A lot of the more useful information retrieval tasks involve a feedback loop. If I’m shopping for a product, I may enter a generic term. Then the system sends me images of products matching that term. Then I tell the system which product image is closest to what I want. Then the system sends me reviews of that product. Then I read the reviews and realize this is not what I actually want…repeat until I’m satisfied.

I can’t do this loop with modern voice assistants.

A voice assistant with contextual conversation skills, and access to an “always on” visual monitor (home projector or AR glasses) would definitely increase utility by 10x or more