If you are the cofounders of Instagram, what do you do as your big second act after leaving Facebook? Kevin Systrom and Mike Krieger say they aren’t yet sure, but in the interim, they have been spurred to action by our desperate need for clear information during the novel coronavirus crisis. It’s called rt.live, a website that tracks the velocity of Covid’s spread. (“Rt” refers to how many people a single sick individual will infect over time.) They have used machine learning and data science to calculate the figure, taking into account factors such as how testing might affect the results.
Unlike the raw numbers of states’ caseloads, Rt is a quick way to gauge the direction of the crisis, yielding at a glance the relative impact of tactics such as sheltering to suppress the spread and relaxing restrictions for economic or mental-health purposes. (The letter R derives from “reproduction.”) Systrom and Krieger say that over a million people have accessed the site, including public officials who have used it to plan their strategies.
The Instagram cofounders are still close friends, and both of them now live, as many of us do, with a measure of caution, awaiting a normalcy that they suspect will not come soon. Though they did not express an eagerness to talk about Facebook, their public pronouncements about their former employer have been diplomatic. (A more turbulent account of the relationship might be found in Nick Thompson and Fred Vogelstein’s 2019 WIRED story, Sarah Frier’s book on Instagram, or my own book, Facebook: The Inside Story.) But they were happy to share the origins of rt.live and what they hope they’ve accomplished by the project. The interview was conducted remotely (duh) and is edited for length and clarity. It was recorded before a report that Systrom had been contacted about taking the job as CEO of TikTok—though he did take a question about that Chinese social network.
WIRED: Tell me how rt.live came about.
Kevin Systrom: Before the pandemic hit, Mike and I were spending a bunch of time on machine learning and asking ourselves, where can we apply this for social good? Where can we apply this to build a business?
Mike Krieger: One thing that clearly emerged from our time at Instagram was the advent of wide-scale deployment of machine learning. I had managed very large teams working on it, but it wasn’t like I was working on it myself. So post-departure, one of the top things I wanted to learn about was how we might use it in future endeavors. I actually took a Stanford class on deep learning with Fei-Fei Li. Andrej Karpathy, who is now director of AI at Tesla, was the grad student who taught the class at one point. His class notes were super helpful.
Systrom: When Covid hit, neither of us wanted to take the time to apply machine learning to starting a business but rather asked, what could we do to help people either understand or deal with this pandemic.
Krieger: Kevin took the baton and started to look at published epidemiological models so we could apply machine learning to populate the values that epidemiologists had already decided were important to think about at this juncture.
Systrom: In order to figure out how this thing was going, I needed to do two things. One is to not think about it as an infection for the entire United States but infections that were local. At the time, everyone was thinking what is THE infection, the monolithic infection, doing inside of the United States. I arbitrarily picked the state level.
Why did you focus on this particular metric?
Systrom: The first thing any machine learning expert wants to do is predict the future, right? But you very quickly realize that it’s very, very difficult to predict human behavior. This has an enormous implication for the path of an infection. It’s like saying you can throw a ball, but you have no idea what the wind is going to be, so you can’t predict it. So I realized that it wasn’t worth our time to try to predict the future. We would focus on the right now instead of the next week or the next two weeks. And, it turns out, that is just as useful a question to ask. If cases are increasing in California, does that mean that things are worse or better, or is it just that we’re testing more? We take all of those things together, and synthesize or summarize what’s actually happening right now into this metric called R.
Maybe we should explain what R means.
Systrom: R is simply a value that says how many people get infected based on one person being infected. So, Steven, if you are the only person with Covid and two people get it from you in the entire course of you being sick, then R is two. If R is 1.0 that means exactly as many people who are sick today will in effect be sick in the future, therefore the infection stays around, smoldering. If R is greater than one, the infection climbs exponentially. That’s what you were seeing around late March, when we were trying to get things under control. Mike and I built this model and asked, how can we share this with the world in the most simple and aesthetically pleasing and effective manner? For a couple of weeks we jammed on putting the site together and then launched it. The idea was to give people a single metric that allowed them to know whether it was getting better or worse in their specific state, because that’s what we felt the world lacked at that time.
