# Information Propagation and Decay in Social Networks

The study of how information, behaviors, and innovations propagate through populations has undergone a profound structural paradigm shift. Historically, scientific understanding of communication and knowledge acquisition relied on linear, cumulative models. However, drawing upon the epistemological frameworks proposed by Thomas Kuhn, the integration of graph theory and computational social science has fundamentally revolutionized this discipline [cite: 1, 2]. The realization that human interactions do not resemble uniform grids or purely random distributions has led to the development of sophisticated topological models that dictate how a signal spreads, amplifies, or decays [cite: 3, 4, 5]. 

Through the lens of modern network science, the dissemination of information is not merely a function of the message's intrinsic qualities, but an emergent property of the underlying graph topology. By examining the mathematical mechanics of graph structures, the theoretical distinctions between simple and complex contagions, the specific differential equations governing decay, and the empirical realities of algorithmically curated platforms, a comprehensive understanding of information life cycles can be established.

## Topological Foundations of Network Science

To quantify the spread of information, populations are modeled as complex graphs, denoted mathematically as $G(N, L)$, where $N$ represents a set of nodes (individuals, accounts, or agents) and $L$ represents a set of links (edges, relationships, or interaction pathways) [cite: 6, 7]. Each link can be further specified by weights indicating tie strength, while each node possesses properties indicating its state or processing capacity [cite: 6]. The architecture of these connections ultimately dictates the efficiency and reach of any propagation event. The evolution of network science has produced three primary theoretical models to explain these macro-structures.

### Random Graphs and the Erdös-Rényi Model

The Erdös-Rényi (ER) model generates random graphs by connecting an isolated set of nodes with a uniform probability $p$ [cite: 8, 9]. In a vast ER network, the degree distribution—the probability $P(k)$ that a randomly chosen node has exactly $k$ connections—approximates a Poisson distribution [cite: 9, 10]. This statistical reality indicates that the vast majority of nodes possess a degree highly clustered around the network's average connectivity, effectively forbidding the existence of massive, highly connected hubs [cite: 8, 9]. 

While random graphs exhibit short average path lengths—meaning information can theoretically traverse the network in very few steps—they lack the dense local clustering observed in empirical human societies [cite: 11, 12]. In real-world social networks, if node A is connected to node B, and node B is connected to node C, there is a high probability that node A is directly connected to node C. This phenomenon, known as triadic closure, results in a high clustering coefficient, a structural feature that the ER model fundamentally fails to replicate [cite: 11, 13].

### Small-World Networks

To reconcile the short path lengths of random graphs with the high clustering coefficients of human social networks, the Watts-Strogatz (WS) model introduced the concept of the small-world network [cite: 8, 14]. The WS model begins with a regular ring lattice where nodes are tightly clustered with their immediate spatial neighbors. Edges are then randomly rewired with a specific probability $p$ [cite: 11, 15]. 

This rewiring mechanism introduces long-range shortcuts across the graph. Even a minuscule fraction of rewired edges causes the average path length $L$ to collapse precipitously, growing only proportionally to the logarithm of the number of nodes ($\log N$) [cite: 11, 14]. Crucially, while the path length drops to match random graphs, the global clustering coefficient remains robustly high [cite: 11, 14]. The resulting architecture mirrors the "six degrees of separation" phenomenon, demonstrating how isolated clusters of dense communities are bridged by sparse, long-distance weak ties [cite: 11, 14]. The degree of "small-world-ness" ($S$) can be rigorously quantified by comparing the clustering and path length of a given empirical network to an equivalent ER random graph, allowing analysts to establish a continuous grading of network topologies rather than relying on binary categorizations [cite: 15].

### Scale-Free Networks and Preferential Attachment

Empirical observations of the World Wide Web, academic citation networks, and early digital social networks revealed that degree distributions often do not follow a Gaussian or Poisson curve, but rather a heavy-tailed power law [cite: 8, 16]. A network is defined as scale-free if the fraction of nodes with degree $k$ follows the distribution $P(k) \sim k^{-\gamma}$, where the scaling parameter $\gamma$ typically falls between 2 and 3 [cite: 16]. 

Scale-free networks are generated through an evolutionary mechanism known as preferential attachment, formalized by the Barabási-Albert (BA) model [cite: 10, 16]. As the network grows iteratively, new nodes are probabilistically more likely to attach to existing nodes that already possess a high degree. This "rich-get-richer" dynamic organically produces a topology dominated by a few massive hubs and a massive, trailing tail of peripheral nodes with very few connections [cite: 8, 16]. 

The scale-free property heavily influences network robustness and information propagation. Because the vast majority of shortest paths flow through the central hubs, the network is highly resilient to the random failure of peripheral nodes [cite: 14, 17]. However, it is structurally fragile against targeted attacks; the removal or incapacitation of a key hub can rapidly fracture the network into isolated, non-communicating components, immediately halting global information propagation [cite: 14, 17, 18]. 

It is necessary to note that recent rigorous statistical analyses have challenged the universality of the scale-free paradigm. An analysis of nearly 1,000 diverse real-world networks revealed that pure power-law distributions are empirically rare. Log-normal distributions often fit empirical degree data as well or better than strict power laws, indicating that while heavy-tailed hub structures are exceedingly common in social graphs, the strict mathematical definition of scale-free topologies must be applied with calibrated caution [cite: 19].

| Topological Model | Degree Distribution | Clustering Coefficient | Average Path Length | Propagation Characteristics |
| :--- | :--- | :--- | :--- | :--- |
| **Random (ER)** | Poisson / Binomial | Low (Approaches 0 as $N \to \infty$) | Low ($L \propto \log N$) | Supports rapid transmission, but lacks the local redundancy needed for behavioral reinforcement. |
| **Small-World (WS)** | Peaked around average | High | Low ($L \propto \log N$) | Highly efficient for simple contagions via weak tie shortcuts while maintaining local community density. |
| **Scale-Free (BA)** | Power-Law ($P(k) \sim k^{-\gamma}$) | Decreases with node degree | Very Low | Hub-dominated propagation; highly vulnerable to catastrophic fragmentation if hubs fail or are removed. |

### Network Centrality Metrics

Understanding how information flows and eventually halts requires identifying which specific nodes or edges exert the most structural influence. Centrality metrics quantify this influence from distinct geometric perspectives:

*   **Degree Centrality:** A strictly local measure counting the absolute number of direct connections a node possesses. In directed networks, this is bifurcated into in-degree (popularity or receptivity) and out-degree (sociability or broadcast capability) [cite: 20, 21].
*   **Betweenness Centrality:** Measures the frequency with which a node sits on the shortest path between all other unlinked node pairs in the network. Nodes with high betweenness act as critical bottlenecks or bridges; they control the flow of communication and have the power to artificially decay information by refusing to pass it along [cite: 20, 21].
*   **Eigenvector Centrality:** A recursive computational measure where a node's influence is determined by the cumulative influence of its neighbors. A node achieves high eigenvector centrality not merely by having many connections, but by being connected to other highly central nodes [cite: 20].
*   **PageRank:** A derivation of eigenvector centrality tailored for directed graphs. It dampens the influence of nodes that indiscriminately link outward, heavily emphasizing the value of inbound links received from other high-status nodes within the graph [cite: 20, 21].
*   **Edge Centrality (ECHO):** Recent research has expanded beyond node-centric measures to evaluate the specific structural importance of individual links. Metrics like ECHO evaluate neighborhood-based optimization objectives, allowing analysts to rank the importance of specific communication pathways even in heavily disconnected or fragmented networks where traditional node metrics fail [cite: 22, 23].

## The Mechanics of Simple and Complex Contagions

The topological properties of a network do not interact with all types of information uniformly. A foundational breakthrough in computational sociology delineates idea propagation into two entirely distinct classes of diffusion dynamics: simple contagions and complex contagions [cite: 24, 25, 26].

### Simple Contagion and the Strength of Weak Ties

A simple contagion operates under the strict epidemiological assumption that a single exposure to an activated node is sufficient to transmit the state to a susceptible node [cite: 27, 28, 29]. The transmission of biological infectious diseases, the spread of straightforward news, the diffusion of benign gossip, or the forwarding of a highly viral, low-stakes image generally adhere to simple contagion dynamics [cite: 24, 27, 30]. 

