An increasingly common refrain in machine learning is “intelligence is compression.” Folks who believe that might bristle at the distinction between learning and compression.
I doubt it because learned features already constitute lossy compression. The question is what kind of compression; lossy vs. lossless, learned vs. unlearned.