AI is Doomed!

Beginning Screen
Human Player Support
Example of AI making choices

Inspiration

The interaction between intelligent agents, its objectives, and the surrounding environment form the basis of many different complex social, economic, and cultural systems that we see in today's society. However, simulating all of society is quite a complex task and so simulating a social game in which the interactions of the players, their objectives, and their decisions affect the outcome of the game and everyone else.

What it does

This game intends to simulate the popular card based social game "We're Doomed!" in which multiple world leaders must escape a dying planet in a limited amount of time . This game involves trust, backstabbing, alliances, and survival which perfectly acts as a abstracted version of how and why individuals make certain decisions.

How we built it

We built a Flask-based multi agent simulation where autonomous AI agents compete in a zero sum environment, learning from outcomes, adapting their strategies, and overcoming each other until a single winner emerges. This is then hosted using render to make a website that can be affected by a python script during use. The program uses API calls to groq.api where we tell the AI about the game state, relationships and previous actions of other AIs and the resources of all the players in order for the AI to make the best choice for itself. The website front end was made using CSS and HTML.

Challenges we ran into

Integrating real AI into the game required replacing a previously scripted, randomized response system with fully autonomous agents capable of reasoning over game state. In parallel, we migrated the entire system from a locally hosted setup to a server based architecture, requiring new back-end coordination and persistent connections to ensure the AI could operate reliably within the live game loop. We also planned to use OpenRouter but struggled with the API calls resulting in us needing to be adaptable to use groq.api.

Accomplishments that we're proud of

A fully functional application that supports complete game mechanics, allows human participation, and enables up to ten autonomous AI agents to interact and compete within the same game environment.

What we learned

AI IS DOOMED!!!! (Also, how API calls work, token efficiency, how to build a website, the limitations of AI agents being used to code and prompt engineering.)