9 comments

[ 1.6 ms ] story [ 21.3 ms ] thread
Since the page doesn't mention this: the title is a reference to "What the Butler Saw", a famous early 1900s (mildly) erotic film.

https://en.wikipedia.org/wiki/What_the_Butler_Saw_(mutoscope...

Aaah I thought it was reference to 'What the dog saw', which is a Malcom Gladwell book. Seems it's also a reference to that film
It mentions it some way down the page.
>What the Robot Saw’s title is a reference to What the Butler Saw films. Neither the butlers nor the Robot could really understand the objects of their obsession. Despite their self-satisfaction, all they had was a squinted glimpse at a peep show.
Glad you got the reference! As ehnto noted, it does explain this down the page. Yep, "the Butler" peeped through a keyhole at something he wasn't supposed see (a woman undressing.) Seductive, but just a superficial glimpse of a person he didn't really know. And so it goes with robots, social media, etc...
I'd be really interested in seeing the source code for this!
This is a good artistic vision, but not quite what I was expecting. I thought this was a stream of the, perhaps pre-rendered, 'view' an algorithm takes of content to decide what to promote. By 'view' I mean like that weird generative art we all saw a few years ago that was created by querying the algorithm for its internal representation of, say, a dog.

This video is specifically content that's not popular, but appears to be rendered (live) with a human artist's own selection (perhaps randomly chosen, perhaps more algorithmic) of cutting, filtering, defocusing, etc. Maybe this difference is more subtle than I imagine. But I think it's clear that although this is probably content that The Robot Did Indeed See, it is not a representation of Twitter's or YouTube's algorithms' interpretations of that content.

Correct, more or less! It's what the robot saw, not how (which is done a lot, but this is not that)... Algorithms pick videos that don't get much human visibility - but their robots saw them.

The algorithms that generate the visuals and sound are generally traditional ('cept for the eyeballs: OpenCV) but use machine learning to make decisions. So the "AI" are editorial rather than generative. See the website for more details..

There are various machine learning algorithms in the mix, but primarily in the back end.. they generate the section labels, talking heads and eyeballs, differentiating sound/music for the sound mix, etc. They also filter "spam" videos. Many of the "AI" algorithms are behind the scenes or not the focus of attention (since it's what the robot saw, not how.) The lower third supers on the talking heads are more what you were probably expecting. But it's something very specific (not my algorithm) and explained on the site.

(comment deleted)