ProNoBot, a Conversational Bot for Understanding Gender and Pronouns
Designing a conversational agent for tension
Introduction
For my Human-Computer Interaction (HCI) course at Emory University, my group and I were tasked with designing and prototyping a bot using Flow XO. However, this bot isn’t any ordinary conversational bot. Instead, we had to design for tension, for uncomfort, for controversy, thus we decided to build a conversation around gender identity and pronouns. Through our bot, users can reflect on their ideas about gender and learn new information in the process. In this post, I discuss our design process, including user testing and implementation, and explain limitations of our project. Readers can visit our chat bot or check out our demo below.
Motivation
Given people’s reluctance toward progressiveness, change, and unfamiliar topics, discussions about gender identity and LGBT+ topics elicit a lot of tension and controversy in many conversations.
For example, the Florida Senate recently signed into law the “Don’t Say Gay” bill, which shows the extent and seriousness of this topic.
Another example, anti-Trans violence has been pervasive in the past few years, revealing more of people’s negative and aggressive behaviors toward gender fluidity and identity exploration.
Tension
Many people only believe in a gender binary rather than a fluidity. Many people may not be familiar with other genders and identities. There are a lot of cases in which people are not malicious or prejudicial but are simply are not informed or knowledged about these topics. Whichever intent, our bot aims at deconstructing and reconstructing what a user knows about gender identity and expression through guided reflection and provided information.
Design
For this design sprint, our group consisted of 4 members: Randy (me), Jesenia, Sophie, and Ruochen.
I. Initial Ideas and Brainstorming
We started brainstorming by thinking about what makes us uncomfortable. So much has changed in the 21st century that one would expect more progressive outlooks, but this is unfortunately not the case. Recent years have shown a lot of resistance by all types of people and groups.
Some of the uncomfortable broad themes we highlighted:
- Politics
- Religion
- Gender
- Sexuality
- Violence
- Work/Research/Industry hazards
Some of the questions or remarks that came from our discussion about personal tensions:
- How do we know when someone is open to talking about certain topics?
- How will I react to a topic? How will other people react to a topic?
- Where or how will a conversation turn when a controversial topic is introduced?
- How do I challenge my own views and biases and truly understand opposition?
From our brainstorming session, we decided to work on gender and pronouns as this is still a new topic in current discourse. And personally although I am a progressive and respectful person, I am not knowledgeable on gender, especially with pronouns, so this will be a learning lesson for the research team as well. We wanted to cater our chat bot to people who were resistant toward “unconventional” gender identities, were unfamiliar with gender identities, or needed challenged perspectives; therefore, we aimed at creating a bot that gives the user space to reflect and provides good resources for learning more about gender.
Idea: a reflective and informational conversation about gender identity, gendered experiences, and pronoun importance
Target Audience(s): conservative people who reject gender fluidity, people unfamiliar with gendered topics
Tools: Figma, Google Docs, Flow XO
We also developed a few “how might we…” statements on conversing about gender identity and pronouns, as well as embedding organic interactions with the chat bot.
- How might we obtain people’s intricate views on gender?
- How might we create a space where people can express their confusion on gender?
- How might we create a space where everyone feels comfortable expressing their gender?
- How might we initially engage people who may be averse to having these conversations?
- How might we keep the ones not feel interesting during the chat?
- How might people come to a common understanding of gender?
- How might we diffuse an understanding of respect?
Our resources, ideas, and designs were maintained in a Google Docs and Figma.
II. Implementation
We designed and mapped the conversational flow in Figma and then implemented the conversation in Flow XO.
We brainstormed some conversations between our team members and carried out initial testing. Some examples are shown below.
Conversational Flow:
In Flow XO, our conversation initiates with introductions and greetings. The bot prompts the user for their name and age. This is done to help personalize the experience and add more human elements to it.
Next, the bot tries to establish some grounds of respect by asking if the user is comfortable sharing their pronouns. The user can branch into “yes” or “no” routes based on their response. Then depending on which branch the user enters, a conversation based on learning more about gender identity and resources or a conversation about one’s experiences with pronouns ensues. There are a few points in the conversation that merge to the same destination, giving users more freedom and not limiting them to the first branch they chose.
Some questions that we asked about gender:
- “How do you feel specifying your pronouns?”
- “How do you think we can create an environment in which people feel safe expressing their gender identity?”
Some resources that we referenced as well as shared with the users:
Our key modes of user input and interaction are choice selections and text responses. The following images illustrate what these look like in the Flow XO interface.
