You can already kind of do this.
There exists GPT-3 already, and there is an API for it.
If you connect it to a speech recognition program or AI (there are plenty of open source ones) and send to the GPT-3 API, and then have a text to speech program (also AI lol) give an answer, then you have a product already.
May need to do some prompt tuning so you get "chatlike" answers rather than a wall of text though.
Well, with rapidly evolving speech and sound recognition technology, not long. There are a few tiny details to figure out here, but overall, AI is already really comfortable with audio processing, analyzing, and processing. In this article https://www.alltheragefaces.com/... you can learn more about some minor challenges with this technology, but also, ways to get rid of them. So, soon Artificial Intelligence will be all over our voice assistants.