It is too bad that the automated speech to text isn’t good enough to allow them to turn this feature off without losing a great deal of high quality subtitling.
Writing high quality captions requires more than just knowing what they are saying, although that is a large part of it. Furthermore, knowing what they are saying may require more than just listening.
(I don't use YouTube, but this applies just as well to television.)
Even if they had captions, it would be impossible to translate certain things (e.g. lyrics, technical or scientific terms) to another language automatically well.
(I don't use YouTube, but this applies just as well to television.)