This has always been present in subfields of AI. For example, in classical computer vision, one had to figure out specific parameters working just for a single

This has always been present in subfields of AI. For example, in classical computer vision, one had to figure out specific parameters working just for a single image or video scene by hand. Machine learning can in theory at least make these parameters learnable at the cost of complexity.