Dynamics of Human-AI Collective Knowledge on the Web: A Scalable Model and Insights for Sustainable Growth

TL;DR

The web’s shared archives now reflect a human–AI knowledge ecosystem where co-creation can both accelerate growth and threaten quality. The authors propose a minimal five-variable dynamical system with state variables $K(t)$, $q(t)$, $\theta(t)$, $H(t)$, and $Q(t)$ to capture how human and AI content inflows, moderation gates, and learning pathways interact. The main contributions are (i) a unified framework that explains regimes from healthy growth to collapse and oscillations, (ii) actionable design and policy levers (content gates, RLHF emphasis, and learning pathways) to sustain growth, and (iii) empirical calibration on Wikipedia and two domain case studies (PubMed, GitHub Copilot) illustrating regime-specific dynamics. The framework provides a quantitative basis for platform-level audits and governance to maintain both the size and the quality of collective knowledge on the Web.

Abstract

Humans and large language models (LLMs) now co-produce and co-consume the web's shared knowledge archives. Such human-AI collective knowledge ecosystems contain feedback loops with both benefits (e.g., faster growth, easier learning) and systemic risks (e.g., quality dilution, skill reduction, model collapse). To understand such phenomena, we propose a minimal, interpretable dynamical model of the co-evolution of archive size, archive quality, model (LLM) skill, aggregate human skill, and query volume. The model captures two content inflows (human, LLM) controlled by a gate on LLM-content admissions, two learning pathways for humans (archive study vs. LLM assistance), and two LLM-training modalities (corpus-driven scaling vs. learning from human feedback). Through numerical experiments, we identify different growth regimes (e.g., healthy growth, inverted flow, inverted learning, oscillations), and show how platform and policy levers (gate strictness, LLM training, human learning pathways) shift the system across regime boundaries. Two domain configurations (PubMed, GitHub and Copilot) illustrate contrasting steady states under different growth rates and moderation norms. We also fit the model to Wikipedia's knowledge flow during pre-ChatGPT and post-ChatGPT eras separately. We find a rise in LLM additions with a concurrent decline in human inflow, consistent with a regime identified by the model. Our model and analysis yield actionable insights for sustainable growth of human-AI collective knowledge on the Web.

Figures (4)

• Figure 1: Overview of the proposed model of human-AI collective knowledge (Eq. \ref{['eq:model']}). Different parameter configurations of the model produce distinct, interpretable growth regimes in the human-AI collective knowledge ecosystem. The model provides a unified view of phenomena such as model collapse, quality dilution and competence inversion, and yields insights on strategies for sustainable growth. It can be calibrated using data to understand real-world platform dynamics as discussed in Sec. \ref{['sec:results']}.
• Figure 2: Examples of different growth regimes in the human-AI collective knowledge ecosystem modeled by Eq. \ref{['eq:model']} (and illustrated in Fig. \ref{['fig:schematic']}) for different parameter configurations. These results yield insights on policy- and platform- level interventions that can ensure sustainable growth of the size and quality of the shared online archive, human skill, and model capacity.
• Figure 3: Real-world case studies related to two different archives that lead to different growth dynamics.
• Figure 4: Model calibration with Wikipedia data for pre- and post- ChatGPT eras (33 months each). The monthly flows (solid) vs. model-implied flows (dashed) are shown for the flow equation (Eq. \ref{['eq:model_K']}). The estimated parameters are given in Table \ref{['tab:wikipedia-calib']}.