Elon Musk Demonstrates How Little He Understands About Content Moderation

Lots of talk yesterday as Elon Musk made a hostile takeover bid for all of Twitter. This was always a possibility, and one that we discussed before in looking at how little Musk seemed to understand about free speech. But soon after the bid was made public, Musk went on stage at TED to be interviewed by Chris Anderson and spoke more about his thoughts on Twitter and content moderation.

It's worth watching, though mostly for how it shows how very, very little Musk understands about all of this. Indeed, what struck me about his views is how much they sound like what the techies who originally created social media said in the early days. And here's the important bit: all of them eventually learned that their simplistic belief in how things should work does not work in reality and have spent the past few decades trying to iterate. And Musk ignores all of that while (somewhat hilariously) suggesting that all of those things can be figured out eventually, despite all of the hard work many, many overworked and underpaid people have been doing figuring exactly that out, only to be told by Musk he's sure they're doing it wrong.

Because these posts tend to attract very, very angry people who are very, very sure of themselves on this topic they have no experience with, I'd ask that before any of you scream in the comments, please read all of Prof. Kate Klonick's seminal paper on the history of content moderation and free speech called The New Governors. It is difficult to take seriously anyone on this topic who is not aware of the history.

But, just for fun, let's go through what Musk said. Anderson asks Musk why he wants to buy Twitter and Elon responds:

Well, I think it's really important for there to be an inclusive arena for free speech. Twitter has become the de facto town square, so, it's really important that people have both the reality and the perception that they're able to speak freely within the bounds of the law. And one of the things I believe Twitter should do is open source the algorithm, and make any changes to people's tweets - if they're emphasized or de-emphasized - that should be made apparent so that anyone can see that action has been taken. So there's no sort of behind-the-scenes manipulation, either algorithmically or manually.

First, again, this is the same sort of thing that early Twitter and Facebook and other platform people said in the early days. And then they found out it doesn't work for reasons that will be discussed shortly. Second, Twitter is not the town square, and it's a ridiculous analogy. The internet itself is the town square. Twitter is just one private shop in that town square with its own rules.

Anderson asks Musk why he wants to take over Twitter when Musk had apparently told him just last week that taking over the company would lead to everyone blaming him for everything that went wrong, and Musk responds that things will still go wrong and you have to expect that. And he's correct, but what's notable here is how he's asking for a level of understanding that he refuses to provide Twitter itself. Twitter has spent 15 years experimenting and iterating its policies to deal with a variety of incredibly complex and difficult challenges, nuances, and trade-offs, and as Musk demonstrates later in this interview, he's not even begun to think through any of them.

My strong intuitive sense is that having a public platform that is maximally trusted and broadly inclusive is extremely important to the future of civilization.

Again, this is the same sort of things that the founders of these websites said... until they had to deal with the actual challenges of running such platforms at scale. And, I should note, anyone who's spent any time at all working on these issues knows that maximally trusted" requires some level of moderation, because otherwise platforms fill up with spam and scams (more on that later) and are not trusted at all. There's a reason these efforts are put under the banner of trust & safety."

Finally, the public platform" is the internet. And trust is earned, but opening up a platform broadly does not inspire trust. Being broadly inclusive and trustworthy also requires recognizing that bad actors need to be dealt with in some form or another. This is what people have spent over a decade working on. And Musk acts like it's a brand new issue.

And so then we get to the inevitable point of any such discussion in which Musk admits that of course some moderation is important.

Chris Anderson: You've described yourself as a free speech absolutist. Does that mean that there's literally nothing that people can't say and it's ok?

Elon Musk: Well, I think, obviously Twitter or any forum is bound by the laws of the country it operates in. So, obviously there are some limitations on free speech in the US. And of course, Twitter would have to abide by those rules.

CA: Right. So you can't incite people to violence, like direct incitement to violence... like, you can't do the equivalent of crying fire in a movie theater, for example.

EM: No, that would be a crime (laughs). It should be a crime.

