ImaginAItion
ImaginAItion
Topics:
Generative AI
ImaginAItion is a web-based game inspired by Telestrations and Caution Signs, designed to help young adults improve their skills in prompt engineering and refine their mental models of AI, particularly regarding biases. Our work is published in ACM CHI* 2025, and I serve as second author on the paper. I led research efforts to understand our target audience and drove iteration cycles to refine the game. Alongside another teammate, I also led the design process in Figma, ensuring the interface was intuitive and aligned with our objectives.
* the premier conference in Human-Computer Interaction
ImaginAItion is a web-based game inspired by Telestrations and Caution Signs, designed to help young adults improve their skills in prompt engineering and refine their mental models of AI, particularly regarding biases. Our work is published in ACM CHI* 2025, and I serve as second author on the paper. I led research efforts to understand our target audience and drove iteration cycles to refine the game. Alongside another teammate, I also led the design process in Figma, ensuring the interface was intuitive and aligned with our objectives.
*the premiere conference in Human-Computer Interaction
Roles:
Researcher, designer, developer
Duration:
Aug 2024 - Feb 2025
Tools:
Figma
JavaScript
HTML
CSS
CSS
The Problem
The Problem
With the increasing use of generative AI (GenAI) systems among young adults (aged 18-34), the need to foster socio-technical AI literacy is crucial.
With the increasing use of generative AI (GenAI) systems among young adults (aged 18-34), the need to foster socio-technical AI literacy is crucial.
Current Solutions
Current Solutions
Existing work in AI literacy fails to address young adults, who make up more than half (56.75%) of ChatGPT users.
Existing work in AI literacy fails to address young adults, who make up more than half (56.75%) of ChatGPT users.
Instead, existing has primarily focused on younger audiences, using game-based learning to simplify AI concepts and teach foundational principles. Tools like Color Conquest and TreasureIsland demonstrate the effectiveness of interactive simulations and gamified eBooks. However, there is a gap in addressing adult learners, who engage with AI in more complex contexts. For young adults, interactive explainers and gamified environments show promise in fostering experimentation, reflection, and awareness of AI biases, offering potential to teach procedural skills while deepening understanding of AI limitations and ethical considerations.
Instead, existing has primarily focused on younger audiences, using game-based learning to simplify AI concepts and teach foundational principles. Tools like Color Conquest and TreasureIsland demonstrate the effectiveness of interactive simulations and gamified eBooks. However, there is a gap in addressing adult learners, who engage with AI in more complex contexts. For young adults, interactive explainers and gamified environments show promise in fostering experimentation, reflection, and awareness of AI biases, offering potential to teach procedural skills while deepening understanding of AI limitations and ethical considerations.
Defining Project Direction
Defining Project Direction
We conducted a focus group with 8 participants to explore how young adults perceive and interact with GenAI technologies. Prior to the session, participants associated metaphors with their experiences, awareness of capabilities, trust, and the impact of over-trust in GenAI, and then explained their choices during the session. They viewed GenAI as a useful tool for automating repetitive tasks but raised concerns about trustworthiness, output quality, and over-reliance. While appreciated for low-stakes tasks like idea generation and simple coding, participants were cautious about using GenAI for complex work due to its incomplete outputs and limited understanding. Metaphors like Ladder and Broken Pencil Lead (shown to the right) emphasized GenAI systems’ utility but reliance on human intervention. Participants also expressed concerns about data privacy, societal dependence, and the need for transparency and education.
We conducted a focus group with 8 participants to explore how young adults perceive and interact with GenAI technologies. Prior to the session, participants associated metaphors with their experiences, awareness of capabilities, trust, and the impact of over-trust in GenAI, and then explained their choices during the session. They viewed GenAI as a useful tool for automating repetitive tasks but raised concerns about trustworthiness, output quality, and over-reliance. While appreciated for low-stakes tasks like idea generation and simple coding, participants were cautious about using GenAI for complex work due to its incomplete outputs and limited understanding. Metaphors like Ladder and Broken Pencil Lead (shown above) emphasized GenAI systems’ utility but reliance on human intervention. Participants also expressed concerns about data privacy, societal dependence, and the need for transparency and education.
Our findings led us to shift focus from trust in AI to mental models of AI and informed usage.
Our findings led us to shift focus from trust in AI to mental models of AI and informed usage.
