Comment by Traster | Hacker News Reader

Comment by Traster | Hacker News Reader

Parent Full thread

>YouTube is still trying to get away with "letting the machines do it", which works for most aspects of their business, but not this.

Citation needed. I mean, they do use AI for lots of stuff, but let's not describe it as working. It has routinely caused massive aggravation for their biggest content creators, it has been one in a group of changes that has basically made Youtube an unsafe monetization platform and driven the creation of places like Patreon,

Sure, this is the only AI fuck up that is going to get them sued in federal court and face charges of exploiting children, but it's not like the other AI stuff is working well. Basically, their AI is doing a bad job everywhere, it's just that this area is where they face legal liability and so they can't pull out their usual "but the machines told us to!" defence.

Heh, I was giving them the benefit of the doubt. They've been running on "algorithms" since 2005, and are still around, so I assume it works for them for some definition of working.

But you have a point. YouTube and Google get a lot of flack for things that are emergent properties due to their lack of humans.

This is survivor bias.

YouTube doesn't release many of the details of the methodology or even the content that they block because it can then be circumvented by the people who create and/or circulate it. It can create a streisand effect which only makes that type of content worse by drawing attention to it.

Some of what you can see is in the child sexual abuse material (CSAM) APIs that Google is now sharing to help other organizations detect CSAM material en-masse in their content. These services that were originally built to do so on GMail, Drive, and Youtube.

AI is really bad ad making judgement calls on things like fair use, advertiser friendliness, controversial content (eg. LGBTQ content), and now whether or not a video is directed to kids. See https://www.cs.princeton.edu/~arvindn/talks/MIT-STS-AI-snake...

To be fair, they're the largest non-pornographic video platform. Making them an endless target for any and all abuse possible to grow channels quickly and earn ad revenue.

It's inevitable that in such an environment the system they come up with ends up making less and less sense over time. It's death by a thousand cuts, and I don't think there's a good way to really avoid it.

Reply on news.ycombinator.com