Kevin Systrom: One Of These Social Platforms May Well Become The Everything App

A conversation with Instagram and Artifact’s founder about social media’s future.

Alex Kantrowitz
19 min readApr 5, 2023

I met up with Instagram founder Kevin Systrom as he was about to take the stage at SXSW in Austin.

Systrom’s been developing a news app called Artifact that uses AI to recommend news articles you might want to read. He’s doing it as Meta takes a similar AI-based approach to filter content on Instagram, transforming his other app from its original use case. Everything seems to be converging into TikTok.

So it was an interesting moment to look back with Systrom at where social media has come from, and where it’s going, especially in this moment of homogenization.

I’d encourage you to listen to the full conversation on Big Technology Podcast. You can find it here:

Spotify: spoti.fi/32aZGZx

Apple: apple.co/3AebxCK

Other apps: https://pod.link/1522960417/

Alex Kantrowitz: Every single social media platform looks a lot like TikTok now. What do you think about this homogeneity? I worry we’ll lose the creative part of the internet when everything starts to use AI to serve you content.

Kevin Systrom: I challenge the premise, which is that everything’s converging, or that that’s a bad thing. When Facebook launched the News Feed, it was a fairly new thing for products, and we take it for granted now. But then services like LinkedIn, Twitter etc. effectively adopt a feed. They wouldn’t exist if the feed weren’t the dominant way to serve information at the time. So it turns out, you can use that to serve people very efficiently.

At that time, the best way to figure out what was relevant to you was to let you choose. You would friend people, and your social network became your filter for relevant content. If I follow you and you post something, I see it. These networks grew to be slightly too large for that to be manageable. If, on Instagram, we just sorted your feed chronologically, the people that posted the most would get the most attention and people would self report seeing too much of one person while missing other people.

Right, Twitter now has the “For You” and the “Following” feed by side for you. In “For You” I get actually interesting tweets. In “Following” I get 1,000 tweets from a single outlet before anything unique.

There’s all sorts of issues, like quality, volume, etc. But my point was, that the feed was a very good way of serving content when the Follow and Friend graph were sparse, when people hadn’t figured out how to get followers and distribution. In general, the system is suboptimal if you go chronological and let people with large followings be the single source of information in your feed. So we started using machine learning to take that content, sift through it, drop anything that’s irrelevant and focus on everything else. Facebook became worth many billions of dollars because of it. But this is also fairly limiting, the only articles that I can consume come from a handful of people who decided to post about them, while there are thousands of high quality articles published every day, it’s very hard to find the good ones. A “For You” feed it is just the next version of a feed, where we go out and find things for you.

One of the beautiful things about TikTok is that literally anyone can become famous. It is the most democratic distribution system because it’s all about the merit of the content, not whether or not that person had built up a huge following beforehand. Every service is asking now, if the content works rather than if someone influential posted about this.

It seems like you’re positioning your new product, Artifact, against Facebook, which is still that old model of what people share.

Facebook and Twitter have figured this out too. When you log into Twitter now you see so much unconnected content. I had some team ran a test on Instagram the other week, I logged in to Instagram and every fourth image is some celebrity I don’t care about, “Do you want to follow this celebrity?” It’s like some partnerships person decided we had to go get a bunch of followers for celebrities. And I think a big part of that is the realization that who you followed initially on a product is not who you are following on a fun story.

What I’m realizing about social media, is that there’s this huge divide between stated and revealed preferences, and Facebook, Twitter, etc, have all realized that to be true. So now as part of your feed experience, you’re seeing that more and more, so everyone’s figuring this out. So rather than positioning against Facebook, I’d say we’re all riding the same wave. It’s just who can capture it first.

Now that all platforms are machine learning-based and are starting to recommend the same stuff to people, what’s the differentiator between them?

Social media platforms, in the United States, tend to be single use. Twitter e.g. is for text, Instagram is for images, TikTok for funny videos, YouTube for videos, you don’t see a lot of convergence. It only kind of works when everybody tries to be someone else. Whereas in Asia, specifically in China, you see a lot of these super apps that do everything, banking, chatting, social media.

