Re:The State of Voice Assistants: Where's the AI?
by neuronDen on Sunday July 07, 2019 @11:34AM
(#20375595)
You don't need NLU to be fully solved for assistants today which work on narrow range on tasks. For example - modern deep learning techniques can easily mark "Send a message to <person> - What are you upto?" and "Check with <person> what he is upto" as similar. Assistants today respond correctly to the first input but not the second.
[ Parent
| on HN ]
Re:The State of Voice Assistants: Where's the AI?
by streetcat1 on Sunday July 07, 2019 @01:32PM
(#20376329)
So I am talking about spoken dialogue system. I.e. system that can converse (I.e. at least 15-20 turns between the user and the system).For example, a system that can book multiple legs flight.
Re:The State of Voice Assistants: Where's the AI?
by neuronDen on Sunday July 07, 2019 @01:45PM
(#20376438)
Microsoft acquired Semantic Machines has the capabilities to do so.
http://www.semanticmachines.com/
Re:The State of Voice Assistants: Where's the AI?
by streetcat1 on Sunday July 07, 2019 @02:32PM
(#20376807)
As I said. This is a problem from the 80s. Why would it would be solved now? and by a specific startup? what changed?. I think that in order to solve this issue we would need some breakthrough in symbolic AI and not just statistical AI. However, most of the research today is done on DNN and their derivative.