You are designing a social robot interaction. You are fighting against the monotone intonation of speech synthesis, against latency in the speech recognition and against the very narrow conversational path you have to lead your users down because “real” AI does not exist yet. How can you avoid your interaction being experienced as slow and flat? How do you make it come to life?

One major part is to figure out who your main interactive agent is. It is easy to make this field blank. You might not even have thought about it, it is plainly “a social robot”. Nothing more.

If your robot is a 2-dimensional, undefined character - a conversation with it will feel flat and uninteresting

Imagine having a conversation with a human who has done nothing, seen nothing, wants nothing and has no particular views on anything. About as fun as having to rugby tackle an angry, drunken football player who has four friends and a running chainsaw.

During a conversation we are collecting everything someone is saying and slowly forming a coherent understanding about who they are. An image of our interests, life experience, what we care about, what we are thinking about is slowly revealed pixel by pixel during interaction. Like downloading a jpg on an early 90’s modem. Of course we often interpret wrong so the image might not be accurate. Again like a jpg on an early 90’s modem. But we are creating a picture.

If someone mentions horseback riding and then ten minutes later makes a comparison to “when you need to slow down a horse” you start consciously or subconsciously to make some assumptions about them. This person rides horses, or at least has an interest. This can lead to more questions being raised in your mind.

Are they from the countryside, did they grow up with horses around? Do they currently have a horse? How do you even deal with a horse if you live in the city? All kinds of thoughts are triggered by seeing some apparent pattern in conversation. What does it mean when there are questions you would maybe like to ask? The interaction is interesting.

For this to happen there must be a coherent person hiding inside who you are talking to. There must exist actual patterns that can shine through during the interaction to create depth and texture.

There must be character

Authors, script writers and actors think about characters and how to use them all the time. In my work on stage most of my creativity is driven by decisions about who I am on stage that were made a long time ago.

Here are 5 steps towards defining a character that I use in my work.

1. Understand that character means limiting options

Giving your robot a character is to limit what it can do, say, and how it can say it. Why would you want to introduce limitations? What is the purpose of thinking inside a box?

Creativity craves constraints

Having constraints makes it so much easier to use your imagination. If I gave you some paper, a pen and asked you to write a story, what would you write? Not so easy is it? The task is wide and undefined. And an undefined problem has an infinite number of solutions. Division by zero is not for me.

If I instead ask you to write a four page long detective story that takes place in a New England forest, what would you write? Think about it for 10 seconds. I bet you already have some ideas. You now have walls to bounce your imagination against.

2. Just make decisions and stick to them

Decisions about traits are what defines character. There are almost no bad decisions. Or at least:

Almost any decision is better than making no decisions at all

Just put time and effort into making your choices work until they do. I have a magician friend who once made the decision to dress in a dragon outfit, be his normal grumpy self and do card tricks pretending to be a magic dragon. On the surface I would say that is not a very good decision. Today he has a really great and successful Vegas show.

3. Describe with 3 words

What are these decisions I speak of? Start by choosing the whole character, or any aspect of it and write down three words to define it. If you think this seems difficult, pointless and bothersome, pick any well loved TV or movie character and see how easy it is to describe them with three words (sitcom characters are especially good for this).

George Costanza – hotheaded, overconfident, self-loathing

Hermione Granger – good, know-it-all, fear-of-failure

Homer Simpson – lazy, dim-witted, family

The idea is not that you have to capture the entire character in three words. Just to establish some basic directions to walk in. You must not always walk with these words, but do not walk against them. The point of decisions is to try to stick with them (are you reading this Dennis?).

4. Flaws are good

When you decide on words, it is good if at least one of the words is something that is a trait that a character would try to hide. A word that is outright or somewhat negative. Vanity, short temper, repetitiveness, overconfidence, repetitiveness, things like that. Why?

Perfect is the enemy of charming

Think of the most famous and well-loved sitcom characters. Homer Simpson. Elaine and Kramer from Seinfeld. Sheldon Cooper. Leslie, April and Ron from Parks and Recreation. They are ALL in some way defined by their weaknesses and faults. Often their flaws are their main defining attributes.

Think of the best conversations you ever had. Were they with people who had no apparent faults or flaws?

Weaknesses make a character more interesting, human, funny and authentic. There is something about showing traits that anyone would try to hide that makes the whole package feel authentic. When something is impeccable it is often too good to be true. You should feel an inclination to read the fine print.

5. Expand the inner world


5 steps to making robot interactions stronger Photo by Dominik Scythe on Unsplash

Go further and try to make some more decisions on various aspects of the character:

  • Motivation:
    Is there anything they want to achieve, in this interaction or in “life?
    This one is really important. Maybe one of very few things that define intelligence in general is that all intelligent things have some sort of goals (again, are you reading this Dennis).
  • Interests:
    What are their hangups, what things would they get stuck on Wikipedia reading about.
    These things will shine through in conversation. If you have made a decision that your character has an interest in the Roman Empire, this will impact the dialogue options you write. References to Caesar or quotes from Cicero will start to sneak into what they are saying. An iceberg will start to form under the tip and people will sense that there are dots to connect, that there is depth.
  • How does the character respond to:
    Being agreed with
    Being disagreed with
    Personal questions
  • Strengths
  • Weaknesses
  • How did they learn what they know about the world?
  • Friends and connections:
    How many times in actual conversation do you find yourself saying “I have a friend X who just…”?
    If you make a random decision that your character has a deadbeat friend called Dennis this might even sneak it’s way into a short blog post.

The more you define the character the easier it gets to write coherent lines for it to say. And a feeling of substance appears in the interaction. There are suddenly underlying reasons for what is said and done and even if people cannot see them, they can feel it (even if these reasons are fictive in the first place). It starts to get interesting.

To write great dialogue you have to know who will speak it. There has to exist a who.

Most of these things are my personal observations from performing, writing and working with robots. I know these ideas have worked out very well for me. How do you think they could impact your work? I am very interested in hearing your thoughts and would love to discuss this more in the comments (especially from you… Dennis).