There’s a very real possibility that using artificial intelligence, some or, generally, one of these companies will walk away with the everything app. The benefit you get of having a large user base, of having the best artificial intelligence for recommendations is enormous.

There’s a very real possibility that using artificial intelligence, some or, generally, one of these companies will walk away with the everything app.

Instagram is so great, because you get on and literally every celebrity and company exists, you can see what they’re posting, the network effects exist there. There’s a network effect in data as well, if you have enough data, your recommendations are so much better than the next company that they could never catch up.

Instagram itself has the machine learning and the user base of Facebook. But they are still not fully committed to that TikTok model. It seems like they’re going to go there, but if they give TikTok this large headstart, how are they going to compete?

TikTok didn’t come out of nowhere, Musical.ly existed, which was fairly popular with the younger crowd in the United States for a while. I was in Shanghai for a board meeting for a different company and I went into Musically’s office. There were 25 kids around 18 years old, all on computers building this amazing product where people would post short videos with music in the background.

Everyone wrote them off, the product wasn’t working and they had to sell. One of the world’s leaders in AI spotted an opportunity, bought it, took all they’ve learned on all of the other products using machine learning and just layered that on top of Musically. So they created Douyin, the Chinese version of TikTok, and TikTok itself.

Reels on Instagram are often described as “Wanna Be TikTok”, but they are pretty good and getting better, the reason for that is that they have more data, 2.3 million people are using Instagram every day.

When Apple Maps launched, everyone thought it was terrible. But things aren’t great overnight, you have to get there and now it’s pretty good. Android for a long time lagged behind iOS. And now people report being just as happy on both platforms. The differences tend to go away over time, as long as the company that’s trying to attack has enough data or enough users.

When you were at Instagram, did you try to acquire Musically?

I did not.

Mark?

We were excited about them, but there was no overture.

Was that a mistake?

When you judge a decision in the past based on information you didn’t have at the time, it’s easy to come to that conclusion. No-one knew ByteDance would do what they did. But yes, if Facebook or Instagram had acquired Musically at that time, TikTok would not exist. That would be positive for all these other companies that are now competing with TikTok, but nobody could have predicted that?

The United States is much more cautious now in terms of approving mergers and acquisitions. The FTC in particular seems to be interested in blocking more mergers of small companies with bigger companies? Does that give and edge to Chinese companies who might have less of a blocker? You are now a startup entrepreneur now, but you’ve also sold a company to Facebook.

I wish that there were some more clear guidelines stated a priori about what types of transactions would be okay, and what aren’t. What are the principles you can use as an operator, whether you are Meta or Twitter, to know what you can do and what you can’t do? Because it doesn’t seem right that a company just can’t acquire other companies.

Do you think there’s a competitive disadvantage vs. China though?

I’m not sure. If Meta or Instagram had tried to acquire Musically, that probably would have gone through. The Figma Adobe thing is super interesting to watch. When they announced that, everyone acted like it would be fine. And here we are, a couple of months later.

But there’s a balance here, you can’t cut off all acquisitions, because the amount of investment and innovation from venture capital comes because people believe there can be exits. Deciding that the only exits you can have are going public, doesn’t make the system work. That takes off a very important part of the feedback in the system, which can be ruinous. Everyone losing their deposits in banks, can that be more ruinous? Probably. There are a lot of forces and headwinds against US startup culture right now that don’t exist outside and in China. So is there is probably an advantage to being an outside player? But at the same time, I don’t know, if you’re, if you’re ByteDance, and you’re trying to do TikTok in the United States, it’s existential to know if they can operate in one of the world’s largest ad markets with one of the highest average revenues per user?

Pledge your support

Should they be allowed to?

It is completely reasonable to use the same logic, the Chinese use to not letting American apps operate in China without impunity, especially if that app, is one of the foremost ways people consume media and information. I worry very little about someone’s looking at what funny videos I watch.

It’s more the ability to manipulate what videos you watch.

I agree. The question is not, is this actively happening? But is there a door that allows someone to do it if they so choose? Because the context now, everyone’s fine, everything’s going well, but what happens in the future? Is there a world where another world power has a very large app inside of a country that they deem competitive and someone could control what you consume? Throwing an election on Facebook is hard, you have to set up all these fake accounts, form groups, then change the topic of the group, you have to hide. It’s easier if you just control what people consume. Someone should pay attention to that threat.

