Captions serve many more people than the ones who need them

Access provisions have a way of becoming everybody's preferred setting.

Captions exist first because some people cannot access audio, and that reason alone is sufficient. Obligations in this area exist, they differ by jurisdiction and by institution, and they should be checked locally rather than assumed. But the pattern worth noticing is what happens once captions are simply present: usage spreads far beyond the group they were provided for, and keeps spreading.

People watching in a shared house at night use them. People whose first language is not the language of instruction use them constantly, because reading a technical term is far easier than catching it in fast speech. People on public transport use them. People with unreliable audio equipment use them. People who process written words more comfortably than spoken ones use them, whether or not they have ever named that preference. The provision made for a specific need turns out to be a general improvement, which is the usual shape of these things.

Practically this means captions should be treated as standard output rather than as a request driven service, and their accuracy should matter to somebody. Uncorrected captions on specialist vocabulary can be worse than none, because a confidently wrong term teaches the wrong word. The correction pass is short and it is best done by someone who knows the subject, which usually means the person who taught it or someone close to the material, and that time needs to be acknowledged rather than assumed.