Yann, Mark Zuckerberg called you five years ago to talk about AI – what did he say then?
I got a call from Mark around September/October of 2013, where he told me, that Facebook was interested in AI and kind of starting an activity in AI.
And that he wanted you to do this?
There were many options about how to go about it: You can buy a company, that is already in the business, you can hire a bunch of young people and try to grow this organically, or you can try to hire sort of an expert – and they eventually chose the last one. So I got that call from Mark, that he was going to stark an activity in AI, because it’s strategically important.
In the end, he convinced you to set up FAIR in New York.
Yes. And now, five years later, AI has become very central to Facebooks operations. Facebook today simply wouldn’t work without AI or, let’s say Deep Learning.
When I log in into my Facebook-Account, where do I meet AI?
First of all, there is all the content that you see and all the content, that you don’t see and both are largely determined by AI-Systems. What you see on your newsfeed are pieces of information, that the ML-Algorithms of Facebook have identified as the most likely to be of interest for you. So, there is a model of your taste inside of Facebook…
…tries to pick the pieces of information that interest you, are most relevant to you or are things that you should know even if those are not things that you enjoy knowing like one of your friends had a death in his family and you need to know about this. That’s not determined just because you are going to be interested in it or click on it, there is a number of criteria.
And the intelligence of that model is created by past data of my own behavior?
And of the behavior of my friends too?
Yes, everything. It’s your relationship to the person posting the piece of information, the topic of the piece of information and whether that matches your interest. For example, let’s say you have a hobby, say..
…I do play chess and am interested in Science Fiction…
…here we go: Maybe you have friends that are interested in Sci-Fi too and they post something about a new SF-novel, maybe the model will show that to you. If the same person posts a picture of a cat and our systems knows, that you are not interested in cat pictures, you are not going to be shown that. Or only one, but not all the twenty pictures, that the person has posted.
And the system is learning. If for example my interests are changing, it will get that?
Yes, absolutely. But it’s a little difficult, because the system has to show you things that it doesn’t know you might be interested in or not. For example, it has to show you Batman and sees your reaction. If the system never shows you any Batman, he will never know if you’re interested in Batman.
So the AI is trained to kind of disrupts me with new things from time to time, to test me?