"Do conversational technologies have a future" is something that an executive might ask herself when considering an initiative to automate call center conversations. It is also something that a developer might ask himself when considering whether to add a new conversational AI-as-a-Service tech to his stack of go-to tools.
I can't help but be expectedly biased when answering this question. This bias does not come from working for a conversational AI startup. Rather, I ended up at a conversational AI startup because of my strong conviction that I should not have to use my fingers to talk to my phone.
- They are command-response interfaces, not conversational. What I mean is that Alexa is great at turning on some music but you can't have a conversation about music with her, augmented with her playing selections of tunes to illustrate what she is talking about.
- Their voices are either too robotic (fail to pass for human) or too perfect (land in the uncanny valley).
- They are limited in the depth of their conversational structures and breadth of digressions.
- They have major issues with remembering the context of the conversation; many (if not most) treat each new reply as a new conversation.
- They are limited in data access.
- They are fully conversational. They don't require commands, instead they extract intents from ongoing conversation.
- They not only sound human, their voices have idiosyncrasies and inconsistencies that, on a subconscious level, identify the speaker as a human to our ears.
- They are able to generate new conversational pathways on the fly, leading to conversations of virtually unlimited depth.
- They hold context of the conversation and, just like a human, can jump between topics and hop back and forth along the conversational timeline.
- They have unlimited data access privileges to the systems they operate and to the wider world knowledge base.