Zero.
Unless by "AI" you mean "software tool", in which case full-text search in Calibre ebook reader, language training with Clozemaster (I don't use the LLM-generated definitions, nor the images--too disturbing, and I dislike the refusal to use human-made art, like that at thenounproject), spam filters for email, Libby for library access (I now refuse to click on the LLM-generated suggestions for related reading), and The Internet, where I prefer sites like text.npr.org and solar.lowtechmagazine.com because they're static.
The main exception I'll easily grant for LLM use is towards equitable access for those with disabilities, and possibly for text-to-speech in general, though both of these things can be addressed with more human-to-human support; we don't have to do everything by ourselves.
I have hearing aids with a deep-learning based noise reduction and speaker isolation/clarity features. So about 14 hours a day. Or have we moved the goal posts again?
The main exception I'll easily grant for LLM use is towards equitable access for those with disabilities, and possibly for text-to-speech in general, though both of these things can be addressed with more human-to-human support; we don't have to do everything by ourselves.