In simple contagions, the speed and systemic reach of diffusion are maximized by network structures that minimize average path lengths. According to Granovetter's seminal theory on the "strength of weak ties," dense clusters of strong ties are highly redundant for information acquisition; individuals within a tight-knit cluster generally possess the exact same information [cite: 24, 27]. Conversely, weak ties—casual acquaintances or distant connections that bridge entirely different, isolated social clusters—act as high-speed conduits [cite: 24, 27]. Therefore, random graphs and small-world networks, which feature abundant long-distance weak-tie shortcuts, are structurally optimal architectures for the rapid, unimpeded spread of simple contagions [cite: 24, 31, 32].

### Complex Contagion and the Necessity of Wide Bridges

In stark contrast, a complex contagion dictates that transmission requires multiple sources of exposure and continuous social reinforcement before an individual adopts the new behavior, belief, or idea [cite: 24, 33]. The adoption of costly personal behaviors, participation in high-risk collective political action, the embrace of controversial societal norms, or the migration to unproven technological platforms consistently behave as complex contagions [cite: 25, 31]. 

The requirement for multiple exposures fundamentally alters how the contagion interacts with the network topology. Extensive research by Damon Centola demonstrates that the very weak ties that accelerate simple contagions actively inhibit the spread of complex contagions [cite: 24, 31]. A single, isolated weak tie connecting two distinct communities cannot provide the redundant exposures required to overcome an individual's innate adoption threshold [cite: 27, 31]. 

The sociological mechanisms necessitating multiple exposures in complex contagions include four primary drivers [cite: 24, 30]:
1.  **Coordination:** Certain innovations or social technologies only yield practical value when a critical mass of an individual's immediate network also adopts them (e.g., migrating to a new encrypted messaging app).
2.  **Credibility:** The veracity of a highly consequential or deeply controversial claim is established not through a single source, but through independent confirmation from multiple trusted, unconnected contacts.
3.  **Legitimacy:** The social risk of adopting a taboo, novel, or deviant behavior is mitigated only when multiple peers publicly endorse it, thereby establishing a new localized social norm.
4.  **Emotional Contagion:** High-risk actions, such as participating in an unsanctioned protest or strike, require the compounding emotional arousal and psychological safety generated by observing multiple network neighbors actively participating.

Because of these inherent barriers, complex contagions require "wide bridges"—multiple, overlapping, and redundant connections shared across network clusters—to successfully propagate from one community to another [cite: 27, 30, 31].

[image delta #1, 0 bytes]

 Consequently, highly clustered networks or spatial lattices, which provide continuous local reinforcement, are vastly superior to random networks for sustaining and spreading complex behaviors [cite: 25, 26, 31].



### Threshold Models and the Speed-Resilience Tradeoff

Complex contagions are frequently mathematically modeled using threshold dynamics. In a deterministic threshold model, a susceptible node adopts a behavior only when the specific fraction (or absolute number) of its infected neighbors exceeds a critical threshold parameter, denoted as $\theta$ [cite: 26, 33, 34]. 

Analyzing threshold models reveals a profound, inherent trade-off between a network's baseline resilience and its eventual propagation speed. Rigorous analytical studies demonstrate that clustered networks are more "resilient" to initial outbreaks, requiring a significantly larger initial seed size of activated nodes to initiate a global cascade compared to simple contagions [cite: 35]. However, once the threshold requirement is met and the initial resistance is breached, the contagion spreads with immense speed and unrelenting stability throughout the dense clusters [cite: 26, 35]. 

Furthermore, the statistical inference and algorithmic reconstruction of hidden network pathways depend heavily on identifying the underlying contagion type. Nonparametric Bayesian methods indicate that reconstructing the exact topology of an obscured network is statistically more accurate when observing a complex contagion on a dense graph than observing a simple contagion [cite: 33, 36, 37]. Because the strict threshold requirement actively suppresses spontaneous, noisy pairwise infections, the resulting behavioral activation cascades provide highly structured, deterministic data that map perfectly to the underlying community overlaps, allowing researchers to reverse-engineer the graph structure with high fidelity [cite: 33, 36, 37].

| Propagation Parameter | Simple Contagion Framework | Complex Contagion Framework |
| :--- | :--- | :--- |
| **Baseline Exposure Mechanism** | Contact with a single active node triggers adoption based on probability $p$. | Adoption requires simultaneous or sequential exposure exceeding threshold $\theta$. |
| **Topological Accelerant** | Isolated, bridging weak ties. | Clustered, embedded wide bridges. |
| **Systemic Tradeoff** | Highly infectious initially, but cascade speed decays rapidly in dense clusters. | High initial resistance (resilient), but cascades are rapid and total once triggered. |
| **Network Reconstruction Viability** | Accurate in sparse networks; highly noisy in dense networks due to spontaneous transmission. | Highly accurate in dense networks; threshold mechanics suppress stochastic noise. |

## Mathematical Modeling of Information Spread and Decay

To quantify the exact rates at which information spreads and eventually dies in a socio-technical network, researchers adapt traditional epidemiological compartmental models, translating the mechanics of biological viral transmission into parameters suitable for sociological dissemination.

### The Susceptible-Infectious-Recovered (SIR) Epidemic Framework

The universal baseline for modeling propagation is the Susceptible-Infectious-Recovered (SIR) model [cite: 38, 39]. The bounded population is divided into three mutually exclusive compartments:
1.  **Susceptible ($S$):** Nodes that have not yet received or adopted the information.
2.  **Infectious ($I$):** Nodes actively spreading or broadcasting the information to their neighbors.
3.  **Recovered/Removed ($R$):** Nodes that have permanently ceased spreading the information, either due to a loss of interest, forgetfulness, or the adoption of a superseding, competing narrative.

In the standard mathematical SIR framework, the basic reproduction number ($R_0$) represents the expected number of secondary infections generated by a single infectious node within a totally susceptible population. It is calculated as $R_0 = p / \mu$, where $p$ is the probability of transmission per contact and $1/\mu$ is the expected recovery time [cite: 38]. An outbreak naturally dies when the susceptible population falls below $1/R_0$, a state known as herd immunity [cite: 38]. 

However, the standard SIR model relies on a critical flaw when applied to social dynamics: it assumes uniform, homogeneous mixing, meaning any node can theoretically contact and infect any other node with equal probability [cite: 40, 41]. Applying SIR mechanics to actual social graphs requires Edge-Based Compartmental Modeling (EBCM). EBCM adjusts the foundational differential equations to account for the empirical degree distribution, clustering, and spatial constraints of the network, ensuring that transmission mathematically can only occur along valid, pre-existing edges [cite: 40, 41].

### The SSEIR Hypernetwork Model and Information Decay Mechanics

While the EBCM-adjusted SIR model effectively explains the expansion phase of propagation, it lacks the parameter nuance required to fully explain the rapid, sudden decay and death of digital information. To address this, the SSEIR (Susceptible Active/Inactive - Exposed - Informed - Recovered) hypernetwork model introduces highly specific state transitions tailored to cognitive and algorithmic realities [cite: 42].

The decay of information—the explicit process by which a viral cascade stabilizes, fragments, and dies—is governed by specific mathematical pathways leading to the Recovered ($R$) state [cite: 42]:
*   **From Exposed to Recovered ($E \to R$):** Regulated by the transition probability $\gamma$. This pathway accounts for users who are exposed to the information in their feed but choose never to engage or share it, reflecting immediate disinterest or algorithmic suppression.
*   **From Informed to Recovered ($I \to R$):** Regulated by the transition probability $\varepsilon$. This represents active spreaders who eventually fatigue and stop disseminating the information.
*   **The Forgetting Mechanism:** Regulated by the transition probability $v$. This continuous variable accounts for the natural cognitive decay of human memory over time, organically pushing both exposed and informed users toward the permanent recovered state [cite: 42].

The absolute rate at which information dies out within the system is represented by the differential equation capturing the accumulation of inactive, non-spreading nodes derived from mean field theory:
$$\frac{dR}{dt} = (\varepsilon + v) I + \gamma E$$

Simulations of the SSEIR model reveal profound behavioral insights. As the recovering rate $\varepsilon$ increases, the peak volumetric number of informed users decreases sharply, and the temporal duration required for the active cascade to reach zero diminishes rapidly [cite: 42]. Most notably, the SSEIR model proves that information traversing modern social hypernetworks exhibits a significantly faster decay rate than biological viruses simulated in a standard SIR model. This accelerated death is attributed directly to the presence of the multiple psychological and algorithmic pathways (specifically the direct $E \to R$ transition via $\gamma$) that allow nodes to bypass the active spreading phase entirely [cite: 42].

