When talking about ‘intelligent’ interfaces, you can imagine lots of things. To make it clear from the beginning: to me, the intelligence level of the chatbots out there is, in my eyes, not yet worthy of calling them ‘intelligent’. But let’s discuss this from scratch. What are conversational interfaces and why are they (or should they be) intelligent?
What Are Conversational Interfaces?
Definition
Conversational interfaces are systems that process natural language and find answers or execute actions (mostly making sense). They are, so to speak, ‘intelligent’ because they have the capability to ‘learn’ from their interactions with humans. In theory. These systems are called Natural Language Processing (NLP) systems.
Chatbots can also talk
Conversational interfaces are not only limited to interacting with humans via written language like chatbots do, but also via spoken language, as, for example, Siri (from Apple) or Alexa (from Amazon). Conversational voice interfaces actually speak and talk back, simulating human language. Some synonyms for these are: chatbots, virtual agents or conversational bots.
In the market, several platforms offer these kinds of specialised conversational interfaces for multiple purposes. Some of them are: Dialogflow by Google, Lex by Amazon, Azure Cognitive Services by Microsoft or Watson by IBM. These platforms provide powerful natural language processing systems.
In this article, we will have a closer look at the basics to design or, better, to architect a conversational interface that should be a bit more ‘intelligent’ (I emphasise: a ‘bit’ more intelligent).
Speaking of Intelligence
At this point, another question arises: When is a chatbot intelligent? Or what makes a chatbot intelligent? The first time we began to experiment with an NLP system, there was the certainty and, at the same time, the disappointment that the thing couldn’t really do something without configuring it to do it. In other words, an NLP system needs lots of configuration and tuning before it can answer with something that makes sense. That’s why it’s difficult to make a chatbot show more ‘intelligence’.
An NLP system needs lots of configuration and tuning before it can answer with something that makes sense. That’s why it’s difficult to make a chatbot show more ‘intelligence’.
Graciela Schütz
Senior UX Architect, Unic
Designing Conversational Interfaces
As I already wrote above, primarily, a conversational interface is a system that creates (hopefully meaningful) interaction in natural language with a human. To do so, the NLP system needs to be configured. The configuration implies a database with phrases and their flows.
Emotions – Consciously Shaping Personality
We humans are sentient beings and as such, emotions are triggered in one way or another when interacting with a system like this. Because a conversational interface simulates natural spoken language, which seems to be real. Our emotions can range from positive to negative and all emotions in between. That’s why it is very important to pay close attention to how we shape the character of a conversational interface. We want to be sure to provide a positive experience for the customer or user. In a way, it feels like one is talking to someone behind the screen.
Creating a Persona
To design a conversational interface or shape its character, it is very important to create a persona. Personas are fictional ideas of a character, which help us to give a face and ‘personality’ to our conversational interface. It’s easier to imagine how the conversation with our conversational interface should be going, if we have a ‘picture’ of it in front of us:
Having a clear idea of goals and skills
I believe that setting the goals and abilities of a conversational interface is crucial for making it more intelligent. The chatbot should have a clear mission and be able to respond with clarity as to whether it can be of help or not. This delimitation is vital when the conversational interface is asked to do things that it just wasn’t trained for. And it can properly react to the situation.
What should the chatbot help with?
Another crucial point it is choosing what the chatbot should be helpful with. There are many chatbots out there that were designed to help with things that are of little value for their users. For example, when asking a chatbot for any kind of information and the only thing it comes up with is a link to a page, which you can easily find yourself just by googling the same thing. That doesn’t make a chatbot worthwhile or intelligent. So, what is it worthwhile for a chatbot to do?Listening to Emily Sappington in the Front Conference (2021) at Zurich, she mentioned that automation and efficiency is where the value of Artificial Intelligence (AI) is. She said, “AI should make us feel like superheroes(!!!)”
And how do we ensure that? In a wrap-up, she gave us the following advices:
Be honest about what you can and can’t do
Set expectations appropriately
Get scenario-focused
Try to be useful and helpful
Have some fun
Later, I’ll come back to that. First, it’s important to have in mind the elements a conversational interface is composed of.
AI should make us feel like superheroes(!!!)
Emily Sappington
Director of Product, Twilio
Understanding the Information Architecture of a Conversational Interface
After thinking of what a conversational interface should do and how its character should be, I need to talk about how the information architecture of a conversational interface looks. NLP systems, being software, work by identifying words in the sentences entered. Those words are broken down into the elements of a conversation. The elements of a conversation with an NLP system are: utterances, intents, entities.
Utterances
Utterances are simply expressions, from humans or the bot itself. Utterances are expressions in spoken language, which we find in any dialogue, as we normally speak. For example: “It's sunny outside” or “I feel hungry” or “I need help finding my security number”.
Intents
Then there are intents. In an ordinary conversation between humans, it’s very normal for a human to identify what type of conversation is taking place. A human can identify whether the other person is making small talk or asking for something or talking about a specialised theme. A conversational interface needs to identify what the user or customer wants or is trying to express. Identifying what actions a customer would like to be executed, being a question or a task, are called the ‘intents’.
For example, in the expression: “How are the stock markets today?”, the intent is simply an inquiry about how today’s stock markets are doing. But in the expression “The stock markets are rising today”, the intent is rather a declaration that today the stock markets are rising.
When the NLP system identifies the intent, these intents are matched with actions and so the action is triggered. In this way, the system responds appropriately, displaying today’s stock market achievements, as an example. Or replying “Swiss Market Index 3 points more than yesterday, Dow Jones 2 points more than yesterday. Yes, stock markets are rising indeed”.
Entities
The entities are pieces of information that are important, to know what the bot should search for in the backend system. Who, what, when and where are the crucial questions here. In the expression “How are the stock markets today?”, the entities are ‘stock markets’ and ‘today’. The system then would search for that.
A chatbot needs to keep up with the conversation when it has answers. And when it doesn’t, just apologise and then keep quiet.
Graciela Schütz
Senior UX Architect, Unic
It’s All About Conversation
So, the final act is the conversation. A chatbot needs to keep up with the conversation when it has answers. And when it doesn’t, just apologise and then keep quiet.
Breaking down what Emily Sappington said, I would give this practical advice:
Delimit the chatbot abilities to the cases it can be really helpful (and not being only a search engine replacement) and communicate clearly what the chatbot can do.
Don't bother to create conversational flows, where you can't keep the conversation flowing. For example the user asks something unexpected, and the bot has no answer for it. I would suggest to analyze the flow of the conversation, to identify at what point the user got the impression the chatbot could be of help.
Try to come up with cases where automation and efficiency are the main goals.
And inject some fun! Let the user smile when interacting with the chatbot.
Conclusion
NLP systems are very helpful and powerful. However, the intelligence of a chatbot should not be taken for granted. The intelligence of a chatbot needs to be designed, configured and trained. The delimitations of a chatbot are crucial for its success as well as the value the chatbot delivers.
Sources
Google Cloud Training, “Contact Center AI: Conversational Design Fundamentals”
Raluca Budiu, nngroup.com. "The User Experience of Chatbots"
Josh Lovejoy, “When are we going to start designing AI with purpose?”
Contact for your Digital Solution with Unic
Book an appointmentAre you keen too discuss your digital tasks with us? We would be happy to exchange ideas with you.
Contact for your Digital Solution
Book an appointmentAre you keen to talk about your next project? We will be happy exchange ideas with you.