RT.live is interesting because at times it doesn’t seem in sync with the caseload numbers you see reported other places. A few months ago, for instance, New York was over 4 on your chart and obviously in trouble. Now everyone says it’s one of the safest places. Yet I checked today and the R was over 1.07. That’s bad! Does that mean the numbers are going to go up?
Systrom: Yes, and in fact they are going up. Think of it as a forest fire—what we’re dealing with in California right now. If you have any enormous block of land on fire and it’s growing quickly, that is really bad. If you have a small plot of land on fire that’s growing at the same rate as that big one, that is also really bad, but it’s less bad because you are starting from a smaller base. So four is really, really bad—it squares with what you felt in March, with New York being a place that was reporting at peak somewhere over 10,000 positive tests per day. But then shelter started, and very quickly you saw infections start to drop. So R went below 1.0, which is good. It means the virus is under control. What you are experiencing in New York right now [with a 1.07 R] is that it’s a small fire growing, but not nearly at the rate that it was back in March. Remember. I said 1.0 is the smoldering level.
Krieger: In mid-July we did a version 2, building out state pages to provide two other pieces of information that were key to understanding the picture. We think of these state pages as showing our work. We started adding how many positive cases you are seeing every day, and also added testing volume. As you triangulate those pieces of information you could say, oh, OK, it’s around 1, so it’s neither shrinking nor growing exponentially and the base is small. So net-net I’m less worried than I was in mid-March when R was well over 1, and there were a lot of cases. But we’re not out of the woods yet either.
You also have a tab that’s kind of a time machine that shows what the picture was like at some past dates. Obviously, in March things looked awful, and then went on a course of improvement. But if I hit the little button that says “two months ago,” it’s clear that things now are worse than they were at the beginning of the summer. We’re going backwards!
Systrom: You’re absolutely right—albeit off a smaller base. If we leave it alone, it will grow to the same levels as it was before. A really interesting point is that there are many states where the end of sheltering, which we show, aligns with either the low in the dip or the point when it crosses 1.0 again. Which is to say that stopping the sheltering clearly reaccelerated cases. That is not to say that it was the right or wrong decision—there are economic costs as well. But if you were just considering the rate of growth of the infection, sheltering helped. And removing sheltering didn’t help. And that is clearly shown through those graphs. If you look at Iowa and Minnesota, it’s very clear that the second the shelter stops, everything goes back up again.
It seems obvious that when people aren’t sheltered, the virus is more likely to spread. But it’s still disheartening that we are going in the wrong direction. Do you feel that we’re not learning anything from this? Who is looking at the data and who is acting on it?
Systrom: We are clearly learning what it takes to manage this, meaning what does it take to get R to a value that is manageable such that we don’t overwhelm our health systems. At the same time, there is an economic cost of shutting everything down and keeping people out of work, keeping people out of school. I’m not the right person to ask about what the balance is between the economic cost and the health cost. But I think in some ways we are also normalizing this state. If we jumped back to how we were seeing things in March, I think we’d be horrified about where we are. But since we’ve been in it for a long time, it’s in some way normalized and therefore we normalize the behavior that leads to these curves.
Krieger: The question to ask as we move forward is, if you’re going to have that balance and interplay between sheltering and reopening, how do you track what’s going on so that you don’t overshoot the opening? The combination of R and case count is a really good one to keep your finger on the pulse and understand how that is changing. And then, two, how do you make sure that your response today is better than your response was three months ago.
Systrom: It is really important to point out that there was a time when the federal government said not to wear masks. I remember wearing a mask out to a local shop and getting yelled at for wearing a mask. But that was three months ago. Now you go out and almost everyone is wearing a mask. So we clearly have learned a lot about what does and does not affect transmission and I think we have changed our behavior such that we can be more open economically and still see the declines in cases we’ve seen, say, in California. But don’t forget we are still seeing on the order of like 4,000 to 5,000 new cases per day in California, which sounds small but that adds up pretty quickly. So we are by no means out of the woods.
