Finite-Horizon Collapse of Platforms Driven by Information Fatigue: An Agent-Based Threshold Model Inspired by Cheap Talk

Abstract

In recent years, within the attention economy centered on social media, the proliferation of sensational fake news and noise aimed at capturing attention (economic and social gains) has emerged as a severe social pathology. Focusing on the defining characteristic of the modern information environment—that the cost of information transmission is virtually zero—this paper adopts the “Dynamic Cheap Talk Model” originating from Crawford and Sobel (1982) as its theoretical foundation, rather than classical signaling theory (Spence, 1973), which separates high- and low-quality information through cost differentials. In a cheap-talk environment with zero transmission costs, the self-interested pursuit of senders boundlessly increases noise on the platform. This study computationally illustrates the mechanism by which this excessive noise irreversibly accumulates “Information Fatigue” among receivers, depleting the cognitive resources (a commons) required to infer the veracity of information, and ultimately leading to the collapse of the entire platform into a “Market for Lemons.” To demonstrate the validity of this approach, we constructed a large-scale multi-agent simulation where high-quality senders and low-quality senders (noise-makers) coexist, and conducted computational experiments. The analysis computationally illustrates that when the proliferation of noise variance exceeds a certain critical point, a finite-horizon collapse of total payoff occurs due to the simultaneous opt-out of receivers, resulting in a phenomenon where even high-quality senders are collaterally driven out (the failure of unraveling). Furthermore, we conducted extended computational experiments incorporating three advanced dynamics: (1) algorithmic amplification of noise (preferential attachment), (2) evolutionary game dynamics of sender strategy adaptation (the chain of degradation), and (3) the delay limits of ex post crowdsourced fact-checking (e.g., Community Notes). The results reveal that algorithmic intervention dramatically accelerates the platform’s self-destruction limit, and the senders’ adaptation process causes the system to experience catastrophic failure before the degradation of all agents is complete. Additionally, we quantitatively identified that to avert collapse through institutional intervention, corrections must be implemented within a strict “Maximum Tolerable Delay.” This research elucidates Gresham’s dynamics of information in the attention economy through complex system simulation, providing simulation evidence for maintaining the hygiene of information spaces and designing sustainable platforms.