duality, and so people understand there really is no distinction between mental and physical health. And so what we’re trying to do is get at those signals that the brain produces to give us this information so we can get clinicians the tools they need to help our troops before it becomes too late.
Tom Temin: So the presupposition is there is some wavelength, some frequency that you can measure, that is a marker for thoughts that could lead to suicide?
Greg Witkop: If you think about it from time immemorial, the way that we find out what someone, how they’re feeling, what they believe to be true, is we ask them questions. They then filter those answers and provide us with an answer. And then we evaluate those answers to gain that information. I’ve been fascinated since I was in medical school, but the brain’s ability to process information in a complex way, take action on that information, and then, frankly, filter up whatever reaches conscious attention. So there’s this whole domain that occurs within the brain where the brain will analyze process and make decisions without there being conscious awareness.
An easy example I like to use is that of a baseball player. If you’ve ever been on a baseball field, you hear the coaches tell the batters get out of your hair and don’t think about it, just let the bat hit the ball. And the reason for that is from the time of baseball leaves a pitchers hand until it hits that glove is somewhere between 500 and 750 milliseconds, it’s very, very fast. And it’s far faster than the brain is actually able to process. In fact, if you start to think about it, your performance will go down. And so there’s a lot of debates in the literature about what exactly constitutes pre-conscious versus conscious, what exactly the different capabilities of these timeframes are.
What everybody pretty much agrees on is anything that is in a time domain, less than 500 to 750 milliseconds, is a time domain the brain can do processing prior to conscious awareness. So our goal is to just explore this time domain, which we know has critical survival value, and you can do incredible things and give us information prior to that conscious filtering occurring, so that we can get the clinicians the most valuable information, frankly information that has never been available before. So it really is more of the capacity of the brain. We’re just using a different aspect of it than we ever have in the past.
Tom Temin: We’re speaking with Dr. Greg Witkop, he’s program manager in the Defense Sciences group at DARPA. And how do you find those brainwaves? How do you measure that activity? Do you put electrodes on somebody’s head?
Greg Witkop: Great question, Tom. And the answer is we’re going to take a multimodal approach. So yes, one of the things that we are looking at is an EEG-type cap, and also peripheral signals that get into this less than 500 millisecond timeframe. And just to give you an idea of what those signals actually are, so you may be familiar with the EEG and EEG tracings that we’ve had for a lot of different reasons. Within the EEG, there are these small deflections called event-related potentials. And these event-related potentials are part of the brain’s processing of that data that has specific semantic meaning.
And I gotta tell you, I have been fascinated by this, literally, from my time in medical school. There was this canonical experiment by [Benjamin] Libet, where he looked at, he asked people to move their finger, as soon as they made a decision to move their finger and look at the clock that was rotated and see what time that was. And what that discovered – this is way back in 1983 – that the brain was actually generating what’s called a readiness potential, which is another one of these event-related potentials, up to 700 milliseconds before the person was consciously aware they decided to make that decision, to make that movement. So they started 20 years worth of philosophical debates about free will and determinism and how the brain works and all of that, right? But they have fascinated me.
Here again, this brain is doing this capacity. When I was in my ophthalmology residency, I was fascinated by this condition called “blind sight.” And blind sight is this condition where your brain receives visual input, but because there’s lesions in the visual cortex in the occipital lobe that processes that information, the person’s not aware that they’ve received that image. Yet, the lower parts of their brain that are responsible for decisions about movement, walking, all kinds of decisions are still getting it. So you can actually have one of these patients where they could have them walk down a crowded hospital corridor, navigate between wheelchairs and beds, and not hit a single thing. And be convinced the entire time that they were legitimately blind, hence the term blind sight. So again, another example of how the brain these signals, these event-related potentials can come into play.
Most importantly, I want to just emphasize this, if you think about what makes us human, ultimately, it’s our capacity for language. And if you think about the way children learn language, it’s phenomenal. They have no, you think about how you and I learned a second or third language, it’s very difficult as adults, right? But as kids, we can learn language both syntax, the order of speech, and semantics what those things mean, without any conscious awareness or attention. And the key to that is, our brain recognizes things like air incongruence, meaning, with these event-related potentials. And so all I’m doing with NEAT is we are using those same signals that the brain uses when you’re a child learning a new language to recognize when a statement is incongruent, or in error. The classic example of that, frankly, Marta Kutas did this back in 1984. She was an EEG and a slide projector, and would present information that said, “I would like a …” and then a blank “with my coffee.” And then she would flash in the word “sugar.” And there wouldn’t be one of these event-related potentials, because that makes perfect sense to the brain. But then she put in this word, “I would like a ‘sock’ with my coffee.” And the brain just went crazy with these event-related potentials. So that’s what we’re trying to capture with NEAT.
