.png)
Meme-opoly
Meme-opoly is a board game that harnesses the power of ChatGPT and of the classic board game to take group activities to another level with customizable tokens, game pieces, and storylines.
Your friends input player descriptions for each other that the AI uses to create humorous personalized moves and scenarios that drive the game forward with unique playthroughs every time.
Meme-opoly
2023
By Annie Dong and Quincy Kuang
Tools: Illustrator, VS Code, Rhino, ChatGPT, Laser Cutter, 3D Printer
How we started:
We wanted to integrate large language models (LLMs) with a physical interface. LLMs excel at certain tasks but lack an understanding of the physical world. Our goal was to design a physical interface to amplify the abilities of machine intelligence to enrich the human experience.
We thought about applications of AI that we hadn't seen yet, and looked into games. Among our initial interests were dungeon crawler games like Pokémon Mystery Dungeon, Hades, and Enter the Gungeon where you progress through the game by entering rooms where you encounter new enemies and obstacles. We liked the idea of changing and adaptable obstacles and level progressions. Concerns: Given the limited time and resources we had to complete the project, a complicated physical display controlled by an AI and a player may not be feasible.
Dungeon crawler games we looked into
We were also interested in role-play games like Dungeons and Dragons, where players can create personalized characters that affect gameplay. We liked the idea of an AI Dungeon Master that could create a customized storyline.
Of course, we also looked at board games. Board games are a classic way to have fun with your friends, and giving them a modern twist while creating an exciting experience for multiple people really appealed to us. Board games also tend to have a set group of choices, like a deck of cards, that can start to get repetitive. It is often randomness that makes board games engaging, so we asked, can we use AI to make board games more unpredictable and interesting?
We combined our favorite parts of all of these games to make a customizable story board game.
For this project, we started off with brainstorming the physical prototype first.
Some of the features we needed to address were what the board looked like/worked and how players were going to interact with the board/AI. We experimented with a board that sensed player locations through a barcode reader, and one that could light up (depicted on left).
But then, we thought about what really valued to us about this game. We wanted it to be accessible, engaging, and customizable. So, we redefined how we imagined board games.
Part 1 (physical):
We wanted to optimize simplicity and customization, so we chose to make a modular board with a game device connected to ChatGPT. The modularity enables both customization and convenience. We started off with foam tiles and a foam prototype that we later connected to ChatGPT.
Prototype 1
User testing of our prototype:
Observations:
-
They had a lot of fun drawing each other on the tokens and inputting each other's descriptions.
-
Waiting for the device to read the AI responses took a long time, and people kept forgetting to hit the button to prompt it - also, hitting that button prematurely would confuse it.
-
Sometimes the responses would glitch out and be irrelevant to the player descriptions.
-
The players had a lot of fun!
With our user feedback, we continued developing our prompt engineering (discussed in part 2). We wanted our default tiles to be visually interesting, but less literal than our prototype tiles that directly portrayed the environment of our original prompt. We researched tile patterns, but realized how difficult it would be to create interconnecting pieces that did not repeat shapes and were visually distinct from each other. We used Adobe Illustrator to experiment with abstract shapes, and created pieces that all had at least one angle that can fit perfectly with every other piece. We laser cut these pieces using different colors of acrylic, and rounded out the corners so people won't hurt themselves playing with them.
.png)
.png)
.png)
.png)
.png)
.png)
One of our test users' favorite parts was drawing each other on the tokens. These portraits of each other ranged from artistically impressive to downright scary, and everyone enjoyed making them. Our prototype tokens were foam tiles that could only be drawn on once, which we thought was wasteful and inconvenient. To make our final tokens, we laser cut white acrylic that can be drawn on with dry erase markers, with a piece of clear acrylic attached with a hinge to prevent the portraits from being smudged when players touch them.
neutral
bad
good
There are 8 tokens total for up to 4 players. According to how many players there are, the same number of tokens serve as buttons on our game device.
For the movable game pieces, we made small removable pieces that can be attached to the other tokens so they can stand.
The components of our ChatGPT-connected game device:
My favorite part of the device is the slidable tray that can hold the dry erase markers, the tokens, and the tiles. The entire game fitting into one device ensures portability and accessibility.
Part 2 (prompt engineering):
Throughout the physical prototyping process, we refined our ChatGPT inputs so there would be relevant responses and smooth gameplay. Using API calls, node, Javascript, and many, many versions of our prompt, we integrated a game narrator that guides the game with humorous turns.
The AI struggled with staying on topic, frequently forgetting what its job was in the context of the game, or saying irrelevant phrases that did not reflect the player inputs. We ended up going through 8 versions of prompts, making large changes every version to improve responses - for example, in earlier versions, the AI would take more than a minute reading out responses, so we shortened its responses to 3 sentences. We needed ChatGPT to be reliable yet customizable, which required a delicate balance of specificity and simplicity. Here is our final prompt:
Prompt engineering section 1 - General rules:
Prompt engineering section 2 - Game description:
Prompt engineering section 3 - Player description:
The whole process was tedious but taught us to think in unique ways to accommodate for ChatGPT's strengths and weaknesses. It challenged me in a unique way - deconstructing how I interpreted language and directions was an undertaking that I didn't expect out of my design projects.
Customizability:
Customizability was one of the most important features of the board game for us. We intentionally made many important features modular - so players can replace them with their own creativity.
These buttons and tiles are modeled out of clay
Awards:
.jpg)
.png)
.png)