1) There’s a huge consolidation around Siri, Alexa, and Google. No one wants to download an app, and you’re never going to win against wake words.
2) Doing the NLU and intent modeling is really hard to get right. All too often, someone comes along, thinks all they have to do skis throw together some regexps or toss the neural net du jour over some text and it will all just work. It won’t. It doesn’t. It’s just really really embarrassing.
Intelligent back off is a thing. Understanding when you don’t understand something is a thing. Conversation management is a thing. UX matters.
Also, trying to fake chatbots with people, and then saying “We’ll just train up a model and get rid of those folks” was never going to work, because you’d always have to have people, because you set your bar at hard AI then. Sure, that’s a decent strategy of you already have a call center that you’re trying to cut costs on, but for a startup, that’s just hubris.
1) There’s a huge consolidation around Siri, Alexa, and Google. No one wants to download an app, and you’re never going to win against wake words.
2) Doing the NLU and intent modeling is really hard to get right. All too often, someone comes along, thinks all they have to do skis throw together some regexps or toss the neural net du jour over some text and it will all just work. It won’t. It doesn’t. It’s just really really embarrassing.
Intelligent back off is a thing. Understanding when you don’t understand something is a thing. Conversation management is a thing. UX matters.
Also, trying to fake chatbots with people, and then saying “We’ll just train up a model and get rid of those folks” was never going to work, because you’d always have to have people, because you set your bar at hard AI then. Sure, that’s a decent strategy of you already have a call center that you’re trying to cut costs on, but for a startup, that’s just hubris.