It does not necessarily mean building another computer chat. There are some applications of neural networks that exist for some time that now get rebranded as AI, such as: optical character recognition (the open source OCR engine Tesseract has been using neural networks for a while), automated and local translation (Firefox Translations, shipped in standard in the recent FF versions).
I think google has a thing that can translate text by taking pictures. You could now write an open source software that does this locally. It can provide solutions that do not depend on cloud, which is positive for SFOS and for the users.
It could also be a better engine for the suggestions when writing messages. You don’t have to know the algorithm behind, it will just make suggestions more contextually relevant.