Greg the Goose

💡 Inspiration

While attending the Hack the North opening ceremony, I caught a beautiful goose plushy that was being thrown into the crowd. It resonated with me due to UWaterloo’s peculiar history with geese. Which gave rise to our idea — a fun game that benefits both human and machine.

🔎 What it does

Similar to the classic game Taboo, Greg the Goose is an AI powered game where you have to describe a selected word to Greg without mentioning the word itself or 3 given related words. All you need to do is think and shout it to Greg and he’ll try his best to guess the word. But there’s a twist! Greg gets angry when he doesn’t win. Will you win with Greg?

🪿 A double win

As people play, Greg collects their responses and uses them to improve its language comprehension. This data set could be of significant use for AI to understand human communication and mindset behind it. Since people are avoiding using obvious words they provide far less articulate responses, raising the bar for Greg to be successful. These training sets can be especially valuable for AI to understand those who may not be able to fully articulate their thoughts.

🔨 How we built it

We fully utilized the capabilities of Cohere's text generation LLM to create our AI powered companion Greg. Greg is provided with a set of our prompts and the user's input, and its task is to generate a single word that accurately reflects the user's description. Throughout this iterative process, the model receives feedback on the correctness of its responses, allowing it to continuously improve. Thanks to Google Cloud's speech to text capabilities, our users get to interact with Greg using natural speech, enhancing the overall user experience.

🚧 Challenges we ran into

During the development process lack of documentation hindered our ability to quickly implement certain APIs. However, we persevered. After a ten hour gamble, we successfully implemented the Google Cloud Text to Speech API. Their wonderful API made the risk absolutely worth it.

🏆 Accomplishments that we're proud of

We are proud of the clean UI and Greg’s lovingly annoying personality. We are proud of coming together and creating an idea none of us could have dreamed of alone. But, most importantly we are proud that we didn’t give up despite the challenges we faced.

📕 What we learned

The Power of Real-Time Data Collection:

We discovered the value of real-time, user-generated data in training AI models. By having users speak and describe words, we realised that this dataset could significantly contribute to enhancing AI's understanding of human communication and the underlying thought processes.

Technical Skills:

Throughout this project, we honed our technical skills, delving into full-stack development. We gained valuable experience by working with a wide range of APIs in unison. Additionally, we enhanced our game design skills, further expanding our technical capabilities.

🪜 What's next for Greg the Goose

Scalability will become a huge focus. As more people utilize Greg, an increasing number of people will enhance their ability to articulate. Simultaneously, it aids in the exponential improvement of AI's capacity to comprehend humans, especially people who may have difficulties verbally expressing their thoughts.