Information Retrieval in the Age of Generative AI: The RGB Model

TL;DR

This work addresses how generative AI tools alter topic-specific information dynamics across web resources and training data. It develops the RGB model, a stochastic, multi-compartment framework with flows that generate, reinforce, and propagate answers among External sources, the Web, Training Sets, Search Engines, and LLMs, using color-coded notions to separate high- and low-quality content. It defines bias-to-quality, finite coupon lifetimes, and four metrics (FIUA, AIRI, FRQ, AIAI) to quantify accuracy, diversity, responsiveness, and autophagy, complemented by mean-field ODEs and discrete-event simulations. Empirical analysis on Stack Exchange data reveals that while high-quality answers can surface rapidly, substantial human effort remains essential, and under high AI-adoption scenarios the risk of misinformation and autophagy grows, highlighting the need for responsible design and verification in AI-enabled information retrieval.

Abstract

The advent of Large Language Models (LLMs) and generative AI is fundamentally transforming information retrieval and processing on the Internet, bringing both great potential and significant concerns regarding content authenticity and reliability. This paper presents a novel quantitative approach to shed light on the complex information dynamics arising from the growing use of generative AI tools. Despite their significant impact on the digital ecosystem, these dynamics remain largely uncharted and poorly understood. We propose a stochastic model to characterize the generation, indexing, and dissemination of information in response to new topics. This scenario particularly challenges current LLMs, which often rely on real-time Retrieval-Augmented Generation (RAG) techniques to overcome their static knowledge limitations. Our findings suggest that the rapid pace of generative AI adoption, combined with increasing user reliance, can outpace human verification, escalating the risk of inaccurate information proliferation across digital resources. An in-depth analysis of Stack Exchange data confirms that high-quality answers inevitably require substantial time and human effort to emerge. This underscores the considerable risks associated with generating persuasive text in response to new questions and highlights the critical need for responsible development and deployment of future generative AI tools.

Figures (9)

• Figure 1: RGB model illustration: Colored circles represent coupons of answers to novel questions. Dashed circles are GAI-generated answers added to the Web. Gray arrows show generation/reinforcement/replication processes.
• Figure 2: Impact of $C_f$ on the coupon addition probabilities of the red, green and blue primary colors. Vertical dashed lines correspond to the values chosen for the system flows.
• Figure 3: Temporal evolution of the FIUA indicator for the three considered scenarios. Shaded areas correspond to 95%-level confidence intervals (as in subsequent figures).
• Figure 4: Temporal evolution of the AIRI (left plot) and AIAI indicator (right plot) for the three considered scenarios.
• Figure 5: Probability over time that the majority of used answers on the Web are irrelevant (left y-axes), and FRQ indicator (right y-axes) in the post-GAI scenario, for initial number of black coupon in the Search Engine equal to 1,10,100.