More than 51% of tweets that violate Twitter’s Terms of Service are now automatically flagged by machine learning systems, Twitter CEO Jack Dorsey said Thursday.
The tweets are then handed to human workers for review, in a process that Dorsey said should ease the burden on people who receive harassing messages on the platform, since they won’t have to manually report as many offensive messages.
“We have made some pretty important strides,” Dorsey said Thursday in a video interview with Fast Company editor-in-chief Stephanie Mehta at Procter & Gamble’s Signal conference. “Just two years ago, the only way we would understand if an account was being harassing or abusive was if someone reported it, which is fundamentally not fair because it’s putting all the burden on the victim.”
The company’s goal is to automatically flag 90% of forbidden tweets so that users don’t have to do their own work of weeding out offensive content to feel safe on the platform, he said in a wide-ranging interview that also addressed this week’s high-profile hack of the service.
“For any tweet that we get reports around or that we see, there’s a team working to figure out if it did in fact violate our terms of service and what to do about it,” Dorsey said.
For years, many Twitter users, especially women and people of color, have complained of harassing messages on the platform, ranging from death threats and sexual harassment to doxxing. Dorsey himself is frequently tagged in tweets complaining about perceived flaws in enforcement on the platform, including calls to “do something about the Nazis” using the service.
Twitter’s move toward algorithmic flagging is part of a push to address hate, harassment, and abuse, which Dorsey said “is our key priority as a service and as a company.” Twitter recently rolled out a feature letting users limit who can reply to particular tweets, with options to allow replies only from people mentioned in the tweet or only people followed by the account posting it. That’s part of a shift to give users more control over the platform rather than forcing them to rely on Twitter to enforce policies, Dorsey said. Similarly, last year, the company unveiled a feature to let users hide replies to their own tweets so that they don’t show up in the thread linked to the original tweet.
As social media has become increasingly central to public discourse worldwide, Twitter and its rivals have faced charges of censorship and favoritism from across the political spectrum in trying to set and enforce policies about hate speech, nudity, misinformation, and other controversial content. Recently, the company for the first time took steps to flag tweets by President Trump. Those include a doctored news video featuring a “fake CNN chyron” that was flagged as “manipulated media,” tweets about California’s mail-in ballot policies that the company said could confuse voters, and a tweet about shooting “looting” protesters the company flagged and obscured for glorifying violence. Generally, Twitter allows posts by world leaders that violate its terms to stay up, although it sometimes makes those tweets harder to find and retweet.
“This was the first time we used it on a U.S. president,” Dorsey said of the tweet about shooting protesters, “but we’ve used it around the world on other global leaders as well.”
Company officials are also having ongoing conversations with advertisers and everyday users about other features that might be useful in building trust, aiming to be guided through community feedback, not just top-down decisions, Dorsey said.
“It’s something that has to constantly be evolved and iterated upon based on everything that we learn,” he said.
The Twitter CEO also announced plans to revamp the platform’s Terms of Service to focus on protecting users, not just the company, and making the agreement more easily understandable.
“That’s definitely been a focus of mine in terms of just looking through all the things that people go through to sign up and use our service and what they’re agreeing to—what they read, what they don’t, and why they might be confused when we do take action,” Dorsey said.
Last year, the company rolled out a new Twitter Privacy Center explaining its rules around privacy and its compliance with regulations such as Europe’s General Data Protection Regulation and the California Consumer Privacy Act.
Dorsey didn’t offer a firm timeline for any changes and said the company is likely to take a gradual approach, unveiling rules updates and seeking feedback rather than revamping the legal agreement all at once. Previous terms updates at Twitter and other social media companies have faced pushback from users, especially when they’ve included expansive legalese that seemed to give the platforms sweeping new powers over users.
“We have to do the work and constantly iterate it,” Dorsey said. “I think if we try to do too much at once and just unveil it, it’s ripe for failure.”