Sci-fi. Dystopia. AI. Copyright issues. This film has it all.
Terence Broad, a London-based researcher, recently completed a dissertation entitled “Autoencoding Video Frames,” which we’ll admit doesn’t sound all that exciting, not initially. Add deep learning and Blade Runner to this mix, however, and you’ve got our attention. Broad built a neural network – a convolutional autoencoder – which taught itself all about Blade Runner (as well as another Philip K. Dick film, A Scanner Darkly), so that it could identify whether video frames were from the film or not.
Then, the encoder started reducing frames from Blade Runner into 200-digit representations… and making new frames out of them.
Why Blade Runner? Well, kinda obvs:
[T]here could not be a more apt ﬁlm to explore these themes [of subjective rationality] with than Blade Runner (1982)… which was one of the first novels to explore the themes of arial subjectivity, and which repeatedly depicts eyes, photographs and other symbols alluding to perception.
How’d it do? Good enough for Warner Bros. to issue a DMCA takedown request to Vimeo over the footage.
Check out the comparison here:
Whether or not androids dream of electric sheep, their riffs on copyrighted work will create a lot of grey areas in the future, to be sure. This one gets the issue off to an interesting start.