August 5 was not a normal day for Kaicheng Yang. It was the day after a US court published Elon Musk’s argument about why he should no longer buy Twitter. And Yang, a doctoral student at Indiana University, was shocked to find that his bone-detection software was at the center of a massive legal battle.
Twitter sued Musk in July after Tesla’s CEO attempted to withdraw his $44 billion bid to buy the platform. Musk, in turn, filed a counter-charge accusing the social network of misrepresenting the number of fake accounts on the platform. Twitter has long maintained that spambots represent less than 5 percent of the total number of users who can monetize, or users who can see ads.
According to legal documents, Yang’s Botometer, a free tool that claims to be able to identify how likely a Twitter account is to be a bot, has been instrumental in helping Team Musk prove that figure isn’t true. Contrary to what Twitter claims that his company was minimally affected by fake accounts or spam accounts, the Musk Parties’ preliminary estimates show otherwise,” Musk said in his counterclaim.
But telling the difference between humans and bots is harder than it sounds, and one researcher has accused Botometer of “pseudoscience” because it makes it look easy. Twitter was quick to point out that Musk was using a tool with a history of making mistakes. In its legal filings, the platform reminded the court that Botometer defined Musk itself as likely bot earlier this year.
Despite this, Botometer has become prolific, especially among university researchers, due to the demand for tools that promise to distinguish bot accounts from humans. As a result, in October not only Musk and Twitter will face trial, but also the science behind bone detection.
Yang didn’t start the Botometer; he inherited it. The project was started about eight years ago. But when the founders graduated and moved on from college, responsibility for maintaining and updating the tool fell to Yang, who refuses to confirm or deny whether he’s had any contact with Elon Musk’s team. Botometer is not his full-time job; it’s more of a side project, he says. He works on the tool when he is not doing research for his PhD project. “Right now it’s just me and my advisor,” he says. “So I’m the person who really does the coding.”
Botometer is a supervised machine learning tool, which means it has been taught to separate bots from humans on their own. Yang says Botometer distinguishes bots from humans by looking at more than 1,000 details associated with a single Twitter account, such as its name, profile picture, followers, and ratio of tweets to retweets, before giving it a score between zero and five. . “The higher the score, the more likely it is a bot, the lower the score, the more likely it is a human,” Yang says. “If an account has a score of 4.5, it means it’s probably a bot. But if it’s 1.2, it’s more likely to be human.”
