Video-based content is already difficult to effectively moderate, because problematic content might be showcased visually, without keywords an automated AI might be able to check. And TikTok users are known for inventing shorthand and coded language specifically to evade censorship. There’s corn — or even just a corn emoji — for discussing porn, or “seggs” for sex. In the world of antisemitism, “H!tl3r” or “that Austrian painter” helps users talk about Hitler without detection.
A while back YouTube disabled the home feed recommendations unless you have watch history turned on. The problem is, under the old behavior if you had history turned off, it’d only recommend videos based on your recent likes and watch later playlist - in other words, it’d recommend things based on what you like to watch rather than what’s popular.
This meant users with watch history turned off got better, more relevant suggestions than those subject to the normal algorithm (where you’d get constant right-wing influencer spam if you clicked on a single vaguely political video). So of course Google disabled it. Can’t have a feature that prioritizes customer satisfaction over engagement, after all!
Which makes total sense from Google’s point of view. They get paid when people watch YouTube. If I log on and see that Lazerpig hasn’t posted a video and then log off, I’m not using YouTube.
But if I log on and they recommend a historical cooking show I’ll watch like ten of them in a row before I remember what I even came there for.