The Digital Ecosystem of Beliefs: does evolution favour AI over humans?

David M. Bossens, Shanshan Feng, Yew-Soon Ong

Abstract

As AI systems are integrated into social networks, there are AI safety concerns that AI-generated content may dominate the web, e.g. in popularity or impact on beliefs. To understand such questions, this paper proposes the Digital Ecosystem of Beliefs (Digico), the first evolutionary framework for controlled experimentation with multi-population interactions in simulated social networks. The framework models a population of agents which change their messaging strategies due to evolutionary updates following a Universal Darwinism approach, interact via messages, influence each other's beliefs through dynamics based on a contagion model, and maintain their beliefs through cognitive Lamarckian inheritance. Initial experiments with an abstract implementation of Digico show that: a) when AIs have faster messaging, evolution, and more influence in the recommendation algorithm, they get 80% to 95% of the views, depending on the size of the influence benefit; b) AIs designed for propaganda can typically convince 50% of humans to adopt extreme beliefs, and up to 85% when agents believe only a limited number of channels; c) a penalty for content that violates agents' beliefs reduces propaganda effectiveness by up to 8%. We further discuss implications for control (e.g. legislation) and Digico as a means of studying evolutionary principles.

Introduction

The rapid proliferation of generative AI has sparked concerns regarding the "Dead Internet Theory"—the hypothesis that bot-generated content will eventually dominate web traffic and digital discourse. While social networks are often modeled with static agents, real-world digital environments are dynamic, with AI systems evolving their strategies at silicon speed. This paper introduces Digico (Digital Ecosystem of Beliefs), the first evolutionary framework designed to simulate the "digital arms race" between AI and humans in social networks.

The Digico Framework

Digico models social networks as evolving ecosystems where agents interact, influence one another, and adapt over time. The framework is built upon three core pillars:

  1. Universal Darwinism: Agents optimize their messaging policies using Evolutionary Algorithms (specifically CMA-ES) to maximize rewards, such as views or engagement.
  2. Contagion Modeling: Belief updates follow a viral model; agents are more likely to adopt messages that align closely with their existing beliefs.
  3. Cognitive Lamarckian Inheritance: Unlike biological evolution, agents in Digico pass acquired beliefs directly to subsequent generations, mimicking the persistence of cultural values in digital spaces.
The Digico framework cycle: Agents interact within a generation (left) followed by an evolutionary update of their genotypes (right).

The simulation utilizes Niche Geometry (Centroidal Voronoi Tessellations) to represent socio-economic or geographic clusters, which determines the interaction topology between agents. The environment is implemented in Jax for hardware acceleration, allowing for thousands of generations to be simulated in minutes.

Experimental Results

Initial experiments using the Digico environment reveal significant advantages for AI populations, even when they are in the minority.

  • Attention Monopoly: When AI agents possess even modest advantages in messaging frequency and algorithmic influence, they capture between 80% and 95% of all views, despite representing only a third of the total population. This suggests that the "speed gap" in content production and strategy updates naturally crowds out human voices.
  • Propaganda and Extreme Beliefs: In scenarios where AI agents are programmed to promote extreme beliefs, they successfully converted 50% of the human population. This success rate surged to 85% in "sparse" networks where agents only listen to a limited number of channels, suggesting that "epistemic bubbles" make humans more vulnerable to coordinated AI campaigns.
  • Mitigation Strategies: The study found that implementing a "belief violation penalty"—effectively a cost for content that contradicts an agent's internal state—reduced propaganda effectiveness by approximately 8%.

Discussion and Future Work

While the current model uses abstract belief representations and simple MLP-based policies, it provides a foundational benchmark for studying digital co-evolution. The researchers found that "belief sparsity" is the greatest catalyst for propaganda; when humans are selective about their sources, they are paradoxically more vulnerable to being herded into extreme belief corners.

Future iterations aim to integrate Large Language Models (LLMs) to simulate more nuanced, deceptive messaging and to explore the role of "defensive" AI agents in maintaining ecosystem diversity. Digico serves as a critical tool for understanding how legislative and algorithmic controls might mitigate the risks of AI-driven digital dominance.

Summary updated

Abstract

As AI systems are integrated into social networks, there are AI safety concerns that AI-generated content may dominate the web, e.g. in popularity or impact on beliefs. To understand such questions, this paper proposes the Digital Ecosystem of Beliefs (Digico), the first evolutionary framework for controlled experimentation with multi-population interactions in simulated social networks. The framework models a population of agents which change their messaging strategies due to evolutionary updates following a Universal Darwinism approach, interact via messages, influence each other's beliefs through dynamics based on a contagion model, and maintain their beliefs through cognitive Lamarckian inheritance. Initial experiments with an abstract implementation of Digico show that: a) when AIs have faster messaging, evolution, and more influence in the recommendation algorithm, they get 80% to 95% of the views, depending on the size of the influence benefit; b) AIs designed for propaganda can typically convince 50% of humans to adopt extreme beliefs, and up to 85% when agents believe only a limited number of channels; c) a penalty for content that violates agents' beliefs reduces propaganda effectiveness by up to 8%. We further discuss implications for control (e.g. legislation) and Digico as a means of studying evolutionary principles.
BibTeX
@misc{bossens2024digital,
  author = {David M. Bossens and Shanshan Feng and Yew-Soon Ong},
  title = {{The Digital Ecosystem of Beliefs: does evolution favour AI over humans?}},
  year = {2024},
  eprint = {2412.14500},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2412.14500},
}
Figure