One of the things that annoys me the most about contemporary voice AI is what I call the ‘Conversational Funnel’. If you’ve ever used a mobile app with voice, it’s that situation where you say something (yes, with your voice) to an app and then the app asks a question back and then you provide it with a little more information and it says something back again. Like the following stilted conversation on an app I used recently:
Isn’t this just painful? Having a voice interface that prompts you constantly is not how human conversation happens. And even worse is the voice interface that doesn’t understand anything other than the response to the question it is asking.
Now, a conversational funnel is super easy to build. The system knows what it's listening for because it is aware of what it is prompting the user.
But, as human beings, we’re prone to changing our minds. And a part of what makes us human is that choices make us happy. And customer happiness is what makes us happy (corny, but true).
Building a system that replicates this conversational funnel is easy and reeks of lazy or incompetent programming. At 17E Tech, we've focused on modeling Leova's (https://www.leova.io) interactions on real human interactions. And just like you'd never end up in a conversational funnel except at the DMV, we wouldn't dare to put our users through something like that.
Building a system that allows and accounts for forks in a conversation, that is able to handle modifications mid-process is the closest that machines can come to an inter-personal interaction and we're proud to have been able to build artificial intelligence that comes so close to this model.
I plan to cover this topic in more depth at a later stage, but for now, I shall bid you adieu and have a great weekend!
Isn’t this just painful? Having a voice interface that prompts you constantly is not how human conversation happens. And even worse is the voice interface that doesn’t understand anything other than the response to the question it is asking.
Now, a conversational funnel is super easy to build. The system knows what it's listening for because it is aware of what it is prompting the user.
But, as human beings, we’re prone to changing our minds. And a part of what makes us human is that choices make us happy. And customer happiness is what makes us happy (corny, but true).
Building a system that replicates this conversational funnel is easy and reeks of lazy or incompetent programming. At 17E Tech, we've focused on modeling Leova's (https://www.leova.io) interactions on real human interactions. And just like you'd never end up in a conversational funnel except at the DMV, we wouldn't dare to put our users through something like that.
Building a system that allows and accounts for forks in a conversation, that is able to handle modifications mid-process is the closest that machines can come to an inter-personal interaction and we're proud to have been able to build artificial intelligence that comes so close to this model.
I plan to cover this topic in more depth at a later stage, but for now, I shall bid you adieu and have a great weekend!
No comments:
Post a Comment