The demos look as exciting as any chatbot demos we saw before, and I think it will fail in practice just like how chatbots fail. With the exception of very few verticals in controlled settings, most real world tasks are way more dynamic and fluid than what we can comfortable code in some state machine. The article pointed it out, too

> One of the key research insights was to constrain Duplex to closed domains, which are narrow enough to explore extensively. Duplex can only carry out natural conversations after being deeply trained in such domains. It cannot carry out general conversations.

The question is after you limit it to "closed domains" narrow enough, where it can still be practically useful. It might help with certain functions in enterprise settings. It will definitely work for spammers because they can work with even 1% success rate.