Tom Temin: So the methodology then is someone is equipped to be having an EEG taken of them. And then there’s also an interview process as in the traditional method, but what it is you’re trying to get at is what happens before the answer, basically?
Greg Witkop: Yes and the thing that we’re doing intentionally here, Tom, is we’re going to have this be an automated process, because we’ve learned there’s so many different variables that occur when there’s an interview, dyadic type of situation, personalities, different perceptions, different concerns. And so we envision this being simply a computer screen that would present different statements to an individual. And then we would watch the brain’s responses to those statements, and look for their assessments much like the sock and the sugar example, to see what they believe to be true about those statements. In this case, would be mental health. So I, instead of putting in socks or sugar, you can imagine and this is just an example, but something to the effect of now that I’ve returned from my deployment, I want to then “blank” my life. Maybe say “I want to enjoy my life, or I want to end my life.” That’s what we’re getting at is some type of screening tool that can rely on these signals that have never been available to us.
Tom Temin: And so how do you know what event-related potential is the one to say, aha, this person is at danger of suicide?
Want to stay up to date with the latest federal news and information from all your devices? Download the revamped Federal News Network app
Greg Witkop: Great question again, there is no specific signal, a signal about a particular risk. Each of these signals would enable us to evaluate what they believe to be true about a particular statement, such as the statement that I just used. In mental health screening there are multiple validated questionnaires around say, depression, anxiety, traumatic brain injury, that these statements already exist. And so what we’re doing is we’re using, we could use these validated instruments, derive what the person believes to be true about those specific statements. And that’s what gives us the insight that is not currently available, because we know again that just like with soldiers don’t like to admit that their back is hurting and so they can’t keep up. And so they’re reticent to go get a MRI to see if they’ve got a herniated disc, right? They just want to keep going with the unit. The same thing with mental health questionnaires, they know that if they respond in a certain way, it might influence or prevent them from accomplishing their mission. And so we’re just trying to give that objective metric like an X-ray or an MRI would do in other types of physical conditions.
Tom Temin: Okay, three questions: The first one, how is this differing from the techniques of lie detection?
Greg Witkop: I am so glad you asked that, Tom. Because in fact, if the BAA [broad agency announcement] which is currently out by the way, and is available for everybody, our proposals aren’t due until May 23. And our abstracts are March 29. So anybody that’s out there listening, please go ahead and listen and make proposals. But let me just say specifically on that is that, what you just described there as far as lie detection or credibility assessment, what we write on the BAA and why I’m adamant about [it] is that it’s a fundamental misunderstanding of this problem, and what we’re trying to do in this program. Because if, again if you think about it, if I ask you a question, and then you respond, there is inherent inevitable conscious filtering of your response. We are trying to get ahead of those conscious responses. So anyone that proposes anything along the lines of credibility assessment or lie detection, is automatically irrefutably out of scope. And in fact that on my proposals day, I told them don’t waste your time and don’t waste mine, because that is not what this is about.
Tom Temin: And then the second two questions are kind of related: How will you assemble a sample to test how all of this works? And then what are the ethics if you find that someone in your test group is in fact a suicide risk, can you act on that, from a medical standpoint, aside from the experiment, and get that person to some kind of treatment?
Greg Witkop: I’m gonna answer the second question first because I think that’s probably the the most important question. So each of these performers will have their own institutional review boards, their own process to deal with those exact kind of issues. And again, our goal here, Tom, our ultimate goal is to get to this idea of suicidal ideation. And to answer those questions. We anticipate the datasets will be much more confined, particularly in the proof of concept phase. So you could imagine questions more along the lines of depression, anxiety, or even frankly, if you were to present a statement to me, and this kind of gets to answering your first question of how do we establish this ground truth? If you were to present a statement to me that said, “I can run a mile in one minute,” I would know with absolute certainty that was not true. If you present a statement to me, “I can run a mile in 20 minutes,” I will know with absolute certainty that that’s true, I can do that. Now, if you presented a statement to me that said, “I can run a mile in 6 minutes and 47 seconds,” phew! Let me tell you, that is not my normal pace at 56 years old I don’t know. Now, if I was trying to save my wife or my child, maybe I could get that amount of speed?