Furthermore, raw network density acts as a massive accelerant for decay. In highly dense, interconnected graph structures like Twitter, the sheer volume of redundant links means users acquire "immunity" (the $R$-state) much faster than in sparse structures. The information burns through the available susceptible nodes so rapidly that the cascade stabilizes and dies almost immediately after its peak [cite: 42].

## Structural Holes and the Anatomy of Propagation Failure

Beyond mathematical decay rates, information propagation frequently halts entirely due to topological dead ends, structural bottlenecks, or absolute network fragmentation.

According to Ronald Burt's seminal structural hole theory, empirical social networks are not monolithic blocks; they are composed of dense, localized communities separated by vast empty spaces known as structural holes [cite: 43, 44]. Nodes that establish links across these voids are known as structural hole (SH) spanners or brokers [cite: 13, 43]. 

These spanners wield immensely disproportionate control over global information flow. Because they act as the sole conduit between isolated communities, they serve as ultimate gatekeepers [cite: 43]. If an SH spanner chooses not to transmit a piece of information (effectively transitioning to the $R$ state in the SSEIR model upon exposure), the propagation cascade suffers an absolute structural dead end [cite: 43, 45]. The adjacent community remains completely insulated from the contagion, and global propagation fails.

Similarly, the failure or deliberate removal of high-degree hubs in scale-free networks can lead to catastrophic cascading failures. If a network is physically or algorithmically partitioned into disconnected sub-graphs, a phenomenon akin to a "split-brain" architecture occurs [cite: 46]. In this fragmented state, information cannot bridge the void, localized consensus models begin to diverge, and the possibility of a unified global cascade is permanently destroyed [cite: 46, 47]. The vulnerability of network clustering to targeted node failure remains a primary reason why viral cascades often die out geographically or demographically rather than universally [cite: 47].

## Empirical Network Dynamics in Contemporary Digital Platforms

The theoretical models of graph theory, complex contagions, and mathematical decay manifest uniquely across different commercial platforms. The specific architectural constraints and algorithmic curation mechanisms of these platforms fundamentally dictate how information lives and dies within their ecosystems.

### WhatsApp and the Constraints of Closed Social Networks

WhatsApp represents a highly specific topological environment characterized by closed, end-to-end encrypted groups and a heavy reliance on strong-tie, peer-to-peer dissemination. Because the platform lacks a public, algorithmically curated discovery feed, information propagation must occur organically across established edges [cite: 48, 49].

To combat the viral spread of coordinated misinformation—which functions precisely as a highly transmissible simple contagion—WhatsApp implemented artificial, platform-wide structural constraints. The most notable interventions include strictly limiting the ability of a user to forward a message to a maximum of five chats simultaneously, and visually flagging highly viral content with a "Forwarded Many Times" (FMT) label [cite: 48, 50, 51, 52]. 

Extensive empirical network analysis of over 10 million messages across thousands of WhatsApp groups in Brazil and India reveals the complex impact of these architectural constraints [cite: 48, 50, 51]. Statistical analysis shows that harmful messages and political misinformation consistently achieve vastly greater cascade depth and breadth compared to benign content, relying heavily on audio and video modalities to bypass textual scrutiny [cite: 48]. 

While the five-forward limit successfully acts as a temporary friction mechanism—artificially reducing the local out-degree of spreading nodes and causing measurable delays in the speed of information spread—it is ultimately ineffective at halting terminal propagation in large, public-facing political groups [cite: 49, 51]. The research indicates that highly motivated users easily circumvent the architectural friction. By manually routing information across groups, they successfully execute cross-group transmission, allowing the contagion to repeatedly jump across structural holes and sustain the viral cascade well beyond the intended limits of the platform [cite: 49, 51].

### TikTok and the Transition to Algorithmic Interest Graphs

If WhatsApp relies entirely on a traditional, edge-based social graph, TikTok represents a complete paradigm shift toward an algorithmically curated "interest graph." This fundamental shift invalidates many classic network propagation metrics.

Data from the Pew Research Center highlights a massive demographic migration, with platforms like TikTok rapidly capturing youth demographics while traditional social graph platforms age [cite: 53, 54]. Network analysis of TikTok's "stitch" conversational dynamics reveals a topology that diverges completely from traditional human social networks [cite: 55, 56]. 

TikTok user interaction graphs display practically zero reciprocity (users rarely interact back-and-forth) and zero clustering (a user's followers do not interact with each other). Instead, the macro-network is composed of massive, extremely sparse, star-like patterns characterized by extreme degree centralization [cite: 55]. The propagation topology is no longer determined by mutual social relationships, but by unilateral, algorithmic engagement with content.

Furthermore, sophisticated Markov model analyses of TikTok's algorithmic feed demonstrate how rapidly the platform acts as a centralized, omnipotent propagation hub. Using automated "sock puppet" audits to track content delivery, researchers found that the algorithm identifies and drastically amplifies content aligned with a user's specific implicit interests at an unprecedented speed. The algorithm frequently locks a user into a personalized, highly aligned content cluster within the first 200 video views [cite: 57]. 

This algorithmic dynamic bypasses the need for peer-to-peer weak ties or wide bridges entirely. The recommendation algorithm itself acts as a universal structural hole spanner, instantly connecting previously isolated users directly to massive, platform-wide behavioral contagions. It achieves near-instantaneous global reach without relying on the slow, organic routing of user-to-user social networks [cite: 56, 57].

## The Impact of Artificial Intelligence on Network Propagation

The integration of Generative Artificial Intelligence (GenAI), Large Language Models (LLMs), and autonomous agents into social media platforms introduces unprecedented variables into the mechanics of information propagation and decay, fundamentally altering both the structure of the graph and the quality of the nodes.

### Autonomous Agents as Centrality Hubs and Market Manipulators

In traditional, human-centric networks, high-degree hubs are typically celebrities, politicians, or institutional broadcast accounts. However, as AI tools drastically lower the economic and computational cost of content generation and automated interaction, synthetic bot networks are increasingly dominating the network topology [cite: 58]. By 2030, industry projections suggest that up to 90% of total internet traffic could be generated by autonomous bots, necessitating entirely new paradigms for identity and access verification [cite: 58].

Network analyses of bot deployments in contentious social media environments (such as during geopolitical conflicts or coordinated corporate targeting campaigns) reveal highly concerning structural anomalies. Despite comprising a microscopic fraction of the total user population (frequently less than 0.3%), synthetic bots achieve massive, disproportionate structural influence [cite: 20]. Bots are empirically shown to consistently rank at the absolute top of both eigenvector centrality and betweenness centrality distributions [cite: 20]. 

By artificially injecting high volumes of coordinated retweets, synthetic interactions, and targeted replies, bots actively manipulate the algorithmic curation engines. They act as synthetic "wide bridges," forcibly sustaining complex contagions that would otherwise die out organically, and actively preventing natural information decay by continuously resetting the forgetting parameter ($v$) in the SSEIR model [cite: 20, 59]. 

Furthermore, as the global AI economy matures, "agentic" workflows—where AI agents autonomously negotiate, purchase, and transact on behalf of human users—threaten to create a highly consolidated, scale-free market structure [cite: 60]. Through a relentless positive feedback loop of preferential attachment, a few dominant, highly capable AI agents will aggregate massive user bases. This dynamic consolidates bargaining power, allowing dominant agents to form impenetrable hubs that exert monopolistic control over the flow of both information and commerce, bypassing traditional antitrust and network competition metrics [cite: 60].

### Generative Content and the Degradation of Network Quality

While AI agents optimize and manipulate the structural flow of information, the widespread deployment of LLMs fundamentally alters the qualitative nature of the information circulating within the network. This creates a severe paradox between engagement volume and systemic health.

Controlled, large-scale experiments deploying GenAI tools (such as automated chat assistants, algorithmic conversation starters, and AI-generated reply suggestions) in realistic social media environments reveal a complex duality [cite: 61, 62, 63]. On a purely quantitative level, AI assistance significantly increases superficial user engagement, expands average comment length, and drastically inflates the overall volume of generated content. In network terms, this effectively injects massive amounts of new nodes and active edges into the discourse graph [cite: 61, 63]. 

However, this quantitative explosion corresponds with a severe, highly measurable qualitative degradation. Human users consistently perceive AI-assisted conversational environments as lacking authenticity, leading to the rapid accumulation of what researchers term "semantic garbage" [cite: 61]. 

