Twitter has rolled out a new feature to prompt users to think twice before sending an offensive post.
The new prompt detects “potentially harmful or offensive” replies to a tweet before a person hits send, according to a Twitter statement on Wednesday, the same day it rolled out the new feature across iOS and Android accounts, starting with accounts that have enabled English-language settings.
Twitter tested the prompt last year, and found that 34 percent of people revised their initial “potentially offensive replies” or decided not to send them at all, and they were less likely to receive offensive replies back. Furthermore, after one prompt, people composed, on average, 11 percent fewer offensive replies in the future.
Since the testing last year, the social media company added that it has made tweaks for “nuance,” such as sarcasm, including considering the relationship between the author of the text and the replier, based on how often they interact, which may indicate a better understanding of the tone of communication.
Other tweaks included adjustments to the technology to better account for situations where language may be reclaimed by underrepresented communities and used in non-harmful ways, and improvements to more accurately detect strong language, such as profanity, according to the company’s statement.
The company said it also created an easier way for users to let them know if the prompt was helpful.
A report by Amnesty International and Element AI, a global artificial intelligence software product company, on tweets 778 female journalists and politicians from Britain and the United States received throughout 2017, found 1.1 million “problematic” or “abusive” tweets sent to them, nearly one every 30 seconds, on average. Women of color were 34 percent more likely to be mentioned in such tweets, and Black women were 84 percent more likely than White women to be mentioned in abusive or problematic tweets.