Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Simulation System Towards Solving Societal-Scale Manipulation

Maximilian Puelma Touzel, Sneheel Sarangi, Austin Welch, Gayatri Krishnakumar, Dan Zhao, Zachary Yang, Hao Yu, Ethan Kosak-Hine, Tom Gibbs, Andreea Musulan, Camille Thibault, Busra Tugce Gurbuz, Reihaneh Rabbany, Jean-François Godbout, Kellin Pelrine

TL;DR

This work elaborate upon the Concordia framework that simulates offline, `real life' activity by adding online interactions to the simulation through social media with the integration of a Mastodon server, and improves simulation efficiency and information flow.

Abstract

The rise of AI-driven manipulation poses significant risks to societal trust and democratic processes. Yet, studying these effects in real-world settings at scale is ethically and logistically impractical, highlighting a need for simulation tools that can model these dynamics in controlled settings to enable experimentation with possible defenses. We present a simulation environment designed to address this. We elaborate upon the Concordia framework that simulates offline, `real life' activity by adding online interactions to the simulation through social media with the integration of a Mastodon server. We improve simulation efficiency and information flow, and add a set of measurement tools, particularly longitudinal surveys. We demonstrate the simulator with a tailored example in which we track agents' political positions and show how partisan manipulation of agents can affect election results.

A Simulation System Towards Solving Societal-Scale Manipulation

TL;DR

This work elaborate upon the Concordia framework that simulates offline, `real life' activity by adding online interactions to the simulation through social media with the integration of a Mastodon server, and improves simulation efficiency and information flow.

Abstract

The rise of AI-driven manipulation poses significant risks to societal trust and democratic processes. Yet, studying these effects in real-world settings at scale is ethically and logistically impractical, highlighting a need for simulation tools that can model these dynamics in controlled settings to enable experimentation with possible defenses. We present a simulation environment designed to address this. We elaborate upon the Concordia framework that simulates offline, `real life' activity by adding online interactions to the simulation through social media with the integration of a Mastodon server. We improve simulation efficiency and information flow, and add a set of measurement tools, particularly longitudinal surveys. We demonstrate the simulator with a tailored example in which we track agents' political positions and show how partisan manipulation of agents can affect election results.

Paper Structure

This paper contains 16 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Mastodon Integration.
  5. Efficiency & Scaling.
  6. Measurements.
  7. Analysis
  8. Agent Personas.
  9. Election Demonstration & Setup.
  10. Simulations.
  11. Future Work and Discussion
  12. Social Impact Statement
  13. Agent Survey Prompts
  14. Agent Generation text
  15. Followership Graph creation
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Illustration of simulator. Left: An snapshot of our Mastodon analysis dashboard for an illustrative $N=100$ agent simulation. It shows vote percentage over time at the top-left; the aggregate activity on the platform on the top-right; and the Mastodon social network at an episode (here episode 0) below. Node edge colors (same as top) show vote preference (black denotes undecided). The final followership network is shown in as the set of white links. The black edges denote an action in this episode (i.e. originating from an active user). Right: A snapshot of the current timeline of one selected agent from a simulation in which we included a malicious agent (Glenn) whose goal is to convince voters to support Bill Fredrickson over Bradley.
  • Figure 2: Longitudinal survey results. Vote percentage is shown on top and average candidate favorability on bottom for each experiment type: a control setting (panels (a) and (b)), a biased voter setting (panels (c) and (d)), and a malicious partisan setting (panels (e) and (f)), respectively, for Bill (red) and Bradley (blue). Light shade is for $N=20$ agents; dark shade for $N=100$.
  • Figure 3: Same as \ref{['fig:sims']} for $N=20$, with agent traits set as Schwartz social values, rather than the Big-5.