From a strict network propagation perspective, this dynamic artificially accelerates the $\gamma$ and $\varepsilon$ decay parameters within the SSEIR model. While AI initially spikes the volume of the active spreading state ($I$), the perceived loss of human authenticity triggers a massive negative spill-over effect. Human users, overwhelmed by the volume of algorithmic noise and synthetic interactions, lose interest and transition to the recovered/inactive state ($R$) at an accelerated rate [cite: 42, 61, 62]. 

Consequently, human-to-human interaction cascades suffer premature death. Research by the Oxford Internet Institute highlights that this AI-driven oversaturation—combined with the threat of computational propaganda and algorithmically generated disinformation—presents a critical threat to democratic information ecosystems [cite: 64, 65, 66]. Ultimately, the network becomes structurally saturated with synthetic content, but functionally dead regarding the genuine transmission and adoption of complex human contagions.

## Sources

1. [www.mdpi.com](https://www.mdpi.com/1099-4300/26/11/957)
2. [www.nas.ewi.tudelft.nl](https://www.nas.ewi.tudelft.nl/people/Piet/papers/TUDreport20111111_MetricList.pdf)
3. [noduslabs.com](https://noduslabs.com/radar/types-networks-random-small-world-scale-free/)
4. [journals.plos.org](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0228728)
5. [www.ijmsi.org](https://www.ijmsi.org/Papers/Volume.5.Issue.7/B05071116.pdf)
6. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC2323569/)
7. [en.wikipedia.org](https://en.wikipedia.org/wiki/Complex_contagion)
8. [mobilizingideas.wordpress.com](https://mobilizingideas.wordpress.com/2019/07/29/how-behavior-spreads-the-science-of-complex-contagions-a-review/)
9. [ndg.asc.upenn.edu](https://ndg.asc.upenn.edu/wp-content/uploads/2022/12/Complexcontagions.pdf)
10. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC11014438/)
11. [economics.mit.edu](https://economics.mit.edu/sites/default/files/inline-files/Speed_Resilience_3_Dec.pdf)
12. [www.lib.uidaho.edu](https://www.lib.uidaho.edu/digital/turning/pdf/Kuhn's%20Paradigm%20Shifts.pdf)
13. [www.ebsco.com](https://www.ebsco.com/research-starters/literature-and-writing/kuhn-explores-paradigm-shifts-scientific-thought)
14. [en.wikipedia.org](https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Revolutions)
15. [philosophyalevel.com](https://philosophyalevel.com/posts/structure-of-scientific-revolutions-thomas-kuhn/)
16. [internationalpolicybrief.org](https://internationalpolicybrief.org/wp-content/uploads/2023/10/ARTICLE8-95.pdf)
17. [greenice.net](https://greenice.net/ai-agent-development-trends/)
18. [www.youtube.com](https://www.youtube.com/watch?v=TXVYErhXsXk)
19. [www.cognizant.com](https://www.cognizant.com/us/en/ai-lab/blog/ai-research-updates-may-2026)
20. [www.youtube.com](https://www.youtube.com/watch?v=DwFD_ffS21E)
21. [www.youtube.com](https://www.youtube.com/watch?v=F1Ihel8Dgqs)
22. [www.researchgate.net](https://www.researchgate.net/publication/381058794_Don't_Break_the_Chain_Measuring_Message_Forwarding_on_WhatsApp)
23. [arxiv.org](https://arxiv.org/html/2505.18099v3)
24. [ojs.aaai.org](https://ojs.aaai.org/index.php/ICWSM/article/view/31372)
25. [www.niemanlab.org](https://www.niemanlab.org/2019/01/whatsapp-limits-message-forwarding-in-order-to-fight-misinformation-and-rumors/)
26. [homepages.dcc.ufmg.br](https://homepages.dcc.ufmg.br/~fabricio/download/complexnet2019whatsapp.pdf)
27. [arxiv.org](https://arxiv.org/html/2403.00025v4)
28. [gtf.fyi](https://gtf.fyi/papers/ovi2025assessing.pdf)
29. [www.pymnts.com](https://www.pymnts.com/news/artificial-intelligence/2025/generative-ai-tools-lower-cost-increase-frequency-bot-attacks-social-media/)
30. [www.researchgate.net](https://www.researchgate.net/publication/390901193_Balancing_trust_and_distrust_in_generative_AI_chatbot_adoption_a_case_study_from_China)
31. [ai.wharton.upenn.edu](https://ai.wharton.upenn.edu/business-generative-ai-conference-2025/speakers/)
32. [en.wikipedia.org](https://en.wikipedia.org/wiki/Small-world_network)
33. [snap-stanford.github.io](https://snap-stanford.github.io/cs224w-notes/preliminaries/measuring-networks-random-graphs)
34. [arxiv.org](https://arxiv.org/html/2403.13215v1)
35. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC2323569/)
36. [www.jsums.edu](https://www.jsums.edu/nmeghanathan/files/2016/01/CSC641-Sp2016-Module-5-TheoreticalNetworkModels.pdf)
37. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC12168247/)
38. [arxiv.org](https://arxiv.org/html/2405.00129v1)
39. [arxiv.org](https://arxiv.org/html/2301.09407v2)
40. [sociopatterns.org](https://sociopatterns.org/assets/papers/PhysRevLett.130.247401.pdf)
41. [economics.mit.edu](https://economics.mit.edu/sites/default/files/inline-files/Speed_Resilience_3_Dec.pdf)
42. [en.wikipedia.org](https://en.wikipedia.org/wiki/Scale-free_network)
43. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC6399239/)
44. [www.math.cmu.edu](https://www.math.cmu.edu/~af1p/Teaching/INFONET/Papers/PowerLaw/dirweb.pdf)
45. [www.stat.berkeley.edu](https://www.stat.berkeley.edu/users/aldous/Networks/boll1.pdf)
46. [hons-web.github.io](https://hons-web.github.io/online/)
47. [sites.duke.edu](https://sites.duke.edu/religion361_01_s2020/2020/04/28/the-math-behind-the-sir-model/)
48. [higherlogicdownload.s3.amazonaws.com](https://higherlogicdownload.s3.amazonaws.com/APS/fc77d112-850f-4db6-997c-e434bd4670b5/UploadedImages/Mickens.pdf)
49. [journals.plos.org](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0246961)
50. [academic.oup.com](https://academic.oup.com/comnet/article/6/6/948/4931473)
51. [www.clevelandfed.org](https://www.clevelandfed.org/publications/economic-commentary/2020/ec-202023-network-enhanced-sir-models)
52. [d-nb.info](https://d-nb.info/1202108008/34)
53. [www.researchgate.net](https://www.researchgate.net/figure/Node-betweenness-centrality-for-node-a_fig3_316944113)
54. [www.csusb.edu](https://www.csusb.edu/sites/default/files/upload/file/2024/Securing_Network_Resilience_Leveraging_Node_Centra%20%282%29_1.pdf)
55. [arxiv.org](https://arxiv.org/pdf/2402.12623)
56. [www.researchgate.net](https://www.researchgate.net/figure/Changes-in-Node-Betweenness-Centrality-and-Eigenvector-Centrality-A-Changes-in-the-top_fig4_388426569)
57. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC12168247/)
58. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC11192313/)
59. [arxiv.org](https://arxiv.org/html/2301.09407v2)
60. [economics.mit.edu](https://economics.mit.edu/sites/default/files/inline-files/Speed_Resilience_3_Dec.pdf)
61. [nwlandry.com](https://nwlandry.com/assets/docs/publications/2024_PhysRevE_reconstructing.pdf)
62. [snap-stanford.github.io](https://snap-stanford.github.io/cs224w-notes/preliminaries/measuring-networks-random-graphs)
63. [braph.org](http://braph.org/braph-1-0/manual/graph-measures/)
64. [memgraph.com](https://memgraph.com/blog/graph-clustering-algorithms-usage-comparison)
65. [www.nas.ewi.tudelft.nl](https://www.nas.ewi.tudelft.nl/people/Piet/papers/TUDreport20111111_MetricList.pdf)
66. [docs.lib.purdue.edu](https://docs.lib.purdue.edu/cgi/viewcontent.cgi?article=2749&context=cstech)
67. [www.researchgate.net](https://www.researchgate.net/publication/350486348_Structural_Hole_Theory_in_Social_Network_Analysis_A_Review)
68. [www.homepages.ucl.ac.uk](https://www.homepages.ucl.ac.uk/~uctpcab/confer/goyal.pdf)
69. [www.kellogg.northwestern.edu](https://www.kellogg.northwestern.edu/research/math/seminars/200405/Goyal042705.pdf)
70. [evelyn0414.github.io](https://evelyn0414.github.io/file/Structural_Hole_Theory_in_Social_Network_Analysis_A_Review.pdf)
71. [networksciencebook.com](https://networksciencebook.com/chapter/8)
72. [wilte.wordpress.com](https://wilte.wordpress.com/2018/10/17/how-behavior-spreads-the-science-of-complex-contagions-damon-centola/)
73. [en.wikipedia.org](https://en.wikipedia.org/wiki/Complex_contagion)
74. [news.mit.edu](https://news.mit.edu/2010/social-networks-health-0903)
75. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC11014438/)
76. [www.gsb.stanford.edu](https://www.gsb.stanford.edu/faculty-research/publications/topological-measures-identifying-predicting-spread-complex-contagions)
77. [passer.garmian.edu.krd](https://passer.garmian.edu.krd/article_158076.html)
78. [www.researchgate.net](https://www.researchgate.net/publication/282831252_A_comprehensive_comparison_of_graph_theory_metrics_for_social_networks)
79. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC10280625/)
80. [arxiv.org](https://arxiv.org/pdf/2312.14306)
81. [arxiv.org](https://arxiv.org/pdf/2502.12519)
82. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC13002914/)
83. [scholarspace.manoa.hawaii.edu](https://scholarspace.manoa.hawaii.edu/bitstreams/5b4fc3b9-3982-4baf-86bd-0b122e122188/download)
84. [www.mdpi.com](https://www.mdpi.com/1099-4300/26/6/486)
85. [arxiv.org](https://arxiv.org/abs/2506.14295)
86. [www.researchgate.net](https://www.researchgate.net/publication/400870282_The_impact_of_generative_AI_on_social_media_an_experimental_study)
87. [en.wikipedia.org](https://en.wikipedia.org/wiki/Complex_contagion)
88. [applied-complexity.org](https://applied-complexity.org/docs/NATCOMMS-2022.pdf)
89. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC11014438/)
90. [arxiv.org](https://arxiv.org/html/2301.09407v2)
91. [economics.mit.edu](https://economics.mit.edu/sites/default/files/inline-files/Speed_Resilience_3_Dec.pdf)
92. [www.google.com](https://www.google.com/search?q=time+in+Provincia+di+Salerno,+IT)
93. [arxiv.org](https://arxiv.org/pdf/2502.18661)
94. [www.researchgate.net](https://www.researchgate.net/publication/387241110_TikTok_and_the_algorithmic_transformation_of_social_media_publics_From_social_networks_to_social_interest_clusters)
95. [informatika.stei.itb.ac.id](https://informatika.stei.itb.ac.id/~rinaldi.munir/Matdis/2024-2025/Makalah/Makalah-IF1220-Matdis-2024%20(88).pdf)
96. [arxiv.org](https://arxiv.org/html/2503.20231v1)
97. [truthout.org](https://truthout.org/articles/tiktok-algorithm-favored-conservative-content-in-2024-potus-race-study-suggests/)
98. [users.ece.cmu.edu](https://users.ece.cmu.edu/~oyagan/Journals/Clustered_Multi.pdf)
99. [pmc.ncbi.nlm.nih.gov](https://pmc.ncbi.nlm.nih.gov/articles/PMC4770194/)
100. [web.cs.ucla.edu](http://web.cs.ucla.edu/classes/spring16/cs111/supp/clusterconcepts.html)
101. [www.math.fsu.edu](https://www.math.fsu.edu/~bertram/lectures/Closure_clustering.pdf)
102. [www.researchgate.net](https://www.researchgate.net/publication/313157204_Vulnerability_of_Clustering_under_Node_Failure_in_Complex_Networks)
103. [resources.geant.org](https://resources.geant.org/wp-content/uploads/2025/10/GN5-2-White-Paper_Gen-AI-in-Network-Management.pdf)
104. [www.sequencr.ai](https://www.sequencr.ai/insights/key-generative-ai-statistics-and-trends-for-2025)
105. [www.networklawreview.org](https://www.networklawreview.org/ai-agents-network-effects/)
106. [intuitionlabs.ai](https://intuitionlabs.ai/articles/latest-ai-research-trends-2025)
107. [www.researchgate.net](https://www.researchgate.net/publication/391276537_Research_on_the_impact_of_generative_artificial_intelligence_GenAI_on_enterprise_innovation_performance_a_knowledge_management_perspective)
108. [www.campaignnow.com](https://www.campaignnow.com/blog/what-pews-latest-social-media-report-means-for-2026-campaign-strategy)
109. [www.pewresearch.org](https://www.pewresearch.org/internet/2022/03/17/mixed-views-about-social-media-companies-using-algorithms-to-find-false-information/)
110. [www.pewresearch.org](https://www.pewresearch.org/journalism/fact-sheet/social-media-and-news-fact-sheet/)
111. [www.pewresearch.org](https://www.pewresearch.org/topic/internet-technology/platforms-services/social-media/)
112. [www.pewresearch.org](https://www.pewresearch.org/internet/2025/11/20/americans-social-media-use-2025/)
113. [www.google.com](https://www.google.com/search?q=time+in+Oxfordshire,+GB)
114. [www.researchgate.net](https://www.researchgate.net/publication/395484792_Computational_Propaganda_Disinformation_And_Democracy_Multidisciplinary_Strategies_For_2025)
115. [www.oii.ox.ac.uk](https://www.oii.ox.ac.uk/news-events/reports/)
116. [www.oii.ox.ac.uk](https://www.oii.ox.ac.uk/research/research-impact/)
117. [committees.parliament.uk](https://committees.parliament.uk/publications/48745/documents/258221/default/)
118. [www.oii.ox.ac.uk](https://www.oii.ox.ac.uk/news-events/oii_tag/media/)

