The research team, led by Lei Lin of the Southwest University of Political Science and Law, together with Jing Tan of Chongqing University and Di Zheng of the Sichuan Academy of Social Sciences, recruited 599 urban residents in China for two randomized online experiments. The choice of population was deliberate. China has more than 900 million urban permanent residents, a group that overlaps heavily with the audiences of official government microblogs, weather applications, and other digital warning channels, and these users are among the most frequent adopters of AI-enabled services. The experimenters used Credamo, a survey platform with verified users, and applied strict quality controls: eligibility was restricted to the highest credit ratings, duplicate submissions from the same IP address or device were blocked, minimum completion times were enforced, and attention and comprehension checks filtered out careless respondents. Of 636 initial responses, 37 were excluded under preregistered criteria, leaving a final sample whose demographic profile shifted negligibly after cleaning.

The scenario at the heart of both experiments was a heavy rainfall warning issued during peak flood season in a Chinese city, a hazard chosen for its familiarity, its low political sensitivity, and the fact that both government agencies and AI platforms plausibly issue such alerts. Warning severity was communicated through the standardized color-coded signals of the China Meteorological Administration, ranging from blue for general risk through