I watch lots of videos on guns, cars, metalworking and history with the former two mostly geared toward history and manufacturing.

I never see (amateur) political talking heads recommended. It's all trash pop-history talking heads, semi-trash documentaries and low brow entertainment related to cars and guns (e.g. demo ranch and whistlin diesel).

I recently (like yesterday) watched a semi-political talking head discuss the economics of OnlyFans after a friend linked to that particular person's analysis so it'll be interesting to see if the algo tries to drag my content toward more talking heads.

Interesting. I imagine there is more to it than just the videos.

E.g. if you live in a generally pro-gun area I think they be algorithm would probably be less likely to assume that it’s worth showing you political content.

I guess its obvious that the algo must work for a large enough number of people and for its optimization function.

You may simply be that % of the false positive/false negative that is a cost of doing business.

I mean - the alternative to this is having Google employ all the out of work journalists and editors and manually curate lists for people. I think the cost of losing a few people is vastly lower than hiring that many people, in each culture and country around the world.

I’m not sure I’m saying anything about whether the algo works or not.

I've had mostly the same experience with youtube. I wanted to get into woodworking during COVID lockdowns, so I was watching a lot of popular woodworking channels that had nothing to do with politics, and was frequently shown Trump ads. Then once I realised how expensive woodworking would be, I started watching videos about game development. After ~1 week, all the Trump ads disappeared, and I started getting Biden ads.

> The moment I watch a gun video, I immediately get shown Ben Shapiro, Jordan Peterson, and The Blaze instead of all the cooking videos etc.

I know recommendations are based on many factors that differ between us (such as location), but I can't reproduce that behavior. If I watch hickok45 or Paul Harrell in an incognito window, I get recommendations for more of their videos along with a few from Demolition Ranch, Forgotten Weapons, and other gun channels. I see no political videos in the recommendations.

Interesting - well those are are the kinds of people I watch.

I’m using AppleTV, but not logged in. It’s possible that it takes some time for that to happen.

Also, I watched a Jerry Miculek Video yesterday, and got no political stuff, so it’s also possible that the algorithm has been improved or they have specifically acted to break this association.

> And yet none of the gun videos I watch are remotely political.

Sure, but to assume that you _probably_ hold conservative beliefs due to having an interest in guns is way more often the case than not. Same is true in an opposite way if you watched some video on abortion rights.

“to assume that you _probably_ hold conservative beliefs due to having an interest in guns is way more often the case than not“

Only in the crudest statistical sense. For example the fastest growing demographic of gun owners is black women, who are overwhelmingly Democratic voters.

And that’s exactly what the complaint about these algorithms is.

That these algorithms are crude statistical approximators of hidden latent biases in the training data? That's their entire method of functioning! That's the entire process of a neural network. It seems you're wanting something that can predict the future, which is currently outside of our matrix multiplication skills.

I’m not ‘wanting’ something.

I’m simply pointing out that the algorithms do in fact steer people towards polarizing content, that is not what they would have otherwise chosen.