But the important point here is this classification would show that I absolutely did not know that it was my, it was indeterminant. And so that’s really what we’re going for, this is try near a classification around specific statements. And so you can see our world we’re developing this proof of concept that you could use a data set and a domain set that would not have the risk factors that you’re so appropriately asked that question about it in terms of suicidal ideation. So again, NEAT is a proof of concept to see if we can do this because it’s never been done, in phase one. And then in phase two, I anticipate we’ll be moving more into that clinical realm, which will have some of those more specific applications that you suggested.
Tom Temin: And finally, is the theory also that if you can identify these potentials that are negative that would show a tendency toward a risk of suicide, can those be altered? That is to say, can you change people’s mental process such that they are no longer at risk at suicide?
Greg Witkop: So Tom, that gets to the very core of what we’re trying to do in two ways. So ultimately, what need is trying to do is determine what someone believes to be true about a specific statement. So while there is no way that we could alter the actual event related potential itself, because again, that’s in this preconscious domain, the absolute goal of this would be that to alter the response in that, if, if someone were to respond to that statement, I want to end my life. And, uh, you and I would hope, and all of our listeners, if we were presented that statement, that our brains will be going crazy right now with incongruence. And that’s an error and no, that is not correct, right? But unfortunately, there are a large number of people that their brain would not show that as an incongruent state. The goal of NEAT is to get people the help that they need, so that instead of wanting to enjoy my life being incongruent, for those folks who consider suicide, where they’re headed, that now after their treatment, it’d be the same as you and I, that if they saw the statement “I want to end my life” that they’d have the same response that you and I would. So that’s precisely what we’re hoping to do, is to be a diagnostic indicator at the early stages, as well as endpoint treatment analysis. In the end, of again, what someone believes to be true about those specific statements. [Does that] answer the question?
Tom Temin: Yes. And at the same time, you will have an objective piece of knowledge that this is a medical condition, and not something because the person is weak or some other attribute that is often wrongly associated with a tendency toward taking one’s life.
Greg Witkop: Precisely right, Tom. Think about this, think about if we treated heart attacks the same way we treat this. So all I tell you, my chest hurts. And we would just expect, oh, well, can you tell is that indigestion? Is that a heart attack? We don’t have any test that we’re going to use, and we just want you to make that decision whether or not this is severe enough, you might actually cause your death. It’s ludicrous! And I’m saying the same thing with mental physical processes is we’re not going to rely on or ask someone to make a diagnosis that they have no way of making. We’re going to provide science and these tools to give them that objective evidence that yes, this is an electrical signal, just like an EKG is an electrical signal. And then you are not weak because of this. This is your body’s response, that is a warning signal that we need to get you help. And that’s precisely what I want to do, Tom. I want to eliminate the stigma associated with mental health once and for all. And I’m hoping this is a step in the right direction.
Tom Temin: And briefly, how does this all get underway? How do you start applying this technological approach in this technique?,
Greg Witkop: Well we are, as again, we right now the BAA is out on the street. We are soliciting performers, we are hoping to form these tightly knit multidisciplinary teams that will have the capability to work in the psycholinguistic realm, as well as these event-related potentials, as well as some very sophisticated machine learning analytic tools to make sense of this information and put these teams together and see what they can come up with. And that’s the beauty of DARPA, right, we do things that hadn’t been possible before. And so that’s the stage we’re at right now is we’re looking for putting really together communities of researchers that haven’t formed before. And so that’s why I so appreciate the opportunity to speak with you today.
Tom Temin: Dr. Greg Witkop is program manager in the Defense Sciences Group at DARPA. Thanks so much for joining me.
Greg Witkop: Thank you, Tom. Have a wonderful day.
Anyone who is in crisis — or looking to help someone else — can call the National Suicide Prevention Lifeline 24/7 at 800-273-8255 (TALK).
Free trained crisis counselors also are available via text 24/7. Text the word HELLO to 741-741.
Starting July 16, everyone across the United States will be able to connect to the National Suicide Prevention Lifeline by dialing 988.