How Elon Musk can fight misinformation on Twitter without censorship

Twitter's algorithmic feed makes it a hotbed of misinformation.

(Image by geralt)

Last month, Twitter user Qasim Rashid tweeted the following:

These numbers are not accurate. The average price of West Texas Intermediate crude oil in June 2008 was $134, not $181.58. In March 2022 it was $108, not $99.76. Gas prices were $4.05 in June 2008 and $4.22 in March 2022. So the markup on gasoline has increased modestly since 2008, but not nearly as much as this tweet suggests. (Thanks to Jeremy Horpedahl for the example.)

But people love to be mad at oil companies, so Rashid’s tweet has racked up 18,000 retweets. As I write this on Monday evening, it’s still on Twitter.

Tweets like this are on my mind as I think about Twitter’s Monday announcement that it had accepted a deal for Elon Musk to buy Twitter for $44 billion.

“Free speech is the bedrock of a functioning democracy, and Twitter is the digital town square where matters vital to the future of humanity are debated,” Musk said in the press release announcing the acquisition.

In recent years, Twitter has developed an increasingly elaborate system for removing various types of harmful and low-quality content from Twitter, such as hate speech, vaccine misinformation, and Donald Trump’s tweets tacitly endorsing the January 6 insurrection at the US Capitol.

Rashid’s tweet apparently doesn’t run afoul of any of Twitter’s rules. But garden-variety misinformation like that obviously isn’t helpful to a functioning democracy.

Conversations about this tend to break down along now-familiar partisan lines, with folks on the left demanding that social media platforms do more to fight misinformation and hate speech, and folks on the right decrying that as censorship. Musk has thrown his weight behind the free-speech side of the argument; at this point there’s little chance that Twitter is going to do more content moderation here.

But there are more options than just taking down misinformation or leaving up. A good starting point would be for Twitter to work harder not to actively promote misinformation. That oil tweet wound up with 18,000 retweets because Twitter is designed to maximize the distribution of highly “engaging” tweets. And engaging tweets are often bad tweets.

The trouble with algorithmic news feeds

When I joined Twitter in 2008, the site showed you every tweet by people you followed in strictly chronological order. Then in 2016, Twitter introduced a new algorithmic feed that prioritized tweets Twitter thought users were likely to care about. This change met significant resistance from users, and Twitter initially portrayed it as optional. But over time Twitter has increasingly pushed users to switch. Today, the algorithmic feed is the default view.

It’s easy to think about this as an innocuous improvement to the user experience. If Twitter knows which tweets I’m likely to find most interesting, why not show those first? But the switch had profound consequences for the kind of platform Twitter would become.

In 2015, I had enough Twitter followers that I could count on every tweet getting at least a few reactions from followers. Some tweets got more reactions than others, and I usually hoped that my tweets would “go viral.” But my main motivation was to share stuff I thought was interesting with my direct followers.

But a few years later, I noticed a growing variation in the level of response my tweets got. If I wrote about a highly engaging topic (say, US politics) I would often get a bunch of likes and some retweets. But if I tweeted about a less exciting topic, engagement would be very low. Sometimes, I’d tweet and get no reactions at all.

The first few times this happened, I wondered if I’d written an especially boring tweet. But now I think the more likely explanation is that hardly anyone sees tweets like this. Once Twitter’s algorithm decides a tweet isn’t engaging enough, it stops putting it into people’s newsfeeds.

The practical result is that Twitter’s software is “training” all of us on the kind of tweets to write. Nobody prevents us from writing tweets on non-engaging topics, but when we do it’s like shouting into a void. So over time, we learn to write in a more “engaging” way—which often means writing tweets that are partisan, inflammatory, or that pander to the biases of our existing followers.

And because so much of our public discourse happens on Twitter, I think it has had a non-trivial impact on our political culture. Twitter is feeding people tweets that confirm their existing biases and make them angry or fearful. When we see tweets from the “other side,” it’s often someone saying something outrageous, accompanied by dunks from our own side. We’re less likely to see tweets that challenge our prejudices or introduce us to topics we didn’t know we were interested in.

This basic insight isn’t new, of course. It’s been a common criticism of social media since at least 2010, when Eli Pariser coined the term “filter bubble” to describe it. But the rise of algorithmic feeds over the last decade has made the problem much worse. A common prescription for escaping filter bubbles is to deliberately follow people with ideological views different from your own. But this doesn’t help if Twitter’s algorithm notices you don’t engage very much with their tweets and stops showing them to you.

Twitter should get rid of the algorithmic feed

If Elon Musk does one thing with Twitter, I hope it’s getting rid of the algorithmic feed. Twitter should go back to showing tweets in chronological order, without regard to how “engaging” they are.

This wouldn’t fix all the problems with Twitter. People would still be tempted to pander to fear and anger to try to get retweets. But getting rid of the algorithmic feed would ratchet that temptation down a few notches. People who just wanted to write tweets for their direct followers, without worrying about virality, would be able to do that again. And that will make all the other challenges Musk will face easier.

Back in 2018, Mark Zuckerberg made this chart illustrating the basic dilemma facing social media platforms. The most “engaging” content tends to be low-quality content. Tweets with negative information about oil companies are highly engaging whether or not they’re accurate. Vaccine misinformation is highly engaging. Donald Trump’s tweets were highly engaging. But a lot of engaging content isn’t necessarily good for democracy or public health.

Zuckerberg understood this problem and offered a graphical solution that, in the abstract, is totally reasonable:

The idea is that rather than pushing engaging content that’s “close to the line” out to more people, Facebook (or Twitter) would try to recognize this kind of content and give it less prominent placement in the newsfeed.

This is a good idea in theory. But in practice I think it ran up against Mark Zuckerberg’s desire to make a ton of money. Because engaging content really does lead to people using the platform more, which is good for a tech company’s bottom line. So as long as Zuck is trying to maximize profits for Facebook shareholders, his employees are going to feel pressures to maximize engagement—which in practice will mean boosting a lot of misinformation.

In contrast, Twitter is about to become a privately-held company owned by the world’s richest man. If anybody could afford to leave some money on the table in order to make a social network less harmful to public discourse, it’s Elon Musk.

And dialing back Twitter’s bias toward virality will make it easier for Musk to implement his broader free-speech agenda. Twitter’s current approach actively spreads engaging content, some of which is offensive to a fair number of Twitter users. That’s bad for business even if Musk doesn’t have a philosophical objection to hosting it. Dialing back Twitter’s pro-virality bias would help keep that kind of content on the fringes where it belongs.

And I think it’s crucial for Twitter to get rid of the algorithmic feed for everyone—not just to maintain a chronological feed as an option for those who want it. It doesn’t do much good for any single user to switch to a chronological feed because all the tweets they read will still be warped by the incentives of the algorithmic feed. To get the full benefits of the chronological feed, everyone needs to make the switch at once, so people once again write for their direct human followers, not Twitter’s algorithm.