I don’t think that we should shut down TikTok. TikTok should want to make it work, too, because they’re enormously successful in the United States.

Let’s talk about this democratization thing. I think that average users have a better chance of getting their post seen in the Follow feed than if they’re in an AI feed, where you only have a few videos, or articles coming to the top. What do you think about that, and what actually has a more equal distribution?

I think you’re both right and wrong. Here’s how I agree with you: It is true that generally speaking, you’re going to have the Pareto distribution, 80% of us are going to be on 20% of the content. That is also true on Instagram and Facebook. I would claim that generally speaking, the long tail is dead in social media, it’s very, difficult to get views on unpopular content. The counterfactual without AI is, just as bad, which is what I challenge you on, I don’t think your Twitter experience is the longtail. Very simplified TikTok work like this: They take content without much information, preflight it to a small number of people. They have so many users, that they can collect enough of these lab tests to decide if they want to show that content to more people. In most cases, it’s a dud, but in some cases, that can be good. So they give it to an extra 1000 people. And then they do that again, and again, until they know who this content is going to resonate with. I think that is more democratic than gaming your way to more engagement by either buying followers or just being famous.

The other part is, I don’t like the idea that people with lots of natural distribution, which are followers, can just decide to tweet about the efficacy of masks for example, which becomes the record. There is a technology that is called Bridging algorithms. The idea is that you for example find out who’s a Democrat who’s a Republican. You see what content tends to resonate with both sides, rather than just resonate with one side or the other and you promote that content. That is much easier to do in an algorithmic world than in a Follower feed. I think the world in which you can use algorithms to dictate what gets distributed is much more powerful, and fair, as long as the people in charge of the algorithm have the right intentions and objectives,. So I think it is more democratic, and that’s a good thing.

We like to think that we’re all interested in high minded news. And when I went through the Artifact signup flows, there were stocks and technology and all that, but there was also dating and romance. And I’m like, oh, yeah, I’m definitely clicking that.

When Instagram started, it was a way to take your photos, and make them look a little bit better, because your camera isn’t as great as it could be on your phone. But as it evolved, you know, a lot of it became about bodies. TikTok also started with lip synching, but a lot of it became about bodies. With Artifact, are we also going to end up in the lower end versus the thoughtful story?

It doesn’t have to be either or. I’ve heard that people love adding super heavy documentaries and thoughtful movies that win a lot of awards to their “Watch Later” list on Netflix. But when they log in, they actually watch reruns of shows they already know, and to me, that’s okay. A lot of people often ask, shouldn’t we force people to read more intellectual stuff? When we do that, how much are we getting in the way of what people just want to do?

Some of my friends have signed up for Artifact, and they do get really good tech news and great news about new healthcare startups from it. But they also really liked Daily Mail and the celebrity section, maybe they’re interested in following along with Kanye drama.

One of the things that Artifact does better than anything else right now is that it doesn’t judge. You can both be a great investor and interested in a specific celebrity. People don’t fit the archetypes that we all have in our heads. Trying to work against that does people a disservice in the long run.

There are certain lines that I don’t want to cross as a company. We try very hard not to distribute publishers that have a history of bad behavior or misinformation. We want to draw the line somewhere with quality, you need to be very careful what you put in the machine. We’re seeing some of that with these chat bots, it takes an enormous amount of time and effort to get them to do only safe things. I see that as a fundamental challenge, but I don’t think that we should place a judgement on entertainment, it’s okay to be entertained by a wide spectrum of topics. Artifact does that. If you want funny dating advice alongside your super intellectual Atlantic article on education, go for it. The nice thing is, machine learning doesn’t judge.

But it also can have compounding effects where it starts to cloud out other stuff and starts feeding you just the entertainment?

