Sandra Matz

Exactly. And you can think of it as essentially this is what we do all the time in our offline relationships. Pretty much any type of conversation that you have is to some extent tailored. You don't talk about the same things or in the same way to a friend or to your kid or to your boss.

So we're trying to replicate this at scale and just say, OK, what is it that you might care about and how can we make saving more appealing to you?

Yeah, so this is essentially research that we did with GPS records. So again, your phone tracks your GPS records pretty much 24 seven. And what we were interested in is whether we could tell whether someone might be suffering from depression or not, just based on these GPS records alone. Now, if you look at the content of some of these traces that we observed, they actually make a lot of sense.

What we found, for example, is that if you don't leave your house as much anymore as you typically did, or there's much less physical activity, you don't travel to as many places as you used to, those are all small indicators that there might be something going on.

It's certainly not a diagnostic tool, but it means that maybe we could be raising a red flag and say, hey, might be nothing, but why don't you check in with some support?

yeah so i don't think it's a deterministic diagnostic tool but it could be incredibly helpful for people for example who know already that they're suffering from depression right so it's like one of these mental health challenges that just pop up time and again and it's really difficult to find your way out of the valley so once you enter the full-fledged depression it's really hard to come back and so if we can get these early indicators of well

maybe it's nothing but here's like a warning system that might alert you to well again there's like these changes in your behavior you're deviating from your typical routine why don't you reach out to someone and see if there's something to it

Yeah. And you can do this in real time. And technically what you could also do if you're really thinking about this as a support system for the person is not just alert the user, but maybe I can give you the opportunity to name two people, loved ones, someone that you want to know that you're having a hard time, even if you're not in a position to tell them.

Yeah, so this is actually one of the projects that I personally care a lot about because there's still so many students dropping out with enormous debt that they never recover. So what we were trying to do is to see if we could predict early on, once people joined university in the first semester, whether we could see if they might be struggling integrating into the system, right?

Maybe they're not finding the information that they should be finding. Maybe they're not embedded in the cohort as much as other people, and they're somewhat on the fringes, not really connecting to the community as much. So we kind of, again, teamed up with a company called Ready Education. They had like a sense of what are the activities that students attending?

Are they talking to other students? Are they part of groups? Are they sending messages? Are they receiving messages? So we looked at all of these data traces. And again, once you combine all of them, you actually have a relatively decent sense of whether someone might be struggling and whether they might drop out at the end of the semester.

Yeah, and for me, what I love most about this is essentially it creates a path to help students. And at the very bare minimum, what it allows administrators to do is identify at risk students, right? So if you see that there's some students who have a higher likelihood of dropping out, maybe you allocate more resources to helping them.

Now, for me, the even more interesting part is that we also get a sense of what is predicting dropout for each individual student. So it could be that I, for example, when I started university as a first-generation student, my problem was that I simply didn't know where all of the information was sitting. I didn't know how to get the literature. I didn't know where to search for information.

And so for me, if that was the prediction that the algorithm had made, administrators could have gone in and said, here's the information that you need. You can pop it up on my app. You can send it in my email. Just make sure that I see what I need to see. Now, there could be other people who know exactly. I know that most of my friends, when I started, knew exactly what they were looking for.

But some of them probably had a harder time integrating with the community and finding the friends and making these connections. So for those people, if we see that that's what's happening based on the algorithm, it's a totally different intervention. So then we're trying to see if we can get you involved in events more. Is there a way to ask other people to connect?

So the moment that you understand why someone is predicted to be a dropout, you can also adjust the approaches that you use to help them.

Exactly. It's the same as targeted advertising, right? So we kind of try and figure out what each person needs at a given point in time. Same for student dropout.

Yeah, so there's really two things that the data has to offer. And I think of it as tracking and treating. So on some level, just all of the data that we generate says a lot about our physical activity, our physical health, but also about our mental health, right? Again, we talked about GPS records that say something about whether you might be suffering from depression.

There's a lot that we can learn about your mental health from what you post on social media, right? So this is the tracking part. But then what I think is really interesting, and it's currently being developed, so I think we're really early stages, is more of the treatment part.

So can I use your footprints to not only surface, let's say, the most relevant interventions to you, the same way that Amazon recommends products and the same way that Netflix recommends movies, can actually an algorithm who knows you based on your data recommend the best treatment for you suffering from depression?