And all the free speech experts scream out in unison at the false notion of fire in a crowded theater."

But just the fact that Musk (1) agrees with this sentiment and (2) thinks that it would obviously be a crime shows how little he actually understands about free speech or the laws governing free speech. As a reminder for those who don't know, the fire in a crowded theater" line was a non-binding rhetorical aside in a case that was used to lock up a protestor for handing out anti-war literature (not exactly free speech supportive), and the Supreme Court Justice who used the phrase basically denounced it in rulings soon after - and the case that it came from was effectively overturned a few decades later, in the new case that set up the actual standard that Anderson suggests about incitement to imminent lawless action (which, in most cases, crying fire in a theater absolutely would not reach).

Anderson then tries (but basically fails) to get into some of the nuance of content moderation. It would have been nice if he'd actually spoken to, well, anyone with any experience in the space, because his examples aren't just laughable, they're kind of pathetic.

CA: But here's the challenge, because it's such a nuanced between different things. So, there's incitement to violence, that's a no if it's illegal. There's hate speech, which some forms of hate speech are fine. I... hate... spinach.

First of all, I hate spinach" is not hate speech. I mean, of all the examples you could pull out... that's not an example of hate speech (and we'll leave aside Musk's joke response, suggesting that if you cooked spinach right it's good). But, much more importantly, here's where Anderson and Elon could have confronted the actual issue which is that, in the US, hate speech is entirely protected under the 1st Amendment. And, we've explained why this is actually important and a good thing, because in places where hate speech is against the law, those laws are frequently abused to silence government critics.

But keeping hate speech legal is very different from saying that any private website must keep that speech on the platform. Indeed, keeping hate speech on a private platform takes away from the supposed trust" and broadly inclusive" nature Musk claimed to want. That would be an interesting point to discuss with Musk - and instead we're left discussing what's the best way to cook spinach.

Anderson again sorta weakly tries to get more to the point, but still doesn't seem to know enough about the actual challenges of content moderation to have a serious discussion of the issue:

CA: So let's say... here's one tweet: I hate politician X.' Next tweet is I wish politician X wasn't alive.' As some of us have said about Putin, right now for example. So that's legitimate speech. Another tweet is I wish Politician X wasn't alive' with a picture of their head with a gunsight over it. Or that plus their address. I mean at some point, someone has to make a decision as to which of those is not okay. Can an algorithm do that, or surely you need human judgment at some point.

First of all, broadly speaking all of the above are protected under the 1st Amendment. Somewhat incredibly, his final hypothetical is one I can talk about directly, because I was an expert witness in a case where a guy was facing criminal charges for literally Photoshopping gunsights over government officials, and the jury found him not guilty. But, also broadly speaking, there are plenty of legitimate reasons why a private platform would not want to host that content. In part, that gets back to the maximally trusted" and broadly inclusive" points.

But, on top of that, none of those examples are hate speech. Hate speech is not, as Chris Anderson bizarrely seems to believe, saying I hate X." Hate speech is generally seen as forms of expression designed to harass, humiliate, or incite hatred against a group or class of persons based on various characteristics about them (generally including things like race, religion, sexual identity, ethnicity, disability, etc.). The examples he raises are not, in fact, hate speech.

Either way, here's where Elon shows how little he understands any of this, and how unfamiliar he is with all that's happened in this space in the past two decades.

In my view, Twitter should match the laws of the country. And, really, there's an obligation to do that. But going beyond that, and having it be unclear who's making what changes to who... to where... having tweets mysteriously be promoted and demoted without insight into what's going on, having a black box algorithm promote some things and not other things, I think those things can be quite dangerous.

Again, in the US, the laws say that such speech is protected, but that's not a reasonable answer. We've gone through this before. Parler claimed it would only moderate speech that violated the law and then flipped out when it realized that people were getting on the site to mock Parler's supporters or to post porn (which is also protected by the 1st Amendment). Simply saying that moderation should follow the law generally shows that one has never actually tried to moderate anything. Because it's much more complicated than that, as Musk will implicitly admit later on in this interview, without the self-awareness to see how he's contradicting himself.

