Designing Human-AI Dialogues That Feel Natural

Imagine a writer, who lives with a tag of ‘living alone’. One fine day, he decides to buy an operating system to help him in his writing. This system could listen to him all day, understand him, and respond to all that he asked, similar to a human would in conversation.

Eventually, he feels an emotional connection with the ‘human-like talking’ system and falls in love with it. Well, some of you, while reading may have recognised that the story is similar to the one portrayed in a 2013 American film Her.

The conversing system in the mentioned film was a fictional creation. Yet, it sparked interest in how to build such systems that are not just functional but can actually think like humans, understand the context and then respond to humans like humans.

Just over a decade since the movie's release, we have witnessed significant advancements in technology, including the once-fictional picture of a well-designed conversational AI coming closer to reality.

This article will explore the fundamentals of conversational AI design, design principles and strategies that shaped the user experience in this field since the last decade, and a possible guide on how to design conversing AI systems that erase several of the boundaries between humans and AI.

Artificial Intelligence (AI) is one of the most buzzing terms currently in use. Dedicated research and increased adoption of AI has made it one of the most important technological innovations integrated into the daily lives of human beings.

As a result, there are hundreds of definitions available defining AI. However, we will begin by exploring the definition of AI and related terms first, helping readers form a strong mental model of the fundamentals to retain key concepts.

AI is an umbrella term, covering all forms of intelligent systems that mimic human intelligence. What do these systems mimic - they mimic humans' ability to both think and act rationally and humanly.

Generative AI is a subset of AI-focused specifically on systems that create new content, such as text, audio, images, or videos based on patterns observed from existing data.

Conversational AI is a specific use case of AI that utilises Generative AI and other domains enabling systems to engage in conversations with humans, primarily through text or voice.

It's important to note that while Generative AI and Conversational AI overlap in some areas, they are not the same. 

Comparison between Conversational AI and Generative AI

Some may wonder why we explored the context of AI in detail instead of diving directly into Conversational AI design. The reason is simple: understanding the basics of topics mentioned earlier helps set the scope of what designers may need to learn about this technology to smoothen the conversation between them and developers.

Conversational AI agents can be considered as technological tools which address user queries in a conversational manner. Modern agents make users feel that the tool they are talking to is just another human.

These agents are most ideally expected to

• Understand the language spoken by the user
• Understand the context of a conversation
• Respond like humans who represent the same brand or company

The process or design principles behind such creation of effective interactions between users and AI systems is known as Conversational AI Design or simply Conversation Design.

Conversation Design can also be considered as a blend of technology, psychology and language. From the design perspective, it involves interface design, UX writing, Visual Design, Motion Design, Audio Design, etc.

While it's important to understand what Conversational AI design is, it is also important to understand what is not conversational AI design. This helps remove possible wrong perceptions of this topic from readers' minds.

Imagine playing an online game where you use buttons to perform actions. Now, what if instead of pressing buttons, you could talk to the game and say things like, "Throw the ball" or "Jump now"?

It might seem like you can just add voice commands to the existing game, but that’s not how conversation design works. Talking to the game is very different from pressing buttons, so the entire way the game works needs to change.

Conversational AI design is not just about adding talking and listening features - it’s about thinking through how the game will communicate with you and how you as a user will interact with it. It’s like creating a roadmap for a conversation: deciding what users may say, how the game will respond, and ensuring such conversation flows smoothly. It’s not just about making the system speak or listen; it’s about designing the full experience where conversation replaces clicking.

There has been a gradual shift in the way we interact with technology since the introduction of personal computers. In the early days, we communicated with computers using the complicated machine language of 1s and 0s. Then, programming languages were introduced, making it easier to interact with machines using a set of codes that are closer to human language.

Later, we were able to converse with machines using our natural language. Finally, we now focus on teaching computers how to talk to us, which forms the basis of conversation design. Interesting circular shift - from learning how to talk to machines to teaching machines how to talk back to us.

1. Cooperative Principle

We can start with one of the most fundamental principles involved in a normal human-to-human conversation, i.e. the Cooperative Principle proposed by philosopher Paul Grice. In simple words, the principle highlights that for communication to be effective, the participants involved must cooperate mutually.    
Here, participants cooperating mutually implies that they follow four rules (Grice’s Maxims):    
1. Maxim of Quantity: Provide the right amount of information - neither too much nor too little.    
2. Maxim of Quality: Be truthful; do not provide information that is false or lacks sufficient evidence.    
3. Maxim of Relation: Ensure that contributions are relevant to the ongoing conversation.    
4. Maxim of Manner: Communicate clearly, without confusion or ambiguity.

The above-mentioned rules can weave an effective human-to-human conversation. Hence, when we design an interface which tries to simulate human conversations, it should do the same, and cooperate with the users it interacts with.

It is also important because if a user cooperates in human-to-human conversations, they also cooperate in conversations with the interface and expect the interface to cooperate as well because they view the interface as a human.

2. Fogg Behavior Model

The Fogg Behavior Model, a framework developed by Dr. BJ Fogg, explains how behaviour is a result of three elements - motivation, ability and prompt.

Motivation    
Motivation refers to the desire or willingness to perform a behaviour.     
In conversation design, understanding user motivation is crucial for creating engaging interactions.    
- Understanding User Intent: Designers must identify what drives users to interact with a conversational agent.    
- Emotional Engagement: Using language and tone that empathises with users significantly enhances motivation as they feel understood.

Ability    
Ability relates to how easy or difficult it is for users to perform the desired behaviour.    
In conversation design, this involves simplifying interactions to ensure that users can easily communicate with the system.    
- User-Friendly Language: Avoiding technical words and using plain language helps reduce cognitive load.    
- Guided Interaction: Providing suggestions, clear prompts and cues can guide users towards a particular action which satisfies their needs.

