Complexity of Social Media in the Era of Generative AI

PERSPECTIVE ON GENERATIVE AI AND SOCIAL MEDIA

Introduction

  • Authors: Liming Pan, Chong-Yang Wang, Fang Zhou, Linyuan Lü

  • Published in: National Science Review 12: nwae323, 2025

  • Date: Advance access publication on 19 September 2024

Overview of Generative AI (GenAI)

  • Recent years have seen a profound expansion of generative AI creativity, notably in language and image generation sectors.

  • Examples of Generative AI Programs:

    • ChatGPT: A large language model that excels in creating natural-sounding text.

    • Stable Diffusion: A text-to-image model capable of producing realistic images.

Impact on Social Media Landscape

  1. Positive Aspects:

    • GenAI facilitates the rapid generation of captivating, tailored content.

    • Enables personalized recommendations, minimizing prior demands for time and human effort.

    • Aids in identifying and filtering harmful content on social media platforms.

  2. Risks and Challenges:

    • Potential for generating false information due to insufficient training.

    • Social robots using GenAI may have a major influence, posing challenges in detection.

    • Centralization of social media: Deployment of GenAI could lead to more centralized or ‘synthetic’ social media experiences.

Modeling Information Propagation in Social Media

  • Understanding information propagation at the individual level is crucial to characterize collective behaviors in the GenAI era.

  • The mechanics of content creation, sharing, and consumption are influenced by:

    • Users’ interests

    • Temporal activation patterns

    • Connection patterns

Changes Brought on by GenAI

  • GenAI agents can create content that aligns more closely with users’ interests, increasing the likelihood of sharing.

  • Temporal Activation Patterns: GenAI may produce content irrespective of human-related temporal factors (like weekdays or holidays).

  • Connection Diversity: GenAI users potentially showcase a broader range of connections, unlike human users who often maintain homogeneous interactions.

Collective Behavior Implications

  • Changes in individual behavior can lead to:

    • Shifts in propagation outbreak thresholds.

    • Influence on public opinion formation.

    • Emergence of echo chamber effects.

  • GenAI agents can be engineered to capitalize on these dynamics to manipulate large-scale information dissemination (e.g., spreading political misinformation).

Social Bot Detection Methods

  1. Historical Context: The concept of social bots has been present since early computing; detection methods have evolved over time due to their potential harm.

  2. Classical Detection Methods:

    • Graph-based methods: Analyze connectivity patterns to identify discrepancies indicative of bot activity.

    • Feature-based methods: Encode user behavior into features for machine learning analysis of human versus bot behaviors.

    • Crowdsourcing approaches: Leverage human effort for identifying social bots via large worker pools.

Modern Challenges

  • AI advancements make social bots appear increasingly human-like, complicating detection processes.

  • Tailored methods to adapt to the unique features of GenAI bots are essential for effective identification.

  • Examples of AI-assisted detection technologies include:

    • SynthID: Watermarking text, images, and audio to identify generative content.

    • GPTZero: Utilizes perplexity and burstiness to determine if a text was generated by AI.

Network Interventions

  • Successful models integrating GenAI bots enhance network intervention strategies to improve dynamics of information propagation.

  • The aim of interventions:

    • Optimize behavioral change.

    • Address issues like polarization and the echo chamber effect.

  • Intervention processes should evolve to accommodate GenAI bots' unique aspects, moving beyond small-scale perturbations to larger, optimized designs.

Future Implications of Generative AI

  • The generative AI landscape is still evolving, and its future impact on social media and society remains uncertain.

  • Continued systematic studies are necessary in:

    • Information propagation mechanisms.

    • Social bot detection technologies.

    • Effective network interventions within GenAI-involved ecosystems.

  • Such interdisciplinary efforts between AI, network science, and related fields are essential for navigating the complexities of social media in the GenAI era.

Funding

  • Supported by:

    • National Natural Science Foundation of China

    • STI 2030–Major Projects

    • New Cornerstone Science Foundation via the XPLORER PRIZE.

Authors Affiliation

  1. Liming Pan: School of Cyber Science and Technology, University of Science and Technology of China.

  2. Chong-Yang Wang, Fang Zhou: Institute of Fundamental and Frontier Studies, University of Electronic Science and Technology of China.

  3. Linyuan Lü: Corresponding author, University of Electronic Science and Technology of China, Email: linyuan.lv@uestc.edu.cn.