III. User Testing & Feedback
We tested our implementation with some classmates and people in our network. Users selected from these close vicinities are not ideal, but their insights were still interesting and valuable given tensions between cis and non-cis gendered discussions.
One user noted the following:
- asking for age is rude and irrelevant
- providing direct links in conversation seemed strange and inorganic
Another user validated that the age question was irrelevant by saying:
I am very young!
instead of providing an age.
Another user shared good experiences with the initial implementation. Some of their remarks are as follows:
“i like the website we’re prompted to check out”
“i also like how the questions sometimes had examples of other people’s responses”
“a productive conversation… did feel like it was short”
The user also wonders what the other conversational branches would have been like if they answered differently.
Another user also felt the conversations were short. Some of their feedback is as follows.
“it was shorter than i thought based on how I answered…”
“None of the questions or responses [felt] off”
likes… “[bot] prompts the user for lengthier responses bc it allowed me to reflect and gather my thoughts”
“Maybe u can provide more resources”
“try to vary the question types if able”
Reflecting on the user feedback and my interactions with these users, their insights are very valuable. Even if the feedback is not inherently negative or demeaning, the good feedback is also important for validating if the chat bot is useful, meaningful, or pleasurable to interact with. Fortunately, a lot of the users expressed interest in the bot. Some feedback helped remove unnecessary or irrelevant features and questions, such as asking for age when we do not really use it, and they facilitated thinking about other questions or approaches that we did not think about previously.
IV. Re-Implementation
After acquiring user feedback, we iterated and refined our implementation to include longer conversational branches and more organic interactions.
Some revisions we made for the iterated implementation:
- Remove age question
- Add more depth to the branches, extend the conversations
- Provide more opportunities for reflective freeform responses
- Offer hypothetical scenarios about gender and communication
- Emphasize and provide recaps of important information
- Integrate more human, organic responses and validators
Conversational Flow:
Some specific questions or remarks we shared:
- “Have you ever been in a situation in which you felt uncomfortable sharing your pronouns?”
- “Thanks for sharing. How do you think we can make people feel more comfortable?”
- “Let me know if you find this helpful.”
- “Okay, thanks for sticking with me so far. Do you mind if we recap a few things about gender?”
- “We have had a nice conversation so far. Is there another topic you would like to know about?”
V. Final Products
After testing and implementing, we created and demo’d our final products as follows:
Our chat bot.
Our demo video.
VI. Limitations
Given our team’s lack of expertise and experience in building chat bots and conversational agents, our bot had both technical and conversational limitations.
We created a bot that attempts to deconstruct and reconstruct the user’s knowledge about gender, identity, and respect.
Building a bot in Flow XO was quite tedious and limited in features, thus we couldn’t deliver personalized bot responses to a user’s unique responses or confusions. Since our bot focuses a lot on self reflection and guidance through these uncomfortable topics, it is difficult to recognize quality and subjective answers given the Flow XO platform. We also tried to streamline this subjectivity as much as possible through broad questioning, such as questions about example scenarios or about the user’s experiences. We also broke down sentiment-based questions through concrete choices — the user first identified how they felt about an idea, concept, or scenario, and then the bot encouraged them to share their ideas in freeform text. This approach is of course more robotic, but it is effective given the platform’s limitations.
Then we tried our best to user test, acquire feedback, and iterate, but we did not know where to find users in our target audience and identity LGBT+ individuals. Although our chat bot is about non-cisgendered people, they are not necessarily our target audience; still, it is valuable to include their insights in authentic and organic portrayals of the types of experiences, conversations, and interactions that they have.
Conclusion
Our chat bot effectively provided a space for people to challenge and reflect on their experiences or perceptions about gender identity, pronouns, and respect. Through interacting with the bot, users are guided through a series of questions based on their experiences instead of being presented with flat interview-like questions. This enhances the user experience and gives the chat bot a more human-like, realistic appeal. From designing the chat bot to user testing and iterating our implementation, I learned how important it is to acquire user feedback in design and development, especially when working with sensitive topics. Simply by iterating our chat bot again, we were able to recognize important missing details and enhance the user experience.
Our bot is a foundation for cultivating respect and empowerment for users and contributes to the bigger solution of understanding gender. Users must ultimately interact with people in the real world to refine the knowledge and insights learned from the bot and better connect with the lived experiences that non-cisgendered people have.
Acknowledgments
I would like to thank my group members, Jesenia, Sophie, and Ruochen, for contributing to this project.
