Paparazzi

A civic participation experiment designed to support national insight initiatives.

Comment

What is Paparazzi?

Paparazzi is a multiplayer 'civic participation experiment' where participants discreetly photograph one another while navigating around real locations. Captured images are automatically processed by an AI system, which evaluates visible behavior and classifies subjects as compliant or non-compliant, under authorisation by the department of surveillance.

Inspiration

We wanted to build something that explores a speculative near-future where AI systems quietly mediate civic participation, compliance, and social value - turning citizens against each other, encouraging suspicion. The goal of the short multiplayer game is to reflect on how AI may reshape, agency, andconsent in the future of world governments. The surveillance-driven theme draws inspiration from George Orwells 1984 and wider themes of 'Big Brother is watching' rhetoric.

How we made it?

Papparazi was built as a full-stack, real-time multiplayer system powered by websockets. Our interface follows official GOV-UK theming, through a self-made component library in Flutter. Most of the backend is written using Python, which connects across 5 AWS Lambda Functions.

  • State is handled through a serverless DB (DynamoDB), storing which players are 'in the game'
  • Sharing images is handled through websockets (and AWS-Lambda, see our diagram)
  • Computer Vision runs on an AWS-Lambda to handle facial recognition, editing, segmentation etc...

Challenges we ran into/Things we're proud of

  • Our Lambda function for computer vision used so many dependencies we had to package it into a Docker image, which was a pain to debug!
  • Continuously re-architecting a fully serverless backend, to ensure our bank accounts were safe!
  • The concurrent use of 3 external dependencies: Gemini, MapBox, and of course AWS
  • Following existing GOV UI themes to provide a 'familiar' experience to participants.
  • Managing real-time location updates between all players on the shared map.
  • Generating high-quality, robust embeddings in small cloud functions (no mediapipe).
  • Establishing the bureaucratic, authoritative tone that AI governance could enforce.
  • Networked, online, public, scalable, web multiplayer (above all!)
  • Creating an orientation based map with live player positions (using web)

What's next for Paparazzi?

  • Full Palantir buyout.
Share this project:

Updates