Designing for conversation

As designers we spend a lot of our time focused on visuals and UI patterns which have huge importance in the work we do today. The thing is the world is changing, and it’s changing fast. AI is not new but its application to modern life has finally arrived. It is no longer science fiction, but our reality. How do we react as designers and incorporate AI into our work?

I finished my PhD in Artificial Intelligence and Human Computer Interaction about 6 years ago. I explored the space between machines using natural language with humans. I’m a designer with a decent engineering brain so I coded everything myself and also did the research around how people interact with conversation. Just as we need to understand html, css, how databases work, etc…to be good designers, as we work with AI, we need to understand that technology as well.

I needed to deeply explore a few new areas for me around:

  • Machine learning: a type of artificial intelligence (AI) that provides computers with the ability to learn without being explicitly programmed.
  • Cognitive linguistics: the study of the meaning behind language and its grammatical form, or how we derive meaning from words.
  • Psycholinguistics: the relationships between linguistic behaviour and psychological processes, including the process of language acquisition, the process we go through to learn all language as humans. If you want to get a machine to learn language, you need to know where to start.
  • Construction grammar: this perspective groups a number of different grammars (the framework of a language) that all subscribe to the idea that knowledge of a language is based on a collection of “form and function pairings”.

I designed a few different concepts around conversation from the ground up and my main constraint was that the machine needed to make its own decision about what to say and how to say it. I didn’t want to use a database that worked to find the closest match to a pattern. I wanted the system to understand the words, the constructs and the intent. I also wanted it to understand tone, as communication is not only what you say but how you say it.

Through this process, I was able to understand some important concepts and to open up to a few new ideas. Although my own system was somewhat limited, it still worked well enough to come 4th in the Loebner prize, the yearly Turing test which evaluates the machine’s ability to exhibit intelligent behaviour.

Machines take me by surprise with great frequency.” (Alan Turning — Godfather of AI)

Why we do it

I like to think that we make machines to better understand ourselves. If I made the machine, is it smart or am I? We’re busy designing machines in our image so that we can more easily connect to them. We often like to have the machine embodied like in the movie “Her”, and we’re also experimenting with it disembodied like in Alexa, even though she has a humanlike voice. All of this is in an effort to better connect the machine with people.

What’s a “smart” machine…

The user interface is a way to visually communicate the input-output that occurs in a system. Instead of designing user interfaces we can design conversational experiences as a way to connect. The way we speak, the nuances, culture, intonation,… are all unique to each of us and important as they introduce complexity when designing these experiences. It’s easy to design a system that mimics, it’s hard to design a system that learns and adapts. In order for a machine to adapt to us and become really useful, we need to invest in our relationship with it. Simply put, the more you interact with it, the more accurate it will become. The same way that the more you get to know someone, the better you will work together.

I learned tonnes about this topic during my research, enough to fill a huge book, but here are the three main things that I think designers would find useful:

1. We connect with similarity

I tested the idea that people respond better when they are mimicked. We do, we relax and feel comfortable when the other person uses the same words, similar sentence length, sentence structure, and mimicked our tone. It creates a sense of familiarity. When designing conversation, it’s important to keep this as a foundational design principle.

We do something similar in our UI design, we try and replicate known patterns and keep things familiar enough for users to get oriented.

In conversation, this is the same but deeper and more complex. Every person perceives language based off of their experiences through life. The system needs to identify this and adapt on the fly, as it becomes familiar to the user. Designing for conversation allows us to build a unique and intimate relationship with people.

2. We like to feel important

Most designers know to practice user-centered design. We now need to move beyond this to engage in a relationship with people. Conversation is about connection, and without it there is an awkward and laboured interaction. The same is really true for user interface design as well, but we haven’t yet moved towards establishing deeper connections. We typically design for task completion first and foremost. With conversation there is no real end or beginning to the interaction, it’s a lot less linear. With conversation you have to build a connection otherwise the person disengages. This is what we do in real life when we engage in small talk with our colleagues before a meeting for example. Our behaviour doesn’t change with a machine. Not yet.

3. We react emotionally

As humans, when we process information there is an emotional response, it is how we interpret the world around us. When we design for emotion we make sure things look good and feel good so that our users have a positive experience. There are emotional moments built into the best user interfaces, a moment to smile, feel relief, feel like a god(dess).

The same is true in conversation, yet it’s subtler. You can’t rely on a cute animation to do the job for you. You have to truly invest in being present and attentive. Fortunately machines are excellent at this while we are not. We often design user interfaces for the best case scenario which means that our design is less adaptable to edge cases. When designing a conversation there can never be a dead end, it needs to be meaningful regardless of the users input. We introduce a lot more unpredictability in a conversation than you would have in a user interface. This introduces new design challenges around making the person feel connected and successful.

Why it’s so exciting

If you’re not convinced yet, designing without a user interface is a fascinating thought exercise. When designing, start with the conversation and see how it evolves without jumping to building out a user interface. Most tasks that you are designing for can be accomplished with a conversation and provide us a better way to connect with people. Additionally, just as it has been important as designers to understand how backend / frontend code works to be able to effectively design, the material we design with is changing. It’s time to explore and become informed on how AI systems are built and how they work, so you can unlock the possibilities we have as designers in this space.

The future is finally here, and you’ll love all the new challenges we’re going to get work work on C:.