Elon Musk is slamming the brakes on his deal to acquire Twitter due to what he considers a huge issue on the platform: the actual number of spam accounts. But detecting bots is a lot harder than it looks.
Botometer, an online tool that monitors the activity of Twitter accounts and gives them a score on how much they behave like a bot, initially indicated that Musk is more bot-like than not when it did the rounds on Twitter this week. The tool scores an account based on how similarly it behaves to a bot, with a higher score meaning the account acts more like a bot. Musk's account received a score of 4/5 when we tested it out Tuesday. When we tested it early Wednesday, his account got a score of 1.4/5. We tried it out once more, and Musk’s account received a score of 0.5/5.
The Botometer highlights just how hard it is to identify bots, especially using only public data. The team behind Botometer wrote in a tweet that it’s hard to give an accurate score because of the high volume of retweets and mentions around Musk’s account, and a lack of data from the Twitter API can sometimes result in an abnormally high score. The Botometer team recommended looking at distributions over individual accounts. “But we agree that the task is hard!” the account wrote.
The Botometer highlights just how hard it is to identify bots, especially using only public data. The Tesla CEO said his purchase of Twitter "cannot move forward" because he thinks, without citing evidence, that the percentage of bots on Twitter is much higher than the 5% that the platform estimates. But determining what accounts are spam and which accounts are actually human beings is complicated. Twitter CEO Parag Agrawal tweeted Monday that the platform suspends about 500,000 suspected spam accounts each week, and its protocol for addressing these accounts is constantly updated to ensure Twitter isn't suspending real people. (Twitter also won't publicly detail how it roots out those spam accounts, but it's safe to say its analysis goes a little deeper than Botometer's tool.)
"Spam isn’t just ‘binary’ (human / not human)," Agrawal wrote in a Twitter thread. "The most advanced spam campaigns use combinations of coordinated humans + automation. They also compromise real accounts, and then use them to advance their campaign. So — they are sophisticated and hard to catch."
Musk has been laser focused on Twitter's spam issue, even going so far as to troll Agrawal on Twitter in response to his thread on Monday. He may be looking to get out of the deal, though Twitter is moving forward anyway. It's clear that determining whether an account is a bot or spammer is not a cut-and-dried process, and Musk's own account proves that point.
Is @elonmusk a bot? One of *the* most advanced algorithms for bot detection thinks so, which illustrates just how difficult it might be to clean up this bird place. \n\n\u00af\_(\u30c4)_/\u00af \n\ntry the tool: https://botometer.osome.iu.edu/\u00a0 and technical references here: https://botometer.osome.iu.edu/publications\u00a0pic.twitter.com/mtfua9RA1L— Chris Bail (@Chris Bail) 1652796001
Update:This post was updated to reflect the inconsistencies in scoring that have been recognized by the team behind Botometer. Updated May 18, 2022.