Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation

2025/08/05 by Maik Larooij, Petter Törnberg, Larooij, Maik +1 · 36 voices · 2 citations
Computer Science · Physics and Astronomy · Social Sciences · #Bridging (networking) #Computational and Text Analysis Methods #Constructive #Core (optical fiber) #Generative grammar #Generative model #Language and cultural evolution #Opinion Dynamics and Social Influence #Prosocial behavior #Psychological intervention #Social dynamics #cs.CY #cs.SI

paper · pdf · doi:10.48550/arxiv.2508.03385

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/08/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05

Abstract

Social media platforms have been widely linked to societal harms, including rising polarization and the erosion of constructive debate. Can these problems be mitigated through prosocial interventions? We address this question using a novel method - generative social simulation - that embeds Large Language Models within Agent-Based Models to create socially rich synthetic platforms. We create a minimal platform where agents can post, repost, and follow others. We find that the resulting following-networks reproduce three well-documented dysfunctions: (1) partisan echo chambers; (2) concentrated influence among a small elite; and (3) the amplification of polarized voices - creating a 'social media prism' that distorts political discourse. We test six proposed interventions, from chronological feeds to bridging algorithms, finding only modest improvements - and in some cases, worsened outcomes. These results suggest that core dysfunctions may be rooted in the feedback between reactive engagement and network growth, raising the possibility that meaningful reform will require rethinking the foundational dynamics of platform architecture.

Citations

• Moving towards informative and actionable social media research
• Do Large Language Models Solve the Problems of Agent-Based Modeling? A Critical Review of Generative Social Simulations
• Simulating Social Media Using Large Language Models to Evaluate Alternative News Feed Algorithms
• Generative Agents: Interactive Simulacra of Human Behavior
• Bridging Systems: Open Problems for Countering Destructive Divisiveness across Ranking, Recommenders, and Governance
• Birdwatch: Crowd Wisdom and Bridging Algorithms can Inform Understanding and Reduce the Spread of Misinformation
• Agent-Based Models in Empirical Social Research

Cited by

• Quantifying the Lifelong Impact of Resilience Interventions via Agent-Based LLM Simulation
• The Significance of User Characteristics for Reposting Prediction on X: A Comparative Analysis Under Distribution Shift
• Segregation Before Polarization: How Recommendation Strategies Shape Echo Chamber Pathways

