By Michael Jensen
Hate speech on X was consistently 50% higher for at least eight months after tech billionaire Elon Musk bought the social media platform, new research has found.
The research looked at the prevalence of overt hate speech including a wide range of racist, homophobic and transphobic slurs.
The study, published today in PLOS ONE, was conducted by a team of researchers led by Daniel Hickney from the University of California, Berkeley.
It clearly demonstrates how a platform initially invented to help friends and family stay in touch has now metamorphosed into a place where hate speech is prolific. This is especially concerning given hate speech online has been linked to violent hate crimes offline.
A long list of promises
On October 27 2022, Musk officially purchased X (then known as Twitter) for US$44 billion and became its CEO. His takeover was accompanied by promises to reduce hate speech on the platform and tackle bots and other inauthentic accounts.
But after he bought X, Musk made several changes to the platform to reduce content moderation. For example, in November 2022 he fired much of the company’s full-time workforce. He also fired outsourced content moderators who tracked abuse on X, despite research showing social medial platforms with high levels of content moderation contain less hate speech.
The following month, Musk also disbanded the platform’s Trust and Safety Council – a volunteer advisory group of independent human rights leaders and academics formed in 2016 to fight hate speech and other problems on the platform.
Previous research has shown hate speech increased on X immediately after Musk took over. So too did the prevalence of most types of bots.
This new study is the first to show that this wasn’t an anomaly.
More than 4 million posts
The study examined 4.7 million English language posts on X from the beginning of 2022 through to June 9 2023. This period includes the ten months before Musk bought X and the eight months afterwards.
The study measured overt hate speech, the meaning of which was clear to anyone who saw it – speech attacking identity groups or using toxic language. It did not measure covert types of hate speech, such as coded language used by some extremist groups to spread hate but plausibly deny doing so.
As well as measuring the amount of hate speech on X, the study also measured how much other users engaged with this material by liking it.
The researchers’ access to X data was cut off during the study due to a policy change by the platform, replacing free access to approved academic researchers with payment options which are generally unaffordable. This significantly hampered their ability to collect sample posts. But they don’t mention whether it affected their results.
A clear increase in hate
The study found “a clear increase” in the average number of posts containing hate speech following Musk’s purchase of X. Specifically, the volume of posts containing hate speech was “consistently” 50% higher after Musk took over X compared to beforehand – a jump from an estimated average of 2,179 to 3,246 posts containing hate speech per week.
Transphobic slurs saw the highest increase, rising from an average of roughly 115 posts per week before Musk’s acquisition to an average of 418 afterwards.
The level of user engagement with posts containing hate speech also increased under Musk’s watch. For example, the weekly rate at which hate speech content was liked by users jumped by 70%.
The researchers say these results suggest either hate speech wasn’t taken down, hateful users became more active, the platform’s algorithm unintentionally promoted hate speech to users who like such content – or a combination of these possibilities.
The study also detected no decrease in the activity of inauthentic accounts on X. In fact, it found a “potential increase” in the number of bot accounts partly based on a large upswing in posts promoting cryptocurrency, which are typically associated with bots.
An important data-driving deep dive
There were a number of limitations to the study. For example, it only measured hate speech posts in English, which accounts for only 31% of posts on the platform.
Even so, the study is an important, data-driven deep dive into the state of X. It shows it is a platform where hate speech is prolific. It also shows Musk has failed to fulfil his earlier promises to address problems on X such as hate speech and bot activity.
As Musk himself said at the White House earlier this week: “Some of the things I say will be incorrect and should be corrected”.
