As artificial intelligence (AI) becomes part of everyday life, we need engaging ways to teach young people about its ethical challenges. We designed EvolveMoralMaze, an interactive game to teach middle school students (ages 11-14) about AI ethics, training data, and algorithmic bias. We tested this tool with 117 students at a Texas public middle school. Both qualitative and quantitative data showed a large increase in students' self-reported understanding of AI. More importantly, students developed an in-depth understanding of how training data shapes AI behavior. However, the game inadvertently reinforced the misconception that ''perfect rules'' can solve ethical problems. This research demonstrates that games can be powerful tools for making complex ethical ideas tangible for this age group.

Artificial Intelligence (AI) is growing rapidly in our society and is now apparent in our day-to-day lives. With the recent burst of interest in AI, many individuals and children may view AI as something mystic and magical. It is important to demystify and introduce to them how AI is made and works. To address this need, we developed a software application that allows children to specify the parameters used by a Reinforcement Learning (RL) algorithm. Then students experience how RL is used to train an AI model to play the game "Snake." This software tool was tested with 71 middle school-age students. Here, we describe the design of the TrainYourSnakeAI application, the approach we used to introduce the associated ideas to middle school children, and how we assessed student learning. Qualitative data collected from students are presented and discussed. We surveyed their knowledge of AI before and after using the application. In this work, our research questions were: (RQ1) How can we create an engaging tool to teach reinforcement learning? and (RQ2) Does using our application foster a stronger understanding of reinforcement learning in children? Our findings indicate that students were able to understand the functionality of reward functions and how agents can learn from the environment using the concept of RL. We found that out of the 51 students who were not previously familiar with RL, 40 were able to provide adequate descriptions of RL after using TrainYourSnakeAI.