Discussions

• We built the simplest possible social media platform. No algorithms. No ads. Just LLM agents posting and following. It still became a polarization machine. Then we tried six interventions to fix socia [bsky, 299 points, 14 comments]
• （感想続き） Blueskyが比較的平和に見えるのは、集まる人々によるもので、現実世界を反映するほど「荒れる」可能性は、あるのかもしれない。なお、この研究は(1) 限定的な状況のシミュレーションである点、(2) 最後の見解はまだ仮説にとどまる点にはご注意。 オリジナル論文は下記で読めます。 Can We Fix Social Media? Testing Prosocial Interventio [bsky, 50 points, 0 comments]
• Maybe social media is unfixable? A new (ongoing) study suggests that "(1) partisan echo chambers; (2) concentrated influence among a small elite; and (3) the amplification of polarized voices" cannot [bsky, 21 points, 4 comments]
• Pro: Someone did the experiment I’ve been wanting to do for years (yay!) Con: The result points toward traditional social media being pretty much irredeemably structurally broken. [bsky, 6 points, 0 comments]
• plutôt que d’injecter de l’IA dans les réseaux sociaux, il est préférable de repenser le fonctionnement du système des réseaux sociaux et l’infrastructure pour les rendre plus démocratiques (on peut r [bsky, 4 points, 0 comments]
• While preparing for a course, I came across this important #preprint by @maiklarooij.nl and @pettertornberg.com: arxiv.org/abs/2508.03385 👈 Their starting point is that on #socialmedia, including @bs [bsky, 4 points, 1 comments]
• Can We Fix Social Media? Testing Prosocial Interventions Using Generative Sims [hn, 4 points, 1 comments]
• Can We Fix Social Media? Prosocial Interventions Using Generative Simulation [hn, 3 points, 2 comments]
• 週刊文春が不親切でオードリータンが挙げてる論文の出典を掲載してないけど、GPTに探させたら以下が見つかった arxiv.org/abs/2508.03385 [bsky, 3 points, 9 comments]
• Testing Prosocial Interventions Using Generative Social Simulation [hn, 2 points, 0 comments]
• Can We Fix Social Media? Testing Prosocial Interventions Using Social Simulation [hn, 2 points, 0 comments]
• In 2025 study, AI agents simulating social media reproduced several dysfunctions, e.g., partisan echo chamber, small elite influencers, amplification of polarizers. IMHO, researchers have simply descr [bsky, 2 points, 0 comments]
• Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation by Maik Larooij & Petter Törnberg, reported by @theaifix.show www.arxiv.org/pdf/2508.03385 | arstechnica.com [bsky, 1 points, 0 comments]
• Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation | arXiv arxiv.org/abs/2508.03385 #openaccess [bsky, 1 points, 0 comments]
• ... of interaction and visibility that define these environments." Fellow social media innovators: Should we accept the challenge from this paper? arxiv.org/pdf/2508.0... 2/2 [bsky, 1 points, 0 comments]
• 🤖 Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation They simulate social networks w AI agents and consistently reproduce: 1. partisan echo chambers 2. concen [bsky, 1 points, 1 comments]
• Selbst KI-Bots werden toxisch: Soziale Medien sind kaum zu retten. Lesenswertes wissenschaftliches Preprint. arxiv.org/abs/2508.03385 [bsky, 1 points, 0 comments]
• Larooij, M., & Törnberg, P. (2025). Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation (No. arXiv:2508.03385). arXiv. doi.org/10.48550/arX... [bsky, 1 points, 1 comments]
• New study shows that even without algorithms, social media breeds echo chambers, attention elites & amplified extremes. (and how hard it is to avoid it). arxiv.org/abs/2508.03385 [bsky, 1 points, 0 comments]
• Some researchers did just that, creating a “generative social simulation — that embeds [LLMs] within Agent-Based Models to create socially rich synthetic platforms…where agents can post, repost, and f [bsky, 1 points, 1 comments]
• ➡️ Die Untersuchung von Forschenden des Istitute for Logic, Language and Computation (ILLC) der University of Amsterdam 🇳🇱 auf der kuratierten Preprint-Plattform "arXiv" arxiv.org/pdf/2508.03385 2/2 [bsky, 1 points, 0 comments]
• Can We Fix Social Media? Testing Using Generative Social Simulation [hn, 1 points, 0 comments]
• Interesting experiment that shows emergent dysfunctional characteristics in simulated social networks "Can We Fix Social Media? Testing Prosocial Interventions using Generative Social #Simulation" arx [bsky, 0 points, 0 comments]
• I risultati dell'esperimento vanno comunque considerati con il giusto grado di scetticismo, dato che lo studio non è ancora stato sottoposto a peer-review, ma è comune interessante approfondire l'idea [bsky, 0 points, 1 comments]
• Link to the preprint. arxiv.org/abs/2508.03385 [bsky, 0 points, 1 comments]
• In a study 500 AI bots ended up forming echo chambers of their own in a simulated "social network". Based on GPT-4o mini and set "personas," they engaged with each other and available content, often f [bsky, 0 points, 0 comments]
• In other words, social media cannot be unf---ed. It'll rot your brain, take away your humanity, and prove addictive...much like the Sackler's Oxycontin. But the owners will get fabulously wealthy and [bsky, 0 points, 0 comments]
• ¿Podemos mejorar las redes sociales? Prueba de intervenciones prosociales mediante simulación social generativa Maik Larooij, Petter Törnberg Ciencias de la Computación > Redes Sociales y de Informaci [bsky, 0 points, 0 comments]
• 元の論文はこちらだそうです。 https://arxiv.org/abs/2508.03385 [bsky, 0 points, 0 comments]
• When University of Amsterdam researchers deployed 500 uniquely‑styled #ChatGPT bots on a simulated #SocialNetwork, they soon noticed familiar patterns. #AI partisan echo chambers concentrated influenc [bsky, 0 points, 0 comments]
• Looks like even synthetic humans turn into polarizing a*holes online. "Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation" arxiv.org/pdf/2508.03385 [bsky, 0 points, 0 comments]
• Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation https://www.arxiv.org/pdf/2508.03385 [bsky, 0 points, 0 comments]
• Echo Chambers, Angry Elites, and AI: Inside the Simulation That Proves Social Media's Problems Are Deeper Than You Think. arxiv.org/pdf/2508.03385 @moskov.goodventures.org @mcuban.bsky.social @reidhof [bsky, 0 points, 0 comments]
• Can We Fix Social Media? Testing Prosocial Interventions using Generative Social Simulation #socialmedia #dysfunction #mentalhealth #toxicity #polarization #llms #simulation #reactive #engagement #net [bsky, 0 points, 0 comments]
• arxiv.org/abs/2508.03385 [bsky, 0 points, 0 comments]
• If engagement-growth feedback drives chaos, what architectural change would you try first—caps on virality, decentralizing influence, or friction for hot takes? Paper: arxiv.org/abs/2508.03385 [bsky, 0 points, 0 comments]

Related

• Simulating Social Media Using Large Language Models to Evaluate Alternative News Feed Algorithms
• Designing social media to promote productive political dialogue
• AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society
• Generative Exaggeration in LLM Social Agents: Consistency, Bias, and Toxicity
• Synthetic Social Media Influence Experimentation via an Agentic Reinforcement Learning Large Language Model Bot