OCR in images is becoming the norm. It was a big relief on social media for blind and visually impaired people. Asking for manually created image descriptions and/or transcribed text was too much to ask from the social web. So ML found one of its first public use cases. Composite image description and OCR. And yes, the quality is bad. But nobody really cared about this, esp. not those offering the feature. Holy heartlessness basically says "it is for free, so be happy that we gave it away. If it fails, that is not our fault."