Controlling bad-actor-AI activity at scale across online battlefields

TL;DR

This paper analyzes the threat of malicious AI activity on social media, predicting escalation patterns and providing mathematical tools for policy decisions to mitigate harms at scale.

Contribution

It offers a detailed social media battlefield model, a Red Queen analysis predicting escalation timelines, and quantitative policy frameworks for containment and removal strategies.

Findings

Bad-actor-AI activity expected to escalate daily by early 2024.

Mathematical models predict where and when malicious AI will dominate.

Policy trade-offs can be quantitatively assessed using provided formulae.

Abstract

We show how the looming threat of bad actors using AI/GPT to generate harms across social media, can be addressed at scale by exploiting the intrinsic dynamics of the social media multiverse. We combine a uniquely detailed description of the current bad-actor-mainstream battlefield with a mathematical description of its behavior, to show what bad-actor-AI activity will likely dominate, where, and when. A dynamical Red Queen analysis predicts an escalation to daily bad-actor-AI activity by early 2024, just ahead of U.S. and other global elections. We provide a Policy Matrix that quantifies outcomes and trade-offs mathematically for the policy options of containment vs. removal. We give explicit plug-and-play formulae for risk measures.