The mechanism of introducing an information bottleneck, e.g. by changing from prose to poetry and trying to recover the prose, seems similar to autoencoder techniques that are popular in machine learning.