The first question I get is usually, “Doesn’t it have this problem with filter bubbles?,” which means you only read articles that support your point of view confirmation bias. That puts you in this loop that makes you only see the stuff you want to see. Wouldn’t it be healthy to see other points of view? Not just in politics, also advice on parenting, schooling, exercise or nutrition… I think the answer is yes. Generally speaking, the optimal strategy with any machine learning system is most of the time to do what you think the user is going to want. Then you set aside some smaller portion of time to kind of randomize the behavior, because that random exploration allows you to discover things you wouldn’t have known otherwise, it’s also a very effective way to avoid what we call tunnel vision. If you’re reading a ton on a specific issue in San Francisco politics, it’s important that we explore potentially other publishers or other points of view, not just the one that you’re reading about. That exploration turns out to be all the value in these systems because it’s easy to serve people what they want, it’s a lot harder to discover the things that will be helpful in the margin.

Share

One follow up about Instagram. The biggest complaint that people have had about Facebook or Instagram is that it harms mental health and puts these unrealistic standards of beauty in front of them. Looking back, do you have any regrets or anything you would have done differently?

I spent an enormous amount of time internally working on specifically these issues, trying to figure out what resources we could surface within the app when you were searching for a specific topic. We had all these early on tags around whether it was self harm, or eating disorders, or mental health issues, and people would do searches on them to try to find content, and then people would form these accounts to talk about these types of issues. The question is, what would you do about it? In some cases, we would try to pop up resources, whether it was suicide hotline, or eating disorder information, we would try to take down self harm accounts. But there’s a lot of information that backs up the idea that if you simply take all of it down, people can’t find their communities and help. So there’s always this balance. But we worked a lot on internally, it wasn’t a thing, we thought about once a year, it was an ever present issue. I wouldn’t say we solved it, obviously, because clearly the data supports that social media can be difficult, in particular, for the vulnerable populations.

Maybe the biggest regret I have is that as founders, we just cared about photography. We just wanted people to take great photos, the commercialization of Instagram is not something I got excited about over time. It’s not something I’m particularly proud of the idea that you have these beautiful influencers and people that show an unrealistic version of the world, and that there’s an incentive for them to become this and show this because then they can get famous and get these deals. The influencer culture is something I’ve struggled with, because it wasn’t the vision for Instagram, it was that the every person could be a photographer and show the world through their perspective. But what it ended up being, is that the power of distribution was funneled into a small handful of beautiful, influential, etc, people. Part of why I’m working on what I’m working on now is that, I hope it would be very difficult for that to be true in publishing, with a machine learning algorithm, because imagine a world where we opened up to Self Publishers, I actually love the idea that we can incentivize a lot of people to become creators in publishing, where they can publish on topics that wouldn’t have been economical to publish on in the past. Maybe these niche topics like, a specific type of gardening or Japanese HiFi audio systems or something, can you create a creator ecosystem that unlocks good for the world, rather than just envy? I do think, there’s this race to the bottom on a lot of these social media platforms now of like, who can get creators and who can get the most influential creators and started on YouTube. But there, it feels different because it’s a little bit more substantial. Like I study out on a music production, and they’re these amazing creators, and it feels like I’m going back to school. That feels very different than travel influencers or fashion influencers which dominate the Instagram world. I don’t want to pick on any particular group, it’s more just that the balance of content in the long run leans towards this unsustainable vision of what a perfect person or a perfect life was.

I was with someone last night and they just said, everyone’s posting on Instagram back in New York City, and I now live in the Bay Area, and I feel like they’re just all living their best lives now. How many times do people have to tell you that your Instagram life is not your actual life? People don’t believe it. We always believed that focusing on people and our personal connections, rather than popular influencers or celebrities would be the right direction. They announced like a year ago that they were just going to focus on creators. I understand why the game theory leads them to do that, but I hope, there’s still a firm belief that the people you know, your family and your friends are the core of what makes Instagram Instagram.

You’ve talked about how news is just a beachhead for you, what you’re really excited about is the machine learning technology. And where are you thinking you might go with this?

That news is super exciting to me. If you think about taking a bet, you want to have an edge. I think the news world and social media is oversold, people discount it, because there have been a bunch of people who’ve tried it in the past and haven’t done well, they’ve been around a while, and they’re not TikTok. That was also true with photos, when Instagram came out. I remember countless people telling me there’s no money in photos, you’ll never do anything. Obviously, that didn’t become true. So what’s different this time around? The reason why I call it a beachhead is, I care so deeply about machine learning in the long run, the way that Mark Zuckerberg might care about the internet when he started Facebook back in the day. It’s not that we don’t care about the beachhead, it’s just that this can be so much more in the long run, if you just see that personalization and machine learning is the thing that’s going to turn every industry on its head in the next five years.