**Sources:**
1. [uidaho.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFTrvD_wNw3SCW1KDBrjH_iKuUWzb50F4vBjehd2svbzhHsg2ZDcz_iy1gy0uCBKkr802VwnUUxeC_FPv6jFj_jrUKUjPdvPHHlje76iuPvIlc39LGWqt9NwkPJz0U7tHzCrhW1qrDa2bzw1qZRHAU-dI20gHN6kk1-Oyh3WDxYgzDDvw==)
2. [ebsco.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFGQuOjTtS8RnxxykzApMHFowHOBPmW3CJ5zo9zRigcD9e9e4lOp_xmwTyHqjw3-h-TWfvoK7T-_l0gbmzmba6z-hSkbH34UxfJf2hQjPaQoahT7QKxf3oa-TsD9K6V9MdY13yfdzKi1GzGQY_HkvX_TQKXR0H9ysFE_C4sOB8gRy_0MrVh9wIslD0fIsdbF7KY5c2x-6tctQsTLDh7Y36pI6tI1X0=)
3. [wikipedia.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEf2Uoxnn13ZQBCpQIcCOdLH-8h1td_sI5okUdqrv1GLTC3Ac9xPVJcwHDjS3vdSa_PEV3xpSIQGbhF7zToPIUoyN9vK0zxr5jpbAGH3jf3DXxmGRpx1vTZwtjNQLP2tXRKsRMLQq6q0LJHvoMbcRE95NeIqtE3b8rBe0k=)
4. [philosophyalevel.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGc25cYdBWUdP88UO9f7afsvttwhV237rWU9zJwuZSISKCj6Aub3JokfLnL9o1kR8nM0NUgXpOqNgHmhkBHu0kWAfLVe-fiZPqXhRZ08k_Rh27aTi3ElWWqJv_iFF5lWIZ_FcYJDDfdb0zpT0Ra6hDwmiEbDaPV538WNcFkd845Arkbqf_Faq1GWg==)
5. [internationalpolicybrief.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEeRWZUvTdYkwG_-QO3FuKsL_yHyWOiGi0uyKmtp0HmFhDX_OcTH9ApQgBbb9Y226Cqur1fgtzSaraLbklrKK81ufPeYQhJlhX35jflCBY_KHuilKhCjhxxiyZp8ryry3wdnf_CNLbsCn2FtqmbcuRQXiqwzoVy_2wGGHkrN886WLObHuS5)
6. [tudelft.nl](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF1HlcYLyjozwM3ugULnKCvCSUmu5R5QTpVHcjMvHtfEjiQcz91IkUPDujwkkdigV8_U8pj-CnrklaYkxzmG0KDVPSolYjsruzTgSANVzzLxurqBm3rwLhZtG1IBU6B_Az5Zxd67yaYbPra0Z1dPme5IkmQlnKinQ5lASZD7zyXpNXIHzMbXsJV)
7. [plos.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFtFnYa0Ou6oOmT0F1tQ9c8kRq636QQtynOcG7QxqhAtUDtx2TnU8CA-8TqtRRKLR9vj49gIjEMPyC-3JRe_jeYOj-IoHN_sW3D_6jbrzqaFISVFaTfsnWv2eoq0owgNm2pHJuwBkrlEjaYb9dSyfR93IFH6QRGhz_67KJ3HWZx)
8. [noduslabs.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEuDykVZsskR4xM6HKpF50__pd-8GXZd13GH_uX6dOQD-dcIG7SiVBkt1qS2rlsjP8viC1aPfi4Y1VwXhN4dZShnNotUzKLlgF9o7cDl__Fhw-7Be6A8utHTX0DuP2tOCF6C24hVijsYPyjGyfbzK7yLufTzsuwNI3YMRbF0gAS)
9. [jsums.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFjtoO8vtlSYunmq6gn9_5jqZXXlBLaT5xRB8Lyoxiymsl2rFlsf4s0uhOAQjteb52_pN1uQnRS9HUsItVDGfVeOBzugR6_kttVAtY5FHrAD8PNBwey_v_JzNiLQPkt1dyM4Qco32dBY83VOiGwQIt3q4GwiiB6R4OtnHaI2_wAG8YJzjCbxfMsT7nTqRmTBqnu7h3rli3aVbGf)
10. [berkeley.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFKF8Q9lw6PznKb_JLUxn-ZvEV4epXfsKB4-2nZ0_eRR-Zzq97g1JUtPQmnEA2UAqelfMyCqhtD01p2IijU-xtqhgT1HKiCXgVAGSKkW3_Aj5x2Z-vuzb8YcKfY3XFAcMC2xXW76KqypKSbYNMpqJrY0pgs)
11. [github.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHvOxqG_RkyRsVPsJSZlkbaBtwZ-9bX8OUETy3khMNSk8elrVlHn-7XFW8RFo9JbSxUAq3yfMHgXnFDrY6Lw_mc1Y_fyzylJSAFryGaJe8TS7svQtOEDx6CMBtjt3xkTimWczGI3y5OPVbGbBJIbvm9E2VccwwI43ovNeAV4_p7VihoUa9Nga4mU1ccZr7RxJ-j)
12. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEOUxMCIrEPMj1yEuUaqO9eECwfkk_PiBzHAQ30AtYjV7gc7EKse-OjHjGK6_Z6gqLYWr-Cx2Iz0AGxNNafu118cCYubptzVILUpcxBauEWa0WHQA22c3TdeQ==)
13. [fsu.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEa3tHJcbfhKXU8xQsnbVOBmlwu7qgox4ASJnQsp-DbvZ94lrI3mJH-XLiF_AwlL59MzHdFys95ssYEjKcrSIfuJlfyzGUCEIwe_eDJxkzuQ7NK3abc93Muum_bYyepq3zuevLpRFfkZckRr5NX1v4X3jGt_mQL_g==)
14. [wikipedia.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHrh0i_RjTIG735BAoc9Jv2byaYaVgedcAiLbCmKU2cl0FF6B3LY33T3CViXQ24BzqeyKuELX0kG6gGfscs_G-Z1oi3q4oKnpOBDZtmwXAkOdG3BDUZkWgZt0MQyWOONltOX-W63UWw)
15. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHVJFo3w26MQA36JdP3sqJHi3avSu_CA2BBx9-l7UBHAWvP205KL9Z2EYH1o0bhW10t6igKFzGhUm8mlywd29MFZ4Pd4k0tAL074WefnsW31hXO8VdVbYQPfWRBQAj6C0YcDS8vNe_5)
16. [wikipedia.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFBJO2FRDP2Ugy6ZSTCdI5OQpLvsCXImIr1uwxxHgOS8nVfon5dJGqjYRxUUofRzSispwDzYdkcpzLO5E9il_cToDfGjT-3-HDuzHOqC_aYaHzx8F68lytFGISqZ6VGaFR2F35FkNQ=)
17. [networksciencebook.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQESyNkTIm4gpjwfPXWO2bobx5AYR7W3cMHZ2HOpQNEYhd_76dngyI-L2DuCSQqCPKjEZNyqTZtZuuEyhve_luD9OP1dO3v_bF8V1K-nNxNTUgZLVpx_uFjmZnZBwr3h)
18. [ijmsi.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFaNQOJ9h9VO0EAc2Lhcz8T56kK42FdQQ6hDs4Ky2iksADFbcV-ocOxosfGd60I2CeV_m_TtlHehL-3bndwi3XLtgjBgb2qLUtB1OEraJ2OIS1kC1p2Uqq78mEHmWydgysdGuiLnW-eNnzsw9OKZkc1qQ==)
19. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGx5vc4-zd4BXmqrfBSZJXOyJTS3XvcLf_IF_CO6FU4NkHcDvJiWQkPfa4WrmW1iUfoqYYdiYOwvF6lnq2RdfeVe_mfJ1nZwNTdUaCpeJ_XMT98IugKfeM9QZWohzosTWzNCNMwhodD)
20. [d-nb.info](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGYVJoxH5htiLXdTV7pzud8ewRqlvs1f2LXDpOOARLc31OcoQIudLt3ehkYlpCy6qcdhWM63QJ6COxr9WQiOI-pPt5LKbv2MwT8oqZu3sAYEsRU4lh7)
21. [mdpi.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH8xhU5h2O_7NEfoxRwWWs5WL4mYBx-P0odNT-qGhGboMWAmme1S_R0so-gqlKP6WC8l1OrwOOw8NNPPEVVe37OKHtd8_FHNihx2ieLuBBlsXtUNLwyQhKCcAnfYFI=)
22. [csusb.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE5FStAGsohg8CL8f23VhIM_Mkow7-I9yt2VzVMVFzuynj9yz1nbCCUdgKS9AITsHbsNf6wNfsxUd4HeLcxfnoXx794EKN0OuZvriDAy56lxbpGFD_LsPrKgm1JlvO3zHiv1XxKha1CDvrQFQ6j8W9jQoe9YUIigY9FK475FNzXI7In9P8RwUmuOsf46G1KQcR2-DvhudGxIt_KwbzJ0r8pXpUnj7Ds805YuaUZPeQAyu4cZA==)
23. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGcuLnk00BJg5RhE7RbEpSQtbcj3ep5lz-jiRzhJ2Dc1Fro5mp0Zj_quAPRtE0FYVvkrZSkE3cHje7-tzxEMsBUTkJ0VD7C-pLKDlNT7KXVyAC1KOwZAQ==)
