As machine learning advances, it becomes progressively easier to transfer styles from one work of art to another. For example, if you supply enough side-by-side examples of photographs and impressionist paintings of those photographs, an ML algorithm will then be able to produce an impressionist painting from any photo.
I’ve been wondering whether we will eventually be able to broaden the reach of ML for style transfer. For example, suppose we took the brilliant 1934 film The Thin Man starring William Powell and Myrna Loy.
Many young people today might find that film inaccessible. It’s in black and white, the styles, attitudes and cultural variations are more than eight decades old, and the humorous banter and double-entendres that once worked so well might seem incomprehensible to Millennials.
But suppose we could run that movie through a style-transfer machine. Instead of a brilliant 1934 film, we might get a brilliant 2019 film. The algorithm would find a modern equivalent for every clever line of dialog, every flirtatious look, every subtlety of class distinction.
We might very well end up with a modern classic. Or maybe it wouldn’t work at all — maybe the result would be simply painful to watch.
That would be interesting too. After all, an outright failure might suggest an upper limit on the powers of machine intelligence to replicate the nuances of human culture.