When I see written content, I see so much more than news. It’s not just what’s breaking that day, but it’s articles on how to be healthier, or it’s information about a musician that you really love, or a critique on a piece of art that you may not have seen in the past. That, doesn’t feel like news to me. Whenever someone calls us a news app, it’s like calling Instagram a photo app, we ended up being so much more than just pretty photos. If you’re going to start with these companies, I like to remind myself, if Apple started and just said, we’re just gonna sell our computer, they wouldn’t be what they are today, they have to build and design beautiful products and delight people. What we’re doing is saying we are first and foremost a personalization company.

We are going to spend the next 5–10 years focused on news. But wouldn’t it be cool if you could use exactly this technology and go outside? Let’s take shopping. The number of articles that are served on artifact today, that are full of products that people want to buy, fashion, electronics, etc. You can find ways where this could become more than just articles by looking into them articles and saying they actually are recommendations on products. That’s just an example. I’m not sure we’ll go that way, but for for the foreseeable future, I am focused on how people consume news. But I am excited about personalization.

Is this going to be like an enterprise thing or you’ll get the data and you have the engine and you might like license that type of technology out to others?

I wish I could tell you. everyone else. At the beginning of Instagram, we didn’t know where we were going. It’s like you have a map and a compass which points north and you go that direction. But part of the fun of this game is figuring out how you can evolve along the way and not knowing in the moment.

There is another Chinese newsreader app called Toutiao. Have you spend some time researching about it?

We focus a lot on generative AI in the United States. According to my experience, in China, they focus mostly on personalization, and social AI. All the papers that come out, are really focused on how to personalize these enormous user bases, how to make sure people are seeing the right content. I think they’re just ahead in the personalization game. News app should work anywhere, not just China. There are things that make it unique to the United States, particularly the publishers, the ecosystem and the history of social media companies for each of those publishers and how they feel about social media.

The bottom line is, your job as an investor and entrepreneur is looking for patterns and trends and ask, “Could it be different here?” Certainly, there’s a lot of touches history that inspires what we are going after, but it’s a different time. They were five, they started a long time ago, and machine learnings as well as the ecosystem has changed dramatically. It’s not as easy to say, oh, that stamp works there. Just stamp it over here. But it’s certainly exciting and inspiring about it in terms of market opportunity.

I just think ByteDance’s strategy of taking what they’ve built and putting it in different verticals is super exciting, they use text as a beachhead. They’ve purchased other companies and tried to layer machine learning on them, to varying degrees of success. I like to think tech is kind of like making movies. If your first one’s a hit, good on you. Sometimes you have to produce a few more before you get another hit. So the question is, do you have the volume of production that allows you to succeed eventually? We’re just trying to start where we see a large opportunity in the United States.

Thinking of machine learning, we all know about image generation, movie generation and text generation. Where is AI going to develop now? It seems like it came out of nowhere, although researchers were aware this was going on Is there anything else that we should be thinking about? Or is are those three key areas to focus on?

I think those are the key areas now. In general, I so many of the resources are pointed at the China generative AI where probably the most progress is made. It’s cool to say, “three Golden Retrievers making sushi” and get an image of it. But the commercialization of that technology, is a different thing. ChatGPT feels like there’s clearly something there that needs to be more than just a chatbot in a window with a gray background.

AI has that property where we all saw cool demos in the last year. I think what you’re gonna see in the next two years, will defy expectations.

Mark Zuckerberg calls you tomorrow and offers you a billion for Artifact. Are you taking it?

Even if I wanted to, it wouldn’t happen, right? We talked about that earlier.

Lina Kahn is on the phone with Zuck giving you the ahead….

I’m not gonna answer the question because I reject the premise. I don’t think it’s possible. Let’s just talk about something else.

--

--

Alex Kantrowitz

Veteran journalist covering Big Tech and society. Subscribe to my newsletter here: https://bigtechnology.com.