TFA briefly mentions it but I can't imagine how "Voice UI" can be even decent without strong AI. In order to be more efficient than inputting commands with a keyboard or mouse you want an "assistant" that will be able to infer what you want from context. And of course you want to be reasonably certain that it'll get it right without having to double check everything all the time.

Also, make it available offline. I don't want $bigdata to know everything I do on my computer. Amazon Echo & friends are way too orwellian for my taste. I don't know if the general public will care about that however.