海角大神

Twitter tests new misinformation labels. Will they backfire?

鈥淒isputed,鈥� 鈥淢isleading,鈥� or 鈥淪tay informed?鈥� As Twitter revamps its misinformation labels for better visibility and utility, concerns arise: Will these labels really help people discern facts? And do they allow Twitter to avoid more important content moderation work?

First Look

Last May, as Twitter was testing warning labels for false and misleading tweets, it tried out the word 鈥渄isputed鈥� with a small focus group. It didn鈥檛 go over well.

鈥淧eople were like, well, who鈥檚 disputing it?鈥� said Anita Butler, a San Francisco-based design director at Twitter who has been working on the labels since December 2019. The word 鈥渄isputed,鈥� it turns out, had the opposite effect of what Twitter intended, which was to 鈥渋ncrease clarity and transparency,鈥� she said.

The labels are an update from those Twitter used for election misinformation before and after the 2020 presidential contest. Those labels drew criticism for not doing enough to keep people from spreading obvious falsehoods. Now, Twitter is overhauling them in an attempt to make them more useful and easier to notice, among other things. Beginning Thursday, the company will start testing the redesigns with some American users on the desktop version of its app.

Experts say such labels 鈥� used by Facebook as well 鈥� can be helpful to users. But they can also allow social media platforms to sidestep the more difficult work of content moderation 鈥� that is, deciding whether or not to remove posts, photos, and videos that spread conspiracies and falsehoods.

鈥淚t鈥檚 the best of both worlds鈥� for the companies, said Lisa Fazio, a Vanderbilt University psychology professor who studies how false claims spread online. 鈥淚t鈥檚 seen as doing something about misinformation without making content decisions.鈥�

While there is some evidence that labels can be effective, she added, social media companies don鈥檛 make public enough data for outside researchers to study how well they work. Twitter only labels three types of misinformation: 鈥渕anipulated media鈥� such as videos and audio that have been deceptively altered in ways that could cause real-world harm, election and voting-related misinformation, and false or misleading tweets related to COVID-19.

One thing that鈥檚 clear, though, is that they need to be noticeable in a way that prevents eyes from glossing over them in a phone scroll. It鈥檚 a problem similar to the one faced by designers of cigarette warning labels. Twitter鈥檚 election labels, for instance, were blue, which is also the platform鈥檚 regular color scheme. So they tended to blend in.

The proposed designs added orange and red so they stand out more. While this can help, Twitter says its tests also showed that if a label is too eye-catching, it leads more people to retweet and reply to the original tweet. Not what you want with misinformation.

Then there鈥檚 the wording. When 鈥渄isputed鈥� didn鈥檛 go over well, Twitter went with 鈥渟tay informed.鈥� In the current test, tweets that get this label will get an orange icon and people will still be able to reply or retweet them. Such a label might go on a tweet containing an untruth that could be, but isn鈥檛 necessarily immediately harmful.

More serious misinformation 鈥� for instance, a tweet claiming that vaccines cause autism 鈥� would likely get a stronger label, with the word 鈥渕isleading鈥� and a red exclamation point. It won鈥檛 be possible to reply to, like, or retweet these messages.

鈥淥ne of the things we learned was that words that build trust were important and also words that were not judgmental, non-confrontational, friendly,鈥� Ms. Butler said.

This makes sense from Twitter鈥檚 perspective, Ms. Fazio said. After all, 鈥渁 lot of people don鈥檛 like to see the platforms have a heavy hand,鈥� she added.

As a result, she said, it鈥檚 hard to tell if Twitter鈥檚 main goal is to avoid making people angry and alienating them from Twitter instead of simply helping them understand 鈥渨hat is and isn鈥檛 misinformation.鈥�

This story was reported by The Associated Press.聽