Months before buying a social media giant, Elon Musk declared to the world he would crack down on hate speech and "defeat the spam bots or die trying".
Two years later, new research shows neither has happened.
[elon]Hate speech on the platform X spiked by 50 per cent in the months after the billionaire tech mogul bought it, according to analysis by researchers at the University of California and the University of Southern California Information Sciences Institute.
Their report was peer-reviewed and published in scientific journal PLOS One this week.
The spike — which included increases in specific homophobic, transphobic and racist slurs — continued from October 2022 through to May 2023, their report said.
Transphobic slurs surged the highest across the platform, according to the report, increasing from an average 115.2 posts per week to an average of 418 per week following Mr Musk's takeover.
This was an increase of 260 per cent.
Homophobic posts increased from 1,310 posts to 1,737 on average per week (a 30 per cent increase), and racist posts increased from 579 per week to 822 (a 42 per cent increase).
"Today, the world faces an unprecedented combination of large-scale emergencies and global interconnectivity," the report's authors concluded.
"Challenges such as these can only be met through cooperation, coordination, and compromise.
"[Social media platforms] serve as vehicles for the dissemination of hate and disinformation, pitting one group against another, splintering communities … and whole nations in ways that both lead directly to suffering and preclude solutions to global challenges.
"Moreover, the information landscape has become so large, complex, and rapidly changing that it is impossible for objective policymakers or journalists to monitor it unaided."
[chart]The report was funded via both the National Science Foundation and the Defense Advanced Research Projects Agency.
Researchers used a two-step process, collecting posts containing slurs and other terms associated with hate speech at five randomly selected times in five-minute intervals per day.
Their "toxicity" model detected whether a comment was "rude, disrespectful, or unreasonable".
The same model has been used previously to measure inflammatory content on X.
Reposts were included in the dataset, because a post's influence on X is amplified when it is shared — reports that appeared more than 40 times were checked manually.
X's hate speech claims 'lack transparency'
A Queensland tribunal ruled last year X was liable for hate speech published on its platform.
Shortly after his takeover of Twitter, which was renamed X, Mr Musk posted on the platform that "despite significant user growth", hate speech on the platform was trending downwards.
Publicly, X has said while it would not heavily police hate speech, it was working to decrease exposure to it for users.
[pie chart]"Musk and X have claimed that overall exposure to hateful content and spam on X has decreased since Musk's acquisition," the researchers said in the new study.
"However, these claims lack transparency, and so must be evaluated independently.
"For example, X could have a fundamentally different definition of hate speech than the broader research community.
"[They] now consider 'cisgender' a slur, despite this being a term widely used … in medical and gender nonconformity communities."
[likes]Dr Timothy Graham, an associate professor of Digital Media at the Queensland University of Technology, said the results were not surprising.
"One of the issues that we have as researchers [is that] Elon Musk has shut down the application programming interface, which let us hoover up huge amounts of data from the platform," he told the ABC.
"We don't have a 100 per cent strong empirical handle on just how bad the problem is."
"But all of the work that's been done, all the expert advice that's been given, all the anecdotal evidence and small scale in-depth research that's being done around the world moment is all pointing to hate speech continuing that surge."
The University of California Berkeley study also noted it was difficult to definitively say what had caused the increased rate of hate speech after Mr Musk purchased the platform.
[hate posts]Some actions by X, including "reducing the size of the trust and safety team", offered a "tempting" explanation but specific events or policies could not be singularly blamed.
The takeover also occurred just before the US midterm election, suggesting there were external factors that could have led to the rise in hate speech.
The authors added there was little information available to researchers when it came to X's moderation methods, and that hate groups often employ "coded" or covert language which may not have been picked up by their team.
"Nevertheless, these potential explanations do not rule out the possibility that direct changes to the platform made by the organisation had an impact," their report said.
"In April 2023, protections for transgender individuals were deleted from the hateful conduct policy, a change that plausible led to fewer transphobic posts being removed from the platform."
Increase in bots and crypto accounts on X
The California team also examined the number of bot accounts on the platform — looking for repeating hashtags, reposting, and accounts posting simultaneously.
They found "coordinated" accounts, or bot accounts, associated with information campaigns increased more than "non-coordinated" accounts following Mr Musk's X purchase.
There was no significant change in the activity or engagement for these accounts, but the report noted "an increase in the number of active bot-like accounts" and "a substantial upward trend in cryptocurrency posts".
[elon 2]"All of these results suggest no decrease, and a potential increase, in activity among in-authentic accounts," the report said.
"This could have grave consequences for user enjoyment of the platform, but also for potential scam victims and the stability of democracies in the face of disinformation."
However, they noted, as with the change in hate speech, external researchers have little information on how X has enforced different moderation strategies.
Dr Graham said it was important to look at what kind of bots were active on the site.
"When people hear the term [bot] they think of a problem, they think of misinformation, hate speech," he said.
"There are a lot of accounts that use scheduling software, that use automation, that are useful.
"There's a lot of good things out there, but the particular problem that X has is very much spam.
"It's a volume problem, where's difficult to basically go onto the site without basically being bombarded with porn bots, bots that make no sense, bots that are just replying to you using what is almost certainly some kind of [AI].
"It just becomes noise after a while."
He added there was also an incentive for users to create multiple spam bots to generate ad revenue from X.
"I think the trend that we're probably likely to see is that X is not going to change its ways," he said.
"It's not even often responding to take down notices, and when it does, it goes to court.
"What we'll probably see is just more polarisation, where you have people with similar political beliefs, with similar understanding of [the harm], moving to platforms like Blue Sky.
"[These platforms are] more moderated and … less centrally controlled by one rich billionaire.
"Everyone else who has those kind of politics and is willing to put up with the spam will stay on the platform.
"What we're seeing is empire building, but a Balkanisation of these empires. And I don't know whether that's necessarily a good thing.
"Time will tell."
