Contents
The Swiss Reto Strobl is responsible for advertising on YouTube. He also relies on artificial intelligence there.
In 2004 Google came to Zurich. And since 2004, Reto Strobl has been working for Google in Zurich. His employee number is correspondingly low: 1200 – out of a good 190,000 Google employees around the world.
As technical director, Strobl is now responsible for advertising that is placed on Google’s video platform YouTube. Ads that can be clicked away after a few seconds. Sounds like a bad deal for advertisers, but half the people stay tuned to the end, says Strobl. “When we started advertising that can be skipped at the push of a button in 2010, it was only about 17 percent.”
The machine learns which advertising is well received
So is advertising more popular today than it used to be? It’s probably just made better: Advertisers have learned what advertising on YouTube has to look like: If you give the audience the opportunity to press the skip button after five seconds, then so much has to happen in those first five seconds for it yet nobody does.
Caption: Reto Strobl studied computer science at ETH Zurich. Today he is responsible for video advertising at YouTube as technical manager. SRF
The fact that advertising can be skipped on YouTube has therefore changed how advertising is done. Just like the music streaming service Spotify (where a stream is only counted if it runs for at least 30 seconds) has an impact on how music sounds today: a lot has to happen right from the start – there has to be a hook that people can listen to binds the song or the advertisement.
AI can even recognize humor today
And just like Spotify, YouTube also knows a lot about its audience: “Our machine learning models are trained with data from user interactions and can thus determine which advertising is best shown to whom,” says Reto Strobl.
The “machine learning” he is talking about is often also referred to as “artificial intelligence”. And such AI is also used elsewhere on YouTube – for example to find out which advertising is particularly successful. So-called “big language models” also come into play here – the AI technology, which also forms the foundation of chatbots such as ChatGPT.
Reto Strobl is enthusiastic about the new possibilities that are opening up: “These large language models are incredible! They can recognize things that would have been impossible before – like humor.” Just two years ago, people said that you shouldn’t even try, after all, humor is far too specific a matter. “But these models can actually determine whether something is meant to be humorous or not!”
Future: Advertising will be generated automatically
Artificial intelligence on YouTube not only knows which advertising videos are particularly well received by the audience, but also what is funny about these videos. Wouldn’t it be possible for the AI to write the screenplay for a clip itself after entering a few keywords for the advertiser? “We are definitely heading towards that,” says Reto Strobl, “in ten years it will definitely be like that.”
And should AI continue to advance, even more would be possible. Then the machine could also generate the advertising clip itself: “That’s a goal we’re working towards – that you just have to say, ‘Give me a beach scene with palm trees and a couple in love going into the water at sunset running.’ And then it’s done automatically.”