24. [wikipedia.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEGdNQtNYSYK6HODI1ZTVR4S_7UDfQWVSPKd_dZ4uqTteEOsYBJ-RqzfXYzn9DER5ec3EarNg1eTHzS4f_KXZ1J-024Z5LXUbupmmboh6JlZ3flPpUVVfkf--jCtn5_vG7ZwBkepw==)
25. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFRARKRfJdbLEMcWiid5jqim0zOby75tst5IurWBSbT9x96xjhZ3qRg_esS6Q4LwkFTN1NXr3Na1o43a8yyscmhUUO4-1RMT4931Ug6M7bJzfZuehj-yBIRth4FN_ALfFLRya9x1ZdNjg==)
26. [applied-complexity.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF6vSKHhz8tJnUYmfqZlzDcSNlj4FgCCAIE3v1-bTxIST0eE9okvShv0VfFiqri_g5kACXLCDKYYzQGhsbKHD6n4y0fSJtlP9H9eMwimQ025cseoADFgioSb_Ec2pTLB7jZ55v1xBP7EBYG-w==)
27. [upenn.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGafiEnmNMnCEgPw3diRdDXVNFLqis1sfonzcJ_jdEwuNiL2vc9UmpYSmw9N9dMCmOqOVNzCERtarDVheTLDOeSH7HXb_bYuYvzeIo8ujAodcfpSEeyCJ78bYUcEsV-aSnebs0BoD6M9ICPnnInRB6tz8stYn18Pmhy2SuNCN0ULA==)
28. [sociopatterns.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH_6YbUY2MgyxJDUlB248IgXB6ARI7pD8XCmBggTn6Z_ddFuAvenHCcfk1wAPempgoJjqEn6OEa78uNO9uodD_PA46FViSHupCLTpSG90rzDZ0gz5UZhRFjsQlglmKbZCHCXKwfZEvKAo2KoB-RT6a_FJgLMqRC2tw=)
29. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGSKCAYOi0dEVYJXfm0jOCL82YgjTuVQCINop5rk3RFym7PUBsL_X3q4-YiirKh1Vvg5KhEuw-kFvvvIjrcthHvXHOKa8mOowN7YfvuOSukgGvtqd5deZ29XSgQsosP41v6mmngYeQciQ==)
30. [wordpress.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG4w-BadhdInvPZnt26y1vmgP47zh4HmGQ6KdoEUxEbZs0NfU6LiuzI3vTWxC_op41InirQXpg9re_F-dD0RHt-PC1MNhxczwCBX_cQky-T3EfWfiaXUKcmHrDAX2qhyalwKzi6uJBAa-0Jx5esinREYm1AGKLRk1HaNfMK_79t9-ocUPbN3jgBUfj3S8tMMNbr3d-TrvED9o1d_2zb8Zcfqs4=)
31. [wordpress.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFfSlrztFiJKdjTOer-pqXA7Sb06Avzw0VRKWMr_cNRswweWzQw3eJmetYw81p3pXYG-bO2yggvezLxA-FLKEVEEQ714pD_0VeGozuaklocTqdUGGl5nthojpI54Al9JrF2CEMcDCXxJ3DeCksN2l0bHAbd6o7pkdS-AKogWxzdSIliVhOHRzkmfh5pQw__bAtiturbDvbaDzFlmJLIjm0zW3yuAkVQoA==)
32. [mit.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHID3dfMv8bpmW5oghj3mlxXRFsU0pgFHCVE8zCMi1HwN6KrcICg5yxR5CCtCEuF99MbFIynEGXbeUBEjDVt6GIek_6xvKUwL8Q80Iwuvg1IaLNCJMhubivnQNdm0nSazBNpY8dtBR8hEFLJQ==)
33. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHYUpRDeKBhphhOFREzzltZmwhv9WfekNT2_12810JY6DP05-4lSf6yXsnMaOYin3_e597_uEjChHWILQYE9nPdmdc2fu6KsAanztJsEPmJQoT2zKthPrcMlE5gWrNvtwnb5sbYYPgqLg==)
34. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFZwoTdLjdLELWjcx6cMzU_ufRS42UGWTN6io0Hu--Mwnf2OODmZMssSDHhTknmIl5cetMsisVXLB7Ha1wU4iGI94d-fN8uuSfaLTcy3_kML0xcKrCqFTaRcg==)
35. [mit.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFt-tD21QhZsOsuSjo4aC6xh9VXlWOCwSe5IJo_kxmMBwiNFQNOWU80MVVEFH5Ff1ZEw4PtmhJKmMYvmEqqvKT4wPnJiyGrkuUg_t-2iEVHA9DULl-CBN5tRo3EAOMSsElDYi7Y7SfnxdH66SR6azNkqbR2CSymgKiZUo-6bs_ijFugHj-KoVtiNKym)
36. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHeRsk5MOHakOYHPKVDq_sLuGGuToOjYPw7nnn7NK3PDRUoydaklsnnV-O0420OG3t01HxE6yZh29SEMH982gFrdtttRPi-54i4ndZD0e-WTnJCHwzIGKeDvA==)
37. [nwlandry.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGbgyvIBo2CascGrIMTUe-XR0BjKiS4aEXyHgHnTTmIiUjn40CrxWctm8wLYr-19YZ2vvwt6WyU6tGPJsbOUr_b7TIjwSR7FESYJ3Oe0xNshiXZxBuBUeDyenhweBFL2l1p6Z_rVo3DppbxisVJzpPNqw_qH8VBk9NnFHVgPoHUDPQsnB0=)
38. [duke.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHWWk2Re-iwDplESQwIAazzscVbQKSxnTNJo-7vbKmjNIsGZYLN9jwgEjBnhkT5XvF9YsCowUWvArg2Hv4ctP6nEVcZAvk0-S-B7cW4hkemELRXiu0Y89g5HXi-ze407djRiKg6FoAr8sHD6S8Q9YwxckA0fFlvx6LNKX3oSJQNNwhkLSbqNQlNEv8h)
39. [plos.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGNmtNdsoCGBfWDMBEze8NjtPF2ahzR0lxbF4ACiNd1n80uai5dVdT87E8mUA7yfPzIfxiDiIjplDFfSrchVik231sd-MLP6OI7GGlEPdKFRjSMKWStWxWUdigfuYRC7oHRU1aN7Ul0HxVwExTNKQVWmI6QX2P0j9HaMcnzqsEh)
40. [oup.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFCQ-9SXu_rCPZ94aedJgzxbfro5m83fOPiww-JqVba8WIM5O72GOa73uKGveXYi-ITNjFGFJwUxYXE2MWn8Cvy2Eda1vSGMOJqepsDk24MC0N1BIoWI7AEmsYFxi4TW89CV7_9F6JJ2zwog5iy)
41. [clevelandfed.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF9ri0oV6MxPMNQj8A7RJCwW4TQJK52HTRRIjtkvHXdL2DDTPHNoX3kaEkJ6CeldIHj8qlXM_sJrSkDWHfcAC2bfXMyJ5kWQY4d9UlZArHk0uU-zcLT-YLCxUuAcfwxZREmNLGssyQURokPpEG4zLdtg47M_-okJF_gz3gLPDD0fP9dejsZKQc25je9eIf0Sw-FOMDoWffXeFQQ5zyScQ==)
42. [Link](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG1FRoz4CVTVtNOrqNnz0iwrV98FO5FXs9p2l3gUtDJxOezz82qMAwWwy4BCNap6Zx5GHSqRJiirreGEwmxv-nCpwwFwHgwB2p--zWz1uLl0Mo-3Wf9hjhA2QgWm6Xq)
43. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGgDTgOQTEcLQDCG4tERFgL3NnzZ23alf_uDB8iX5r-u6GWkcttHlXzKnVyy_TZtfRKmK9GsHhx9L8NQVQvnaGPeM2X--Olg9GrK5mhwbHUdnk5pPmY4YJ7txKDCr7FHjDgTzmOYo5TCGBCCx-63EAaCKNE3URXTfk6M_KFM_m2QdAZuTiAD36htwwfF7kl8xszfrBSGoScfEsYTW33JyZhi9Hy)
44. [github.io](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGIsnSkYlhgz_2yXSzltbu3NIf08SA4pXP71_eaX0B2JKYonfzgWLY43BsRNyO_-weo2z-oIFDE4r27H6NX_LYalbtjvlHOfBav0VMhPwTL23ZjqqFXxrzOd5GS8hIehm94t0N1LUlj3HuJuL7Xz5nQIwoeLvvavwvhho2jiL9AFV-oh5NSJKza9dUhjUHbFtgp3nBMsP0=)
45. [ucl.ac.uk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGq0Ts9t2sIuwoZnh9B7sm7V4eLRi97uZ_OAMNoYESSTaEcrM_F33s4ffdEo1pPePiiZwwFuuAcpcqiFqI7r3UGie9GvQ4pnwxJdp4tqexvRcT_TzN7ElKogTEZrQ_cgB7CAEhi__ePSurCV8NlCyA=)
