UNDERSTANDING CONVERSATIONS AI MICROCOURSE

CONVERSATION DESIGN TOOLS

Tools and techniques for conversation design.

Our Designing Conversations microcourses are supported by the Conversation Design Institute, the world’s leading training and certification institute for conversation design with students in over 50 countries around the world.

What tools can we use to design conversations?

In the WHAT IS A CONVERSATION microcourse, we looked at how conversation is cooperative and some of the rules for human conversation.

When we looked at creating conversations between machines and people, we saw the importance of personality, gender and vocabulary. In the CONVERSATION DESIGN microcourse, we learned how conversation designers could use this information to build natural human-computer conversations.

We also learned how a conversation designer works out what are called intents - a set of questions the human might ask an AI about different topics and then designs the responses - what the AI says in return.

In the DIALOGFLOW microcourse, you’ll have the chance to program your own intents and responses using Google Dialogflow, so let’s learn more about how intents work in Dialogflow.

To get started, watch this video from Google, which talks about chatbots and how Dialogflow works. There are lots of really detailed videos in this series you can watch, too.

INTENTS

Our digital person, Hailey, is programmed using Google Dialogflow.

Here are the intents that Hailey can recognise when she acts as a guide to the AI microcourses. (Hailey is called an agent in Google Dialogflow terms).

You can see that there are 15 intents, including what is called the Default Fallback intent (what she says when she doesn‘t recognise anything, for example I didn't get that. Can you say it again?).

There is also the Default Welcome Intent - how Hailey will start a conversation when you say something like Hi! or Gday. She might say:

Good morning. Or maybe afternoon, depending where you are.

Time really doesn't matter to me, although since I'm run by a computer, time is very important. It's just that my time is different to yours.

This is just one thing you might find strange, but there are lots of things where we are alike.

Why not ask me something about A I, or about what you can learn.

You can even ask me to tell you a joke.

You can go ask her and see if you can get her to say this.

Let’s look in detail at one of the intents, the ethics intent.

Google Dialogflow lets you create versions of the intent to ensure that when the person types something the same or similar, it will correctly recognise what was intended.

But in addition, it also builds an AI model of the agent’s conversations by doing training: it tries to learn from the training phrases you gave it to figure out if it can work out other variations. This is because you can’t specify every variation of a question that a person might ask.

Also, Dialogflow will allow you to import conversations between people which have been written down. It can use these as training material to learn from.

Your written down example of a conversation simulating CourseBot in the CONVERSATION DESIGN microcourse could be used like this.

RESPONSES

Next let’s look at the responses to the ethics intent.

You can see here we have provided Hailey with one response which talks about what you will learn in the AI ETHICS microcourse.

We could have provided alternatives which gave different answers, or variations of the same answer.

This is very useful when we want to see which responses are the most effective.

DESIGNING CONVERSATIONS USING GOOGLE DIALOGFLOW

Using Google Dialogflow is pretty simple.

It’s a matter of creating intents and responses that allow chatbots (and digital people like Hailey) to talk naturally. There are lots of chatbot building tools you can experiment with.

But as we saw in the CONVERSATION DESIGN microcourse, the real skill is making conversations appear natural.

Google Dialogflow, like many other tools, provides some clever things to help.

For example, if your bot is one people interact with to buy clothes, it might make sense to design parts of the conversation for that specific task. So if people will be asking about sizes, Dialogflow’s entities can deal with it.

Here Dialogflow lets you create a size entity to define all words associated with sizes that the users might type - and all the variations (or synonyms - words with the same meanings).

This means that when someone types extra small, it will be understood as S in a size chart, so you order the right thing.

CONVERSATION DESIGN: ART AND SCIENCE

But even with those tools, conversation design is as much art as it is science.

It needs you to think carefully, not just about the technology, but about how people will react to it and use it.

This is also a subject we’ll talk about in the AI ETHICS microcourse - why sometimes it’s better to think about people, not only technology.

You’ll be able to try out designing different conversations - either as chatbots or for digital people - in other microcourses.

SOME QUESTIONS TO CONSIDER

How many intents do you think you would need to recognise a question like What’s the weather?

Are there intents that you would find it hard to write, and why?

What might be the best kinds of conversations you could use to train Google Dialogflow?

AI MICROCOURSES HOME