There's then a slightly more interesting discussion of open sourcing the algorithm, which is its own can of worms that I'm not sure Musk understands. I'm all for more transparency, and the ability for competing algorithms to be available for moderation, but open sourcing it is different and not as straightforward as Musk seems to imply. First of all, it's often not the algorithm that is the issue. Second, algorithms that are built up in a proprietary stack are not so easy to just randomly open source" without revealing all sorts of other stuff. Third, the biggest beneficiaries of open sourcing the ranking algorithm will be spammers (which is doubly amusing because in just a few moments Musk is going to whine about spammers). Open sourcing the algorithm will be most interesting to those looking to abuse and game the system to promote their own stuff.

We know this. We've seen it. There's a reason why Google's search algorithm has become more and more opaque over the years. Not because it's trying to suppress people, but because the people who were most interested in understanding how it all worked were search engine spammers. Open sourcing the Twitter algorithm would do the same thing.

Chris then gets back to the moderation process (again in a slightly confused way about how Twitter trust & safety actually works), pointing out that the algorithm" is probably less of an issue than all the human moderators, leading Musk to give a very long pause before stumbling through a bit of a word-salad response:

Well, I...I... I think we would want to err on the side... if in doubt, let... let... let the speech... let it exist. It would have... if it's.. uh... a gray area, I would say, l would say let the tweet exist. But... obviously... in a case where perhaps there's a lot of controversy where perhaps you'd not want to necessarily promote that tweet, you know... so...so... so... I'm not saying I have all the answers here, but I do think that we want to be very reluctant to delete things and be very cautious with permanent bans. I think time outs are better than permanent bans.

But just in general, like I said, it won't be perfect but I think we want to really have the perception and reality that speech is as free as is reasonably possible and a good sign as to whether there is free speech, is is someone you don't like allowed to say something you don't like.' And if that is the case, then you have free speech. And it's damn annoying when someone you don't like says something you don't like. That is a sign of a healthy, functioning free speech situation.

Again, so much to unpack here. First off, that approach of when in doubt, let it exist" has almost always been the default position of the major social media companies from the beginning. Again, it's important to go back to things like Klonick's paper which describes all this. It's just that over time anyone who's done this quickly learns that fuzzy standards like when in doubt" don't work at all, especially at scale. You need specific rules that can be easily understood and rolled out to thousands of moderators around the world. Rules that can take into account local laws, local contexts, local customs. It's not nearly as simple as Musk makes it out to be.

Indeed, to get to the spot that we're in now, basically all of these companies started with that same premise, realized it wasn't workable, and then iterated. And Musk is basically saying I have a brilliant idea: let's go back to step 1 and pretend none of the things experts in this space have learned over the past decade actually happened."

And, again, Twitter and Facebook - just as Musk claims he wants - tend to lean towards time outs over permanent bans, but both recognize that malicious actors eventually will just keep trying, so some people you will have to ban. But Musk pretends like this is some deep wisdom when every website with any moderation at all knew this ages ago. Including Twitter.

Second, his definition of free speech is utter nonsense (and ridiculously got a big applause from the audience). That's not the definition of free speech and if it is, then Twitter already has that. Tons of people I dislike are allowed to say things I dislike. You see that all over Twitter. But that's not a reasonable or enforceable standard at all without context. The problem is not someone I dislike saying something I dislike" the problem is spam, abuse, harassment, threats of violence, dangerously misleading false information, and more. Musk not understanding any of that is just a representation of how little he understands this topic.

Anderson then asks Musk about what changes he would make to Twitter, leading Musk to basically contradict everything he just said and go straight to banning speech on Twitter:

Frankly, the top priority I would have is eliminating the spam and scam bots and the bot armies that are on Twitter. You know, I think, these influence... they make the product much worse.

