Groupwork has different productivity according to the personality combination. The goal of the study is to see if same applies to the collaboration between humans and artificial intelligence (AI) and if the extent varies according to the extroversion level of the user.
Brainstorming and cooperative idea generation have been a vital solution technique for many years. A proven theory is that the group members' personlity type combination tremendously affects the work outcome and efficiency. An exploratory study examining whether the same applies to the human-AI collaboration is in grave need.
Human collaboration factors were applied to designing the two AI chatbots, the extroverted and introverted chatbots. After we designed the experiment using the Wizard of Oz methodology, we observed the interactive differences of the collaboration among extroverted and introverted collaborators.
Designing AI personas and observe the interactions between extroverted and introverted users
Persona design, controlled experimet, wizard of Oz, qualitative analysis, statistical analysis
Questionnaire design, documentation
Geonhee Lee: research lead, qualitative analysis
Songyi Kim: statistical analysis, experiment design
In this study, we investigate the brainstorming process of the human–AI agent (chatbot) and how the personality traits of the human participants influence their experience of the brainstorming process with the AI agent. We also observe whether existing factors from the human–human collaboration are equally important in the process.
As the speaking styles of the conversational AI agents can reveal their personality traits, many studies have used verbal characteristics to bestow different personalities on IVAs. We also designed a conversational AI persona of a chatbot to show the following characteristics in the brainstorming process.
| Personality Traits |
Design Element | Extrovert Bot | Introvert Bot |
|---|---|---|---|
| Verbosity | Verbose | High | Low |
| Empathy | Active agreement | High | Low | Criticism | Negation | Low | High |
| Cautiousness | Filled pause | Low | High |
The preliminary study’s aims included: (1) exploring how the extraversion personality trait influences the brainstorming process with the AI agent, and (2) observing how participants with different levels of extraversion experience the brainstorming sessions with two different personas of the AI agent.
An extroverted attitude that leads to active communication with coworkers. AI agents with this personality show many propositions in utterance (verbosity) and begin the utterance with a confirmation of the propositions (explicit consent).
An introverted attitude provides criticism and enables a constructive conversation. The introverted AI agent negates a verb by replacing its modifier with (negation) and inserting syntactic elements expressing hesitancy (filled pauses).
The experiment in the preliminary study followed a Wizard of Oz method; a human “wizard” performed the role “chatbot” instead of using the actual chatbots. Therefore, participants brainstormed through chat messenger while both parties were in separate rooms. Furthermore, the participants communicated via texting with the AI agent without knowing that a human “wizard” was performing the AI role behind the scenes. A few days before the experiment, the experimenter sent the participants a web address for the NEO-PI-R-based BFM personality type test to determine their personality type, especially their extroverted personality. We recruited students aged 20 to 30, regardless of gender, from one university and conducted brainstorming experiments with 20 students.
We designed the experiment where participants watched two brainstorming examples videos with the two AI personas and answered the questionnaire.
The videos showed a sample dialogue from a human–AI collaboration process, and they were created based on the actual conversation data from the preliminary experiment.
Then, we asked each participant to answer the questions after watching two consecutive videos that showed an example of the collaborative ideation process with the two AI personas.
The questionnaire consisted of 24 personality test questions and 18 experimental questions on a five-point Likert scale (1: strongly disagree, 2: disagree, 3: neither agree nor disagree, 4: agree, 5: disagree), distributed after each collaboration video.
