Here's a thread: generative artificial intelligence has made breathtaking gains in the last four years. For instance, these are images of imaginary bedrooms and imaginary album covers created by a neural network. 
(For more on this, and these examples, see: jebruner.com/2018/03/optimi…)
Here's a generative adversarial network learning to "write" handwritten digits. imgur.com/X04j1Cv
Here's a neural network that can reconstruct low-resolution images into higher-resolution images, in the manner of those silly "enhance, enhance" scenes in police procedurals.
Here's a neural network that can translate black-and-white images into color, or satellite images into street maps.
Here's a neural network that can turn line drawings into photorealistic images of cats. (Even terrifying, unfathomable ones.)
Here's a neural network (@DeepDrumpf) that can imitate Donald Trump, with all his fixations and preoccupations. It even at-mentions Fox News and makes inappropriate comments about different ethnic groups.
Here's a neural network that can synthesize a video to match any audio. (Say goodbye to any kind of public trust in video footage.)
This entire arsenal of techniques is only ~4 years old! So, obviously AI-generated content is going to make its way into our entertainment streams.
Now, consider what happens when generative AI gets crossed with the vast amount of interaction data that we leave on platforms like YouTube, Kindle, Netflix, etc. YouTube, for instance, can show you a graph of audience engagement, second-by-second, for any video you upload.
Video platforms and high-volume creators can understand the impact of every single frame of a video on engagement, and can tweak their content to keep viewers engaged just a little longer.
As in, for instance, those horrifying children's videos that James Bridle wrote about. medium.com/@jamesbridle/s…
Video creators start by making lots of knockoff videos, then begin to vary them to figure out what engages viewers. They follow audience interest into producing bizarre, nonsensical content.
This is still human-driven engagement-seeking. Now imagine what happens when a giant computer cluster starts churning out AI-generated videos in bulk, creating tiny variations and measuring audience reactions, then reformulating and trying again.
It'll become feasible to do this at the individual level. Every time you stop mid-video, you tell YouTube something about your preferences—preferences that could guide a generative algorithm toward creating some new content that's just a little more engaging to you in particular.
Remember Polar Express? It used motion capture to digitize and animate Tom Hanks. Somewhere in LA there's a hard drive with a big corpus of Tom Hanks expressions, waiting to be reformulated into a new movie.
And that movie could be highly personalized: maybe it'll be a fast-cut action picture for some viewers, and a slow-moving romance for others. And to anyone who watches it, it'll be much more engaging than its human-made counterpart could ever have been.
A lot of people today talk about feeling addicted to apps, or social media, or the Internet in general. Wait until there's an infinite supply of perfectly-tuned content waiting for you.
I think this is what David Foster Wallace anticipated in Infinite Jest: entertainment so enthralling that it pushes everything else out. Read the excerpt here: jebruner.com/2018/03/optimi…