46. [ucla.edu](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGa6VF7jT4oBY-etojQTj4QGb2qwATtf0DpX8rEImQSYD-jPvoKJR7lja8tl7gm8m1w7neflzsviUC85YMkYednJ09HOP0r8daBXz2YpKXeF_pMaS60lFFTEIC8XT7FN5KCq_5KTEEMB7NJVo3CVGOuTmg-c_ARqa5nPoPR1g==)
47. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEOY0MZYIEEzhLv0ytMAcGL60p_44KOxP97UI9CxG969AQQeghIHkgtmr7TaIkaM00QeV1qdQPoC1Cvm2QcN12aT8tqqQ_8XWyjGT_BovMaXAh8FoGEtnHTN14UhrXbi_Xc39IDwtU_yZoPpEYDV3wcb6fDD9LBz6u8ufmAlA65qWJVVtPC-VYNQ8SuN8P9iMcM480ScCG1xkD02OYrjvzEGaqT4TOIdoAkK2w=)
48. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEjYF4vMxAaYDY1cEzPwLH4mcH2SD5mWlTxBk0P9XdlYWpT7V_VTee2Cd6vz_4UDzK-MBbHgRnuANkqDDC4em8L1DiI-jHXEKSr27RC0HD_C0F4_5SHQ0H-Kw==)
49. [ufmg.br](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHPqvlXfqdnTavRQeqqNSqHw2Q_RvlZZlwKF6Jb76k3PYp7lebpJjb3-6NYwF-WTHp2alSoq4impSPYZefiDk8IM_lJXwGOgwWtEYPOOv45syBpG1sg_h-gU6mx2yWC7rymG1UqUkikGn6eyIdNVeldTOZNlaj7iF9pLOoSW5BiAsE=)
50. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFXaKY8pS8_8KIVMsskgDkMeN1oBlRc33y8dLo4SSyg_OpuylSfavHumm7td9lKpG65MAI757Ja-8PspEWMbdOF9-16fuuByfQ96ohRiloxgODiwBZVLNbMiGjBz0SiBnbhVI9wRpEtdLg7Xq5wXfY1y3x_-8iRS3ZFOyLb7Ydl3MtwMFHjiWGMQTm9RbyszX3cWoxJE9d1lRKqDA_XN-uh4fN-5gqNJw==)
51. [aaai.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEX2BshLmXUmN5osRxiOrtrrXwtsBtc5bImCZtWCiiEz-FpWLjkkqStbAiH-65o_A9gw9TJWjhuSPLUkqzxFwbLJOT47NCHPjsCtmGbV0QdImdqrBfgJ6jOeCFXuRG4-OEDavsFKDItJ_J8oiyR)
52. [niemanlab.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGZT6xRwZai-tCklL_L_bqqkdtlkvnsd3i-rCfBCQ6yL5eIwhrCLRhNomzTIy9QKBk8ZNPIwthTV04SbIgLV8jMKUe87x0yi0NoV4zMdMKTxtt4elcD14pLfq0N7DA0ILJu-7TyT0pGslEerYQEyD1ltQI11p2fP9SayF-jn_3ZiMdHxKeUvWVA1HmOkl3219W5VcgFMTodwBN3Z5MaVlsAeqcUAU95uQ==)
53. [campaignnow.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQG2AjkDJq0gRge4MF3XE-chPNz_FpkV14YFnYTkzl2R1WfKiqc_Kz0yTPljk1_lvbS8XG0KtKTkS3L576RlDy3n5eFoiprdVecRtNLUJk2jVZS2WFSFajxcQ0QW2N3axqpznw4CqxK3ejBBPgMxrNlmcGiuf0QrI8Mi9e3CNgrQmBKKMxO2IWbhbS3tdQOAM3UsEd8n59VMQGuTeGE=)
54. [pewresearch.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHbPxbCov0DOjZ7gl-EkXFHtZnxLNIJqXR462UqDtuetvCoTQjlHAFiKVUan9luqcZ4kQIrmAn62vlnzcr8asOiRBJGt7xy3mCE0qldiLehRnRhkAMhUcgFHare7cGPl94sSREFDnZ9ip602Z1C5FCWqEtVJPEIniUeqlf20WWR-SGGoDODyg==)
55. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHet_HO0KW-CciXXObgT3ZGSX7qUM36OUwxfBE6lhBoOClS0-PYzCPzVbtPzndqhj-X5ol2SlQZvq961r1rR5XRfH_E6TYwER2WdzsVcvpdOJiyotXM5w==)
56. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHX4oXE_5D8RkfssXfg1nSg7rLCHuQDT2ABmcmR4Q4H2qOEYQHW8pyNClHmhqRH_8CCzakEHwkbsRBYKUfUrmpKPINfWXYIOFAnhj5zCalbLjJgrpI8N4M1dwiAxT4Ltf6hP7608lhXbA_fCiOet6_5USTJBNQGBjGQ5iDZZ2Ajhzd1Oc1j2nXygW6X7Xa8yc0TMnAvxmCqDvNOz4SYN4ulci-iFSaque8ExwLhjzmQZ04yG35OGvj150ZG0-2zFZ4cfeBEddxPhxayRzu4jdkPnTIUYI8puLsjunlg)
57. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFPKZIIj_oxQjrPfbutLJOUXNfC2yc84NTOnvojsLPg8_EXT6dJyUsDBXZZyA1FYIddSMw174VmleGGLSNdtJIAz5XAbU2_34ZqJFGAlFecTtAHptuY83rNDg==)
58. [pymnts.com](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEWim6XIo8r6o2QdIZ0VHhU6Bn6DsTnbdcWl-gj6gJk13xddprit8P7Sc_RYbdQ-BTtZtz6MRv3uwHGDbfut9g9MywVyEbhWzVKmQi8kDO6H9WqobWixUkqNNo0sLiJ15TrnrVzzOEOjd1qsFW-Q5mSQTvAmCpRiwOAo6MN68kltcD0Bujc7pE3F8QZMocOiz6F2GH1stwMD1-f4_V7A2eGdok0ExSDVbj3SDe9Bra-ZYzRvMmXiYnBkLPk)
59. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGVpZtblmTDCP4ZHHYP-rwd7r6ixgY9-SaKsK5QQ-5dME3Ko-8hSXyOn04wNrfof34bdrAFhbAtYUnIDWwQZPQRtg59FHK6Z2fOGlhAeyx7qif2pq09MouIfe3Iur1MEDE5cBUY-Wu84ygSA7Bs3Id7tIPeoNTtncGz6djLaUITpZFR0VzqHw36hBV7vXHixg==)
60. [networklawreview.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQEnTRReFDUh9ZArCR7VlhJE2z1FO7GR6OASVISJ8BGFtvyyhpcBL92-ohU5BtS5xBiN7shhbqC-gyDatoG48XEKMCRsIRMxoHF1c1S_9EUtZsZpi6Go3uDXy3wuIwsuKfPpQh-CR8KYt-jNKpgHk4Fp1g==)
61. [nih.gov](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGtkt7wjZIws0yLMzLrjwdWEAUIbkSKOFHqhr5fIp0b5n-z_z2ygtr2aSWtb4WFr_MFv3TIiA-cNxQUusprRK93fg6bXsj0YaydWKrny5rFK9aq1LrBfv0UHSHh_tx_XNVqlwirmbRxDw==)
62. [arxiv.org](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFG1Ki2krGwPC8PagqhnTJAQYj1xtuWkaUWvn7LkKOK1Sad3fDSp7eT4W6dUqH9basXvHSIUxjW2L-AE8IfVxEQG90mN5WFVg2pCp_T7bKHVZtNf5ir4Q==)
63. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQH8N9Am3sW3Lm3XAne_hJbVk8Nemw2996q2KkMfTQQk7ZF3X5XlBTN2K70DieWDndr75Blj6YNB9g6rUxRlJCuaIh_MctzWUvkCpqTvpgqGZRBWzYeSZdS2lZxbZSw806hGRXs74Z15YIftpsbYvoo_jQzm_7q41Qy8rMGbKF7rRREoz3Cb6AwZR0pBu4ydhonG9pH8jBjAkuvHFTTvKqW_ZL-JQ1EIYLks6A==)
64. [researchgate.net](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE6xEUOA5P2_EfM148NLNW9-08ItWi0Dd665RpItsYu5EUPBexMYVJBkda6KcFFJlLT-tEIFSA2BEY27n1GLY8X5EGnzreBm2HjH6oxAqSPMVFy9tmdO_VH-klbDN0Scj-G9Y_3MprE4oCs0pph6eq_CRxTthpFSyWTItqHIXxTxCJcSi-UldAD3CpbexYKU8Nw6k48qDO61Ns_My5xwZ7CFoWzq_DmPwT4swCkwQ6Z25MqE3QwcovMWF80ZhxnvNZUBLC0XQ==)
65. [ox.ac.uk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQE_033SNTS7BQvcAXYbntPmmoRBWhfbV8h985ckiCpPbGub96BXH39GYRxjL6suVM96sE0FMXfQL-wBLUKgwt8VKseE2dthdapB5bvMjTT-jrYK2E64bR8PyM8urqykesW-9Wso4jvj3A==)
66. [parliament.uk](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFeRe4cGCwp3Y2wxSKKtI7Qw_nkI4EIhIDIGCx1HDftxh5nnfDGXNiLhZf76TTXXRssdiQu-by2hhsDsFqaq4z5JBwacBE97VCgkz1m4yNux6HdZ2_HNVPALKu-ooYW-rNdACwdBC3YsByEjpIYgXbxeZnLXrp7ISm2lWPPwje_rLMDqg==)