Screenshots of text are now parsible by bots. It’s a problem for Captcha users, but it’s also a problem for people who post screenshots to get around language filters… right?
To start off describing the phenomena I’m trying to describe, we need background. Advertisers control a lot of the internet because they spend a lot of money. Websites relying on ad revenue can’t afford to peeve off the advertisers. Users of certain websites who want to make money are often also siphoning off of the revenue the advertisers are bringing, and if they damage their relationship with the website, they risk losing access to that revenue stream. This environment disincentivizes swearing and other PG-13 content, and is the reason why the internet has slowly gotten simpler and simpler.
We can actually see the first traces of this during the Youtube “Adpocalypse”, where someone realized a Coca-Cola commercial was playing on an old racist cartoon. Following some bonkers logic, this meant that A) Coca-Cola was “fine with” being associated with racist videos and B) Youtube and possibly the original uploader of that racist cartoon were making money off of racist content. Coca-Cola reacted very strongly, Youtube reacted equally as strongly, and now Youtube demonetizes or censors videos that contain certain content.
And, as a private company, this is arguably their right – demonetizing isn’t even silencing the original poster, it’s just removing the ability to make money off of it. This makes creating content for fellow adults instead of an all-ages crowd (and doing so profitably) pretty tough, but it’s not like the video isn’t there, it’s just not making money.
This is one company, and it’s one way of handling it. Other companies had other ways of handling it, but the general thread of “protecting the advertisers’ interests” runs through all of them. All of the super-major content platforms, therefore, are incentivized to lean more towards kid-friendly content, or at least content that’s not strictly adult-only. But what is adult content?
Here’s the problem. Does talking about a murder for a minute in a five-minute video on TikTok make the video unfriendly to advertisers? Does describing a “spicy” scene in a book set off the auto-censor and shadowban you? Does having the word “suicide” on screen make you into a threat to children?
And if it does, and you find common euphemisms, do the common euphemisms do the same thing? Are you actually sparing anyone any grief by censoring yourself?
And then, furthermore, are the words bad because they’re bad words or because they describe morally impure actions? Why are we censoring at all??? I saw the word “stole” get this treatment. Like, in a sentence, “and then Pooh Bear stole an apple from Piglet, and he was very sorry”, the word “stole” got censored with random letters.
There’s a question of what this even does. We have autocorrect. Common misspellings get added to moderator dictionaries so nobody can call someone a slur by misspelling it in chat forums. The word “mvrder” isn’t just sneaking by. It’s not like doing this pseudo-censorship is the same as actual censorship; if a kid scrolls into a video describing a graphic crime, they’ll definitely still be able to connect the dots, and so will the auto-moderator programs screening content. Furthermore, most of these euphemisms are working off of the assumption the entire video and all of the audio is getting picked apart.
Kids re-uploading memes maybe don’t want their parents to think they were swearing, sure. It’s a little bit of a silly leap in logic to make, but it’s possible. Teenagers and young adults attempting to blow up on TikTok, however, are going beyond that and making up words to replace no-no words like suicide or murder, seemingly purely to appease an algorithm they think is detecting swears and then de-prioritizing the video the swears are in. And then, from there, also censoring words that are written on a picture in the video. Parsing audio is one thing – having a moderator bot scan the automatic closed-captions after the video is uploaded could happen if TikTok cared about that. Video-scanning for specific words, not even to ban users, just to de-prioritize them, would take a ton of processing power. You’re allowed to swear on TikTok – that’s kind of the bargain they made with being for the 13-and-over crowd. Advertisers know who they’re advertising to, roughly. If anything, an advertiser is probably going to take a meme with a bad word in it over a thirst-trap produced by someone who may be underage, because TikTok doesn’t have any real way of dealing with those or the potential problems they present.
So where does the superstition come from? Are these videos actually performing worse, and if so, is it really the words alone? Does posting content that’s full of swears and complex topics deprioritize it in the algorithm, or is it possible that people themselves are just not clicking “like” on a picture that has the F-word in it 10+ times, because they know their relatives can see them do that? The truth is that some of these algorithms are so complex that nobody can say for sure what makes X result more likely to appear for someone than Y result within a given set of people who have similar habits. The way advertising in the US works, advertisers have oodles of data on basically everyone on TikTok, Facebook, Instagram, et cetera – and that data was initially gathered by the apps themselves surveying browsing habits. We cannot know, and that makes it all opaque and confusing.
