Scale

From Tournesol
(Redirected from Large-scale)
Jump to navigation Jump to search

An important feature of today's algorithms is their scale. Namely, such algorithms now affect billions of individuals on a daily basis. This is a critical consideration for ethics, as any small improvement of a large-scale algorithm affects positively billions of humans.

A few figures

YouTube receives around 500 hours of new video per minute YouTubePress Statista, and represents around 1 billion hours of watch-time per day YouTubeBlog-17 YouTubePress. YouTube's recommendation algorithm is responsible for 70% of views on the platform cnet-18.

In the US, the average time spent on social medias is around 144 minutes per day Statista.

Facebook removes around 6 billion fake accounts per year Statista Forbes-19 Reuters-19.

Google also removes billions of 'bad ads' per year, which corresponds to ~100 per second SearchEngineJournal-18 VentureBeat-19.

Scalability requires algorithms

At such scales, content moderation by humans is extremely costly. It is also hard to uniformize Radiolab-20, which may lead to unfair treatments of different contents.

Perhaps most importantly, it leads to post-traumatic stress disorder for moderators who must review hours of violent, sexual and disgusting videos on a daily basis BBC-18 TheVerge-19.