As promised in my intro post here's a piece on future and AI.

I can easily imagine in far future computers being able to every kind of video imaginable. Like for example, the leader of some country declaring war on another country. Or saying something that would ruin that country's reputation (for a while). Or ruin that leader's reputation. Yeah, in far future...

It turns our that future is pretty near. The research in generative models (neural nets that "generate" stuff, usually text/video/audio) is going at full speed right now. And then regular AI enthusiasts pick the research papers and do those crazy things! Look at this:

(credit Gene Kogan, see this twitter post)

This is an application of this paper.

Of course this is not there yet, you can clearly see it's a generated content. But how long do you think will it take a bug company to polish this approach enough to be indistinguishable for regular video?

Of course video alone is not enough to fool humans! Oh no! So we're good on this front, right? Not so fast. Listen to this.

Again, you can clearly hear the voices are generated. But pretty soon we won't be able to tell the difference. The company behind this clip will make sure of that (that's their business).

It turns out that in not too distant future we won't be able to trust the videos we see. Ironically enough we'll need AI help to detect frauds. But maybe it can also make us re-thing how our governments are being run? Do we even need presidents? Food for thought.

Take care!