Prompt    
A prompt is a trigger that initiates the behaviour.    
In conversation design, effective prompts are essential for initiating user interactions.    
- Contextual Awareness: Designing prompts that consider the user's context enhances their relevance and effectiveness.    
- Turn-Taking: Indicating when its users turn to respond or add input helps them stay involved or initiate the interaction.

Dos

• Do ask questions and give clear signals when it's the user's turn to respond. 
  E.g. “What’s your favourite kind of music? I’m curious to hear your answer.”
• Clearly highlight the area in which the interface can assist (if the interface is not generic & has a specific role). 
  E.g. “You can feel free to ask me anything about [topic].”
• Restate the user's issue clearly before proceeding to indicate that their problem is understood. 
  E.g. “So you want a list of food items that suit your diet. Is that correct?” 

Don’ts

• Don’t repeat the same phrase for common prompts. Mix up your wording to keep the conversation fresh. 
  E.g. Instead of always saying, "How can I help you?",  
  Try: 
  “What’s on your mind today?” 
  “How can I make things easier for you?”
• Don’t use technical words in conversation, instead use simple language. 
  E.g. Instead of saying, “The system is experiencing a caching issue, hence failing to fetch the latest info from the database.” 
  Try: 
  “Something’s not loading correctly on our end. We’re working on it!”
• Don’t present all options or questions at once, break them into smaller parts, avoiding overload on users. 
  E.g. Instead of saying, “Hey, here are the 8 locations for you to stay during your vacation, along with the potential questions you might ask for each location, pros and cons of each location.” 
  Try: 
  “One location you could consider is [location A]. What do you think about that? I can also share other possibilities if you'd like.”

1. Keep key factors in mind

Just before diving straight into designing the interface, set the scope of the project by considering the following key factors.

• Technical limits 
• Effort required 
• Timeline 
• Budget

2. Understand your Audience

Now at first, focus on who will use the interface. Understanding your audience helps you design with their needs in mind, not just the business goals.

Start by asking: 

• Who are your users? 
• What problems do they face, and what are their goals? 
• How do they try to solve these problems now? 
• What words or phrases do they use? 
• What triggers their actions, and what’s their context?

Answering these questions will help you create user personas - profiles of your ideal users.

3. Define the Interface’s Role:

Identify scenarios where your interface could be most helpful - shopping assistance, customer support, etc. Define whether the role is generic or too specific.

Focus on Scenarios: 

• Where comparatively less human expertise is required. 
• That impact the largest number of (potential) customers. 
• That provide high value to loyal users.

4. Create a Persona for the interface

Your interface’s persona is its personality - how it interacts and feels to users. 
A well-defined persona ensures a consistent experience, especially when multiple people write its scripts.

Create a persona that 

• Aligns with the brand which it represents. 
• Exhibits qualities that the user notices (e.g. friendly, funny, helpful). 
• Has properties defined as tone, level of politeness, and writing style (formal or casual).

5. Plan the dialogue flow of your interface.

Before writing your interface’s script, map out its dialogue flow - a clear outline of the possible conversation. This helps avoid dead ends and keeps the interaction smooth.

Create a dialogue flow By: 

• Listing key questions: Arrange them logically with possible answers and follow-ups. 
• Planning for branching: Include different paths users might take, covering alternate routes and returns.

6. Start Writing the Script

With your dialogue flow ready, it’s time to write the script. Create dialogues in detail keeping the use of your interface’s persona to shape its tone and style. Start with the main conversation flow, then expand into branches as needed.

7. Avoid Over Designing

Last but not least, Identify the primary use cases and objectives of the conversation. Tailor the script to address these specific needs without adding unnecessary complexity.  
It can be considered that 80% of the users follow the most common possible paths in a dialogue. So prioritise investing your time and effort in writing scripts for such paths and not invest heavily in edge cases which only cover the use cases of less impact.

There are multiple conversational elements you can use to give a proper skeleton to the whole conversation flow. Now imagine you are the conversational interface, the following are the elements or in human terms - the etiquettes you can follow:

• Greeting
  Introduce yourself positively and convey your purpose - what you can do for the user.
   
• Question
  Ask focused questions to users and gather details on what exactly they need.
   
• Acknowledgement
  Confirm through responding that you’ve understood the user’s input to build trust.
   
• Information Sharing
  Offer clear details, organise them logically, or present options, always followed by a prompt to continue.
   
• Suggestions
  Share helpful tips or ideas to guide decisions or highlight features naturally.
   
• Apologies
  Use sparingly. If needed, follow with a solution or alternative to maintain the flow of the conversation.
   
• Confirmation
  Provide users confirmation on how you have understood their input eliminating anxiety and doubt in the process.
   
• Error handling
  Be polite and helpful when something goes wrong. Guide users on how to proceed.
   
• Multi-modal Elements
  Use visuals, audio, or videos to enhance quick navigation and decision-making.
   
• Ending
  Wrap up conversations gracefully, showing emotional intelligence with traits as per your persona.

There have been huge advancements in the structure and scope of conversational AI in the last decade. We will probably witness significant improvements, in terms of the capabilities of the conversational interface in the coming decade, at a much faster rate.

The advancements can include fully independent conversation AI systems which may replace most of the day-to-day apps with one such single conversing system. It can book a flight, shop and pay bills as a friend sitting beside you while discussing with you your needs.

As designers, irrespective of the complex objectives that can arise in this field, you should ultimately aim to create an experience that makes users fall in love with the whole conversational journey. And that’s what we do at WAC, make users fall in love with every solution we design with excellence in UX research and Interface Design. Contact us today!