nostalgebraist:

About those creepy algorithmic kids’ videos:

That Medium article about them is just bad.  Badly written, badly argued, conflates a lot of things because they strike the author as “wrong” even if some are much more explicable than others.

Most of the videos he cites are based on very simple forms of children’s edutainment, like little exercises that teach color names.  It makes sense that a certain breed of content creators would say “we can generate these automatically by swapping in different characters, and the more videos we have including popular characters, the more clicks we will get.”  So you get color name exercises with every possible permutation of some Disney characters.

It’s admittedly a lot weirder that people will act out the very same templates in live action.  Maybe the animations were notably lucrative, and some people thought “if we do the same things in live action, that will seem more appealing and we’ll steal their clicks for any given search term.”  There has to be some extra cost associated with live action (one does wonder how much the actors and cameramen are paid), but maybe it’s worth it.

Where it started to strike me as weird was in the very last section.  There, he talks about a type of video that is a lot longer, and has narrative elements and not just color/name games.  There are hundreds of these, closely following remixes of the same basic scene templates, which individually make little sense and are presented in random sequences with no greater plot.  This is a typical entry in the genre:

There are some color-game scenes, although here they are more elaborate and bizarre (as superheroes and supervillains swim in a swimming pool, the Joker sprays them with paint; we learn the name of each paint color).

There are also other scene archetypes that recur: “action scenes,” composed of many repeated shots in which one character punches/kicks another out of the frame, and “chase scenes,” in which some monster chases characters and wreaks havoc until it is defeated.  Sometimes archetypes are combined: in the video above, the heroes are chased by a shark which randomly changes color, and we learn the names of the colors.

At first, I thought these were made entirely by computer, and that the “scene archetypes” were completely pre-made scenes into which character models could be swapped, Mad Libs-style.  But after watching some more of these videos, I realized that can’t be it.  The scenes are too different and too well-tailored to their differences.  In a chase scene with a massive snake, it slithers along a street, knocking aside cars, and then destroys a building; in a chase scene with a shark, it rises from the ocean and we see shots of sailors on capsizing boats.

So some human sat down to animate a chase scene with a shark, and came up with a pretty good one (at one point, the shark grabs a motorcycle in its jaws).  Was that part of some pre-existing library of templates?  But what about the other shark chase scene, the one that teaches you colors, where the shark rises out of a swimming pool and then swims through the air?  Did someone make that archetype for a pre-existing library?  Or was it for that video specifically, given how these videos seem to like sharks?

The kicker: people do live-action renditions of this stuff, too.  In addition to the animations, which are produced by multiple channels, there are multiple channels making the dream a reality:

The adherence to the template is very careful.  Yet the actors, as they must, add little human affectations – you can see them struggling to figure out what “act the part” means in these senseless parts.

Maybe this is all because someone made some money on this template, once, and everyone else jumped on the bandwagon.  Everyone knows the template gets clicks, and so everyone is trying to get a piece of that pie.  And so we have hundreds of lovingly exact renditions of the template, even though the template is terrible.  (Kind of an accidental satire of Hollywood.)

a lot of the “eerie” stuff I’ve seen is pretty close in patterning to the goanimate Caillou/“grounded” stuff we know real kids will make given computer animation tools