Intent-based Generative Bots vs Open-ended Generative Bots

In my view, there are two kinds of generative bots you can build using Dialogflow CX.

The example demo bot you saw in the previous lesson – which helps you choose between Dialogflow and GPT – is an intent-based generative bot. Your goal is to map as many of the user utterances as possible to an actual intent, and you want to steadily increase the ratio of number of mapped utterances over number of generative fallbacks.

This is because when you map an intent, it becomes much easier to do all of this:

• design followup actions
• extract entity values
• control conversation state
• review the answers for accuracy

It is also possible to build another type of generative bot using Dialogflow CX and I will call it open-ended generative bot.

Here, we do not have an explicit goal to map as many of the user utterances as possible to an actual intent.

For example, suppose you wish to build a bot which can help build a bot which can help with reminiscence therapy.

Such a bot does not have to be intent-based. Instead, you usually want to keep the conversation going for as many turns as possible.

I will also be explaining how to build such a bot in a future lesson so you can see the various tradeoffs involved in building open-ended generative bots.