Initial Ideation for Solutions
Initial Ideation for Solutions
Shifting our focus to improving young adults' mental models of AI, we brainstormed game ideas using "The Thing from the Future" and "Tarot Cards of Tech" design tools. These helped us factor aspects such as inclusion and accessibility, as well as highlighted important considerations for our game around autonomy and interpersonal connections. We then each came up with 15 one-sentence ideas, meeting and combining our ideas into a drawing and guessing based game focused on prompt engineering and mental models. After we had the final idea, we consulted with 2 “experts” in the field for validation and further insights. One of the experts suggested adding an adversarial role and player reflections to explore mental models, while emphasizing the need for a taxonomy on GenAI bias. The other expert highlighted the challenges in measuring bias mitigation, suggesting a focus on raising awareness and literacy,
Shifting our focus to improving young adults' mental models of AI, we brainstormed game ideas using "The Thing from the Future" and "Tarot Cards of Tech" design tools. These helped us factor inclusion and accessibility, as well as highlighted important considerations for our game around autonomy and interpersonal connections. We then each came up with 15 one-sentence ideas, meeting and combining our ideas into a drawing and guessing based game focused on prompt engineering and mental models. After we had the final idea, we consulted with 2 “experts” in the field for validation and further insights. One suggested adding an adversarial role and player reflections to explore mental models, while emphasizing the need for a taxonomy on GenAI bias. The other highlighted the challenges in measuring bias mitigation, suggesting a focus on raising awareness and literacy,
We aimed to enhance AI literacy via a drawing and guessing game.
We aimed to enhance AI literacy via a drawing and guessing game.
Low-Fidelity Prototyping
Low-Fidelity Prototyping
After pitching our narrowed ideas, we consulted with our professor, who introduced us to a game called Caution Signs and recommended that we prototype multiple game concepts. Acting on his feedback, we developed three prototypes inspired by Telestrations, Pictionary, and Caution Signs. During this process, we explored integrating GenAI into the existing game mechanics and experimented with various guessing styles, such as using lettered blanks (e.g., _ _ _) or selecting from a word list. We also tested different prompt structures, drawing from the original games.
Testing revealed that the most effective guessing style was selecting from a predetermined list, as freestyle approaches diverted focus to one’s ability for quick recall. It also highlighted the optimal prompting structure involved using two pairs of nouns and adjectives. In creating the nouns and adjectives, we applied the persuasive design principle of intermixing by creating bias-inducing and non-bias-inducing combinations. We also found that timed structures encouraged quick thinking and relying on implicit biases, so an untimed reflection stage was necessary for deeper insights on prompt engineering methods and bias recognition.
Low-Fidelity Prototyping
After pitching our narrowed ideas, we consulted with our professor, who introduced us to a game called Caution Signs and recommended that we prototype multiple game concepts. Acting on his feedback, we developed three prototypes inspired by Telestrations, Pictionary, and Caution Signs. During this process, we explored integrating GenAI into the existing game mechanics and experimented with various guessing styles, such as using lettered blanks (e.g., _ _ _) or selecting from a word list. We also tested different prompt structures, drawing from the original games.
Testing revealed that the most effective guessing style was selecting from a predetermined list, as freestyle approaches diverted focus to one’s ability for quick recall. It also highlighted the optimal prompting structure involved using two pairs of nouns and adjectives. In creating the nouns and adjectives, we applied the persuasive design principle of intermixing by creating bias-inducing and non-bias-inducing combinations. We also found that timed structures encouraged quick thinking and relying on implicit biases, so an untimed reflection stage was necessary for deeper insights on prompt engineering methods and bias recognition.
Mid-Fidelity Prototyping
Mid-Fidelity Prototyping
After establishing initial prompting and guessing structures, we turned our focus to the role of GenAI and how to balance AI involvement with human engagement. We explored six different iterations, each with varied roles for AI and humans, such as having AI draw or guess all the time, or mixing human and AI roles based on chance or selection. We found that allowing humans to choose between GenAI-generated images was the most effective in refining prompts. We also experimented with the number of adjectives presented, settling on a set of six adjectives for an optimal challenge. In parallel, we explored player roles, testing a version where one player guessed while two teams created prompts. However, this setup led to disengagement for the guesser, prompting us to shift to a structure similar to Caution Signs, where players were always engaged in either drawing or guessing. This led to the decision to establish three roles within a round to keep all players actively involved and engaged.
After establishing initial prompting and guessing structures, we turned our focus to the role of GenAI and how to balance AI involvement with human engagement. We explored six different iterations, each with varied roles for AI and humans, such as having AI draw or guess all the time, or mixing human and AI roles based on chance or selection. We found that allowing humans to choose between GenAI-generated images was the most effective in refining prompts. We also experimented with the number of adjectives presented, settling on a set of six adjectives for an optimal challenge. In parallel, we explored player roles, testing a version where one player guessed while two teams created prompts. However, this setup led to disengagement for the guesser, prompting us to shift to a structure similar to Caution Signs, where players were always engaged in either drawing or guessing. This led to the decision to establish three roles within a round to keep all players actively involved and engaged.
Iterative testing led to a three-role structure, with GenAI aiding the drawer.
Iterative testing led to a three-role structure, with GenAI aiding the drawer.
