Date

2021.12 - 2022.02

Tools/Engine

Unity (ML-Agents)

Roles

Game Designer, Developer

Featured in K-arts Metaverse showcase

Ideas

“What does it mean to behave like a human, and what significance does creating such behavior hold? How is humanity itself constituted?”

The foundational idea emerged during a workshop on AI reinforcement learning. Observing AIs learning and adapting directly within an environment—growing through failure—sparked a series of questions for me. It made me consider the forces that shape us as individuals, and I became particularly interested in the role that words, sayings, and sentences play in our lives.

Mechanics and Development

Guiding players to reflect on their own struggles with compassionate detachment.

Players step into the role of an observer, rather than actively participating. By stepping out of direct control and supporting MyeongRan’s journey, players are encouraged to reflect on their own struggles from a compassionate, external perspective, fostering empathy toward their inner challenges. 

Players input words based on six categories: words they want to avoid, cherish, challenge, or push away, words that hover around them, or words that chase them. They can either enter words directly into these categories or drag and drop from a pre-listed selection on the screen. MyeongRan, the player’s avatar, then steps into the scene, ready to face both the emotional and physical obstacles created by the chosen words. Each word transforms into an obstacle, moving according to its category’s essence. MyeongRan’s task is to navigate through these word-based challenges, overcoming each obstacle to reach the door at the end of the path. The players watch as MyeongRan leaps forward on their behalf, confronting the obstacles until finally reaching the door, embodying the journey of overcoming emotional hurdles.

-

I began development by strategically placing floors and doors, aligning text, and setting random generation spots for text. I then implemented functionalities based on the symbolic nature of the texts: colliding, leaping, lingering, and pushing. Utilizing reinforcement learning, I conducted over 8 million training iterations for the AI, Myeong Ran. However, I hoped that Myeong Ran would not become a perfect AI. I wanted it to resemble us, prone to mistakes and failures. Therefore, I halted training when Myeong Ran’s success rate was neither too perfect nor too immature, striking a balance that reflected human imperfection.