Andrea Jones-Rooy

So that that way you're removing the user choice and you're just seeing what the algorithm gives you if you have the same decision.

And again, probably as you're listening to this, you're thinking now, and I had this as I read it, you're like, well, that's not how we interact on YouTube.

We interact on YouTube in this interactive way.

But the challenge is this endogeneity where it's,

Sort of like a media and politics problem.

Like, am I watching Fox News because I'm conservative or am I conservative because I'm watching Fox News?

This study is saying, what if we just put Fox News in front of people and see what's going on?

Or what if we put a bunch of TV channels in front of other people and we just let them click on the second one that's recommended, you know, or whatever.

I'm going to jump to the results.

Happy to talk in detail about more methods because it's all super interesting and very clever.

Imperfect.

It's just one experiment.

There's a lot more research to do.

Here are some findings.

I have four big findings for you.

One, YouTube users seems like, across all these studies, it seems like YouTube users are in what they describe as mild ideological echo chambers, but that those echo chambers seem to be driven by user behavior.

Again, in the way they've split this out, it does look like the interaction of the user with the algorithm is the key generator here.

There are kind of these echo chambers where it's like a bunch of videos with ideological similarity are kind of hanging out and in different spots, and you're more likely to find one versus the other, but they say it's actually pretty mild, and ending up in an echo chamber seems to be human-driven, okay?

Number two, generally, and this will come as a surprise to no one, recommendation algorithms prioritize content similar to what the user is already watching, and it does produce content rabbit holes.

So we have mild ideological echo chambers, but genuine evidence of content rabbit holes.