Um, nearly all of those are legal (the scam ones are a bit more hazy there, but the spam ones are legal speech). And just the fact that he acknowledges that they make the product much worse underlines how confused he is about everything else. Dealing with the things that make the product much worse" is the underlying point of any trust & safety content moderation program - and tons and tons of work, and research, and testing have gone into how Twitter (and every other platform) tries to manage those things, and they all pretty much end up at the same place.

To deal with the spam and the scams and the things that make the product much worse" you have to have rules, and you have to have enforcement that deals with the people who break the rules, meaning that you have to have people knowledgeable about content moderation and who are able to iterate and adjust, especially in the face of malicious actors trying to game the system.

But it's quite incredible for him to say pretty much leave it up if it's legal" one moment, and the next moment say his top priority is to get rid of spam. Spam is legal.

And, again, as anyone who has lived through (or read up on) the history of content moderation knows, platforms all went through this exact process. The process that Musk thinks no one has actually done. They all started with a fundamental default towards allowing more speech and moderating less. And they all realized over time that it's a lot more nuanced than that.

They all realized that there are massive trade-offs to every decision, but that some decisions still need to be made in order to stop making the product worse" and to figure out ways to build maximal trust" and to be broadly inclusive." In other words, for all of Musk's complaining, Twitter has already done all the work he seems to pretend it hasn't done. And his solution" is to go back to square one while ignoring all the people who learned about the pitfalls, challenges, nuances, and trade-offs of the various approaches to dealing with these things... and to pretend that no one has done any work in this area.

Every time I post about this, Musk's fans get angry and insist I couldn't possibly understand this better than Musk. And, again, I actually really admire Musk's ability to present visions and get the companies he's run to achieve those visions. But dealing with human speech isn't about building a car, a robot, a tunnel, or a rocket ship. It's about dealing with human beings, human nature, and society.

None of this is to say that, if Musk does succeed in the bid, he doesn't have the right to make these massive steps back to square one. Of course he has every right to make those mistakes. But it would be a disappointing move for Twitter, a company that has been more thoughtful, more careful, and more advanced than many other companies in this space. And it would likely wipe out the important institutional knowledge around all of this that has been so helpful.

I know that the narrative - which Musk has apparently bought into - is that Twitter's content moderation efforts are targeted at stifling conservatives. There is, yet again, no actual evidence to support this. If anything, Twitter and Facebook have bent over backwards to be extra accommodating to those pushing the boundaries in order to use Twitter mainly as a platform to rile up those they dislike. But, from knowing how much effort Twitter has actually put into understanding interventions and how to build a trustworthy platform, I fear that what Musk would do with it would be a massive step backwards and a general loss for the world.

Incredibly, there's a pretty good analogy to all of this earlier in that video. At the beginning, Anderson plays a snippet of a taped interview he did with Musk a week ago (when they weren't sure if he'd be able to attend in person). And in that interview, Anderson points out that Musk predicted to Anderson five years ago that Tesla would have full self-driving working that year, and it obviously has not come to pass. Musk jokes about how he's not always right, and explains that he's only now realized that just how hard a problem driverless artificial intelligence is, and he talks about how every time it seems to be moving forward it hits an unexpected ceiling.

The simple fact is that dealing with human nature and human communication is much, much, much more complex than teaching a car how to drive by itself. And there is no perfect solution. There is no congrats, we got there" moment in content moderation. Because humans are complex and ever-changing. And content moderation on a platform like Twitter is about recognizing that complexity and figuring out ways to deal with it. But Musk seems to be treating it as if it's the same sort of challenge as self-driving - where if you just throw enough ideas at it you'll magically fix it. But, even worse than that, he doesn't realize that the people who have actually worked in this field for years have been making the kind of progress he talked about with self-driving cars - getting the curve to move in the right direction, before hitting some sort of ceiling. And Musk wants to take them all the way back to the ground floor for no reason other than he doesn't seem to recognize that any of the work that's already been done.