by Dario Nardi
Have you played the Sims™ or enjoyed other computer games with virtual characters you can interact with? I am using virtual characters to explore personality. Mainly, I observe college students talking to and programming two characters named Truman and Olivia. This is a novel approach to exploring personality. The programmer sets up a character’s personality, then observes how others interact with the character. There are plenty of type-related patterns that match up.
Conversational Basics
The interactive character, or “bot”, has some “intelligent functions” When we say “hello” at the start of conversation we expect a return greeting. A “hello!” in the middle of conversation means something else! Other intelligent functions include:
– doing math (what is 178 plus 146?)
– knowing the time (is it morning of afternoon?)
– defining common things (what is a horse?)
We also hope people remember what we say, speak up when they don’t know something, make intelligent inferences, and have their own identity and interests! Sample Dialog:
input: what is my name
bot: I don’t know your name
input: it is Dario
bot: hello Dario, nice to meet you. Thank you for telling me your name
input: am I male or female
bot: your name is Dario, so I infer you are male
input: very good, what is your name
bot: my name is Truman
Personality Patterns
I quickly saw the value of personality type to understand the range of interaction between the students and characters. Here are some observations.
Gender: Gender often shows in the first minutes. Males tend to mention sex or tease the character. Females tend to mention people (their sister, boyfriend, etc) and console the character.
Cognitive Processes: To date I’ve identified a few ways people engage characters in conversation that match various cognitive processes.
Extraverted Feeling: People using Fe prefer easing into a conversation with the character, opening with greetings, using cues to change topics, and forgiving software bugs. The focus is having a nice experience. I learned to add more conversational skills about people, relationships and values!
Introverted Thinking: People using Ti prefer questioning the character and testing for weaknesses. They focus on figuring out the software’s limits. I learned the characters should respond to missed social cues, criticism and endless questioning. If the person is too rude (all Ti, nothing else) the character shuts down its own program!
Extraverted Intuiting: People using Ne prefer juggling many conversational threads at once, with each statement referring to several ideas with little need for a goal. Vague responses, light sarcasm and liberal use of metaphor are welcome. I learned to allow the character to give multiple responses at once and keep “opening up” the conversation.
Introverted Sensing: People using Si prefer focusing on one thing at a time and expect the conversation to have a purpose typical of known software, such as a help or search system. They dislike non-literal non-specific responses. I learned the character should state an agenda and do a series of questions or other task such as launch programs.
Temperament: I’ve noticed temperament patterns. Rationals ask conditional questions such as, “if a cow is an animal and an animal is alive then what?” They explore logical inferences: “So I just said X, what does that mean?” Rationals want to see the character “think”, and like evaluating it capabilities. In contrast, Idealists tend to make values statements such as, “how can happiness mean success?” They enjoy when the character learns something unique about them or from them, or better yet, when it comes up with interesting insights. They evaluate the character based on meaning that comes from interaction.
Non-Verbal: When asked to describe the characters’ personalities, non-verbal cues seem to have a huge effect. This suggests that we can improve our type-watching skills with conscious attention to extraverted Sensing, to literally notice the non-verbal cues in addition to abstract interactional patterns.
Feel free to explore further at www.socialbot.com. In the future, I hope characters will notice type patterns and adapt accordingly.