Are any public officials actually using your site as a gauge to see how far they can go in opening up and letting some more economic activity take place?
Krieger: Yeah. It’s like the early days of Instagram, when a noteworthy celebrity using it would light up our networks of people reaching out. Probably the biggest moment was when Andrew Cuomo began using Rt.live in his daily briefing.
Systrom: I think governors and the federal government are having such a hard time because when things are their best you get a little too confident. And a tiny bit of opening allows the infections to rise again. But it goes the opposite direction, which is if you lock down hard, you might overshoot. It is a very complex problem. We want to be one data point of many for these officials to use.
I heard that you both thought that by now we would have the virus under control and there would be no further need for rt.live. Obviously, that is not the case. Has this experience made you more pessimistic about how long it’s going to take to get back to “normal,” if we ever do?
Krieger: I find reason for optimism in the fact that we have agency here. In some states, the lockdown and the reopening didn’t line up exactly with the R moves. Even before statewide lockdowns, often people were taking their own precautions. Also, even before some states fully reopened, there were people who were not taking precautions and being looser. Covid is not this movie-like thing that we have absolutely no control over. While I don’t think we’re going to go back to any kind of normal any time soon, I am heartened by the fact that California has a really high adherence to masks, for example, and when I go out I can see people are being careful and understanding how they can continue to have economic activity happen while in the midst of this. Still, this is here to stay for a while. The numbers don’t lie. If your number is R or above, it just won’t go away by itself.
Systrom: I’m with you, Mike. I am optimistic because I believe we’ve found an equilibrium between the economic and the human cost—which is a terrible place to be—that we can count on until we get a vaccine. At the same time, I am unhappy that it has taken so long to get to the point where simple things like wearing a mask are now acceptable or even encouraged. Simple things that might have saved thousands of lives. The pessimism is in the disjointed effort, the lack of a cohesive approach. I’m not just saying that it’s the federal government’s role to do this, I’m just saying the fact that we have so many varied opinions on what should be black-and-white science. The fact that it has become politicized. Most countries would look at this moment and say this is a time for unity and coming together and fighting this thing. Instead, it has driven us apart. That is the most disappointing thing, other than obviously the massive human loss.
I’ve got to ask you one question about social media. What do you make of what’s going on with TikTok?
Krieger: I wish I knew what was going on with TikTok.
Do you consider TikTok a powerful, once-in-a generation product?
Systrom: Clearly on the engagement side. I don’t think we in social media have seen a company grow as quickly as TikTok. I don’t think it’s a flash in the pan; I think it’s here to stay. What’s interesting is this pattern, that as much as you think one company is big and grooved in and immovable, the next generation comes and threatens its position. And then you have the next generation and then the next generation. It’s cool to see the speed of innovation happening in social media, so you’ve now got many different options for people to express their creativity and be entertained. TikTok, I think, is the latest example in that chain, and I don’t think that chain will end any time soon.
Maybe your former boss Mark Zuckerberg can use that statement of diversity to bolster his case that Facebook shouldn’t be broken up.
Systrom: That’s not for me to decide, nor is it for me to interpret. What’s important is that people should continue to be able to start a company. Mike and I were two guys with little or no real programming experience in a coworking space, and we were able to develop a next-generation social media platform that scaled to over a billion people. Regardless of whether it’s social media or some other vertical, that should continue to happen. The world needs to right now decide whether or not we’ve set up the context and the environment for that to continue to happen, because that is uniquely American and Silicon Valley, and something that makes the technology economy in the United States so special.
Krieger: I couldn’t agree more.
More From WIRED on Covid-19
- ? Want the latest on tech, science, and more? Sign up for our newsletters!
- In China, GitHub is a free speech zone for Covid information
- How to deal with the anxiety of uncertainty
- Why contact-tracing apps haven’t slowed Covid-19 in the US
- 44 square feet: A school-reopening detective story
- How to help your kids head back to school, in person or remotely
- Read all of our coronavirus coverage here