Former Musk ally explains resignation from Twitter, cites “dictatorial edict”

Former safety chief Yoel Roth's advice for Musk: "Humility goes a really long way."

During Elon Musk's first two weeks as the owner of Twitter, he seemed to have an ally in Yoel Roth, who had been the social network's head of trust and safety for seven years. Roth defended some of Musk's early actions and touted Twitter's work to crack down on hateful conduct. Musk pointed his followers to Roth's tweets explaining Twitter's content moderation work and said Roth has "high integrity"—even though they disagreed on whether former President Trump should have been suspended.

"I recommend following @yoyoel for the most accurate understanding of what’s happening with trust & safety at Twitter," Musk wrote on October 31.

But Roth quit on November 10, a day after Musk rolled out the $8-per-month Twitter Blue subscription that came with verification checkmarks. Musk had ignored prescient warnings from Roth's team and was forced to backtrack from paid verification after scammers eagerly paid $8 for checkmarks that made it easy to impersonate prominent accounts. Musk's revamped Twitter Blue was set for a relaunch this week but was delayed again.

Roth recalled how the paid-checkmark mess led to his resignation in an interview yesterday with Kara Swisher at a Knight Foundation conference. "It was not an easy decision," Roth said, continuing:

I was weighing the pros and cons on an ongoing basis. I knew what my limits were and by the time I chose to leave, I realized that even if I spent all day every day trying to avert whatever the next disaster was there were going to be the ones that got through. And Blue verification got through over written advice prepared by my team and others at Twitter. We knew what was going to happen. It's not that it was a surprise. It failed in exactly the ways we said it would.

Musk rules by “dictatorial edict”

The above quote and other portions of the interview can be seen on YouTube. For parts we haven't been able to view, we're relying on news stories about the Roth/Swisher interview.

"Before Musk took over Twitter, Roth wrote down several commitments to himself that would trigger the decision to quit. One limit, he said—one that was never reached—was that Roth would refuse to lie for Musk," a CNN article said.

But another one of Roth's limits was reached. "One of my limits was if Twitter starts being ruled by dictatorial edict rather than by policy... there's no longer a need for me in my role, doing what I do," he said, according to Reuters.

That echoes Roth's statement in a New York Times op-ed on November 18. Roth wrote that "many of the changes made by Mr. Musk and his team have been sudden and alarming for employees and users alike, including rapid-fire layoffs and an ill-fated foray into reinventing Twitter's verification system."

Noting Musk's "impulsive changes and tweet-length pronouncements about Twitter's rules," Roth wrote that "Musk has made clear that at the end of the day, he'll be the one calling the shots. It was for this reason that I chose to leave the company: A Twitter whose policies are defined by edict has little need for a trust and safety function dedicated to its principled development."

In yesterday's interview, Swisher pointed out that Roth defended Musk shortly after his October 27 purchase of the company. "One of the things you said at one point, you tweeted that Twitter was actually safer under Elon. Do you still feel that way?" Swisher asked.

"I don't," Roth replied. He said his earlier tweet came as Twitter was fighting a trolling campaign organized on 4chan. "They were like, 'let's go to Twitter and test the new limits of Elon Musk,'" Roth said.

"I tweeted a graph that showed clearly there was... a trolling campaign, and we shut down the trolling campaign, and we took steps to build technology that addressed that type of conduct automatically and proactively," he said. The new technology "reduced the prevalence of hateful conduct on Twitter relative to the pre-troll campaign baseline, which is great. That unequivocally was a win for trust."

But several things changed under Musk's leadership after that. When Swisher mentioned Twitter abandoning its COVID misinformation policy, Roth replied that "one way of streamlining the work of trust and safety, I guess, is to have fewer rules," and he even called Twitter's public statement on the change a "minor win" for transparency because "I wasn't really expecting clear announcements about policy changes."

But the COVID "policy change is really bad and damaging," he added. "You can certainly streamline things but that doesn't mean that malicious activity is going to get less complicated. It doesn't mean trolls are going to stop. You can't bury your head in the sand."

Roth: Machine learning can’t replace humans

Roth said layoffs were less heavy in his department than in the rest of the company. But the mass layoffs and resignations under Musk make it harder to moderate the platform despite advances in automated moderation technology, he said:

You can't use ML [machine learning] for all of it, you can't automate it, there is no 'set it and forget it' when it comes to trust and safety. I think Twitter's challenge going forward is not can the platform build machine learning—sure, they can—but are there enough people who understand the emergent malicious campaigns that happen on the service and understand it well enough to guide product strategy and policy direction. I don't think there are enough people left at the company who can do that work to keep pace.

Roth also discussed the hectic week preceding his resignation. There were the US elections, which Roth said Twitter handled pretty well. "They were not free of malicious activity, but nothing went spectacularly wrong on Twitter. That means we did a good job," he said.

There were Musk's mass layoffs and "the return to office email... you have people who should be spending their time thinking about how to deal with hateful conduct who instead are wondering, 'Do I really need to go to the office; I don't live anywhere near a Twitter office; am I still employed?' So you have all of the HR-related stress on top of this," Roth said.

Musk rolled out paid checkmarks around the same time. Roth said the basic idea of requiring payments to raise the cost of scamming works in some Internet contexts but not on Twitter in the way it was implemented.

"The problem is that the way it was rolled out and the way that it was implemented, and especially the dynamics of an extremely online, trolling-heavy platform like Twitter, is that it went exactly off the rails in the way that we anticipated and there weren't the safeguards that needed to be in place to address it upfront," he said. It was "a bad decision" made against the advice of Twitter staff, he said.

Swisher concluded the interview by asking Roth if he has any advice for Musk. "Humility goes a really long way," Roth said. Swisher laughed and said, "That one, I gotta tell you, that's not going to happen."

"I don't think so, either," Roth said.

Source