High-Fidelity Prototyping
High-Fidelity Prototyping
After refining the roles and gameplay, we conducted a playtest with three participants to gather feedback on the game mechanics and enjoyment. Key findings included difficulty with scoring, confusion over goals between phases (e.g., choosing a challenging prompt versus generating an easy-to-guess image), and players finding the time for generating images too short and stressful. Based on these insights, we clarified the scoring mechanism, integrated tutorials, and tested different timings, settling on 40 seconds for prompting, 60 seconds for drawing excluding GenAI’s image generation time, and 20 seconds for guessing based on the provided image.
We developed the final prototype in Figma, with me and another team member leading the design process. Inspired by the game's name, we incorporated whimsical elements to evoke creativity, featuring a vibrant landscape with a minimalist layout to create an approachable feel.
Our design aimed to reflect GenAI's potential while acknowledging its limitations and biases.
High-Fidelity Prototyping
After refining the roles and gameplay, we conducted a playtest with three participants to gather feedback on the game mechanics and enjoyment. Key findings included difficulty with scoring, confusion over goals between phases (e.g., choosing a challenging prompt versus generating an easy-to-guess image), and players finding the time for generating images too short and stressful. Based on these insights, we clarified the scoring mechanism, integrated tutorials, and tested different timings, settling on 40 seconds for prompting, 60 seconds for drawing excluding GenAI’s image generation time, and 20 seconds for guessing based on the provided image.
We developed the final prototype in Figma, with me and another team member leading the design process. Inspired by the game's name, we incorporated whimsical elements to evoke creativity, featuring a vibrant landscape with a minimalist layout to create an approachable feel.
Our design aimed to reflect GenAI's potential while acknowledging its limitations and biases.
Key Challenges
Key Challenges
The biggest challenge was coordinating the physical and the digital.
The biggest challenge was coordinating the physical and the digital.
Defining GenAI’s role involved multiple iterations. With many materials, like the scorebook, being physical, balancing digital AI input proved difficult. As we refined gameplay, we kept the goal of creating a custom platform in mind, adjusting GenAI’s role during testing to ensure seamless integration and enhance the player experience.
Another challenge was balancing engagement with goal alignment.
Exploring GenAI results was a key focus, so we tested various iterations, including ones where AI drew or guessed all the time. However, these extremes didn’t engage the human players. To improve this, we developed versions where AI played a more limited role, allowing humans to retain control over the AI-generated results thus ensuring that human players still were having fun.
Key Challenges
The biggest challenge was coordinating the physical and the digital.
Defining GenAI’s role involved multiple iterations. With many materials, like the scorebook, being physical, balancing digital AI input proved difficult. As we refined gameplay, we kept the goal of creating a custom platform in mind, adjusting GenAI’s role during testing to ensure seamless integration and enhance the player experience.
Another challenge was balancing engagement with goal alignment.
Exploring GenAI results was a key focus, so we tested various iterations, including ones where AI drew or guessed all the time. However, these extremes didn’t engage the human players. To improve this, we developed versions where AI played a more limited role, allowing humans to retain control over the AI-generated results thus ensuring that human players still were having fun.
My Learnings
My Learnings
1.
Leading research with systems thinking
Leading research with systems thinking
As one of the primary researchers, I led a significant amount of the research and the synthesis of associated findings. In the process, I made sure to consider the larger impacts of even the smallest of decisions that we made about our game.
2.
2.
Applying persuasive design principles
Applying persuasive design principles
The course introduced multiple principles through lecture format, and the project allowed us to do hands-on applications of these concepts. Some specific ones we relied on were obfuscation and intermixing. We also manipulated System 1 versus System 2 thinking to ensure that people had fun while still engaging with the underlying goals of our game.
Next Steps
Developing ImaginAItion
As the primary developer of the team, I will lead the effort to translate our ideas from the Figma prototype to a web game using React and HTML / CSS.
2.
Further testing
Once the web game is developed, we plan to continue testing and refining it based on observed results. As part of this process, we will document our findings in a research paper, where I will serve as a co-first author, and submit to relevant conferences to share our insights.
My Learnings
1.
Leading research with systems thinking
As one of the primary researchers, I led a significant amount of the research and the synthesis of associated findings. In the process, I made sure to consider the larger impacts of even the smallest of decisions that we made about our game.
2.
Applying persuasive design principles
The course introduced multiple principles through lecture format, and the project allowed us to do hands-on applications of these concepts. Some specific ones we relied on were obfuscation and intermixing. We also manipulated System 1 versus System 2 thinking to ensure that people had fun while still engaging with the underlying goals of our game.
Next Steps
1.
Developing ImaginAItion
As the primary developer of the team, I will lead the effort to translate our ideas from the Figma prototype to a web game using React and HTML / CSS.
2.
Further testing
Once the web game is developed, we plan to continue testing and refining it based on observed results. As part of this process, we will document our findings in a research paper, where I will serve as a co-first author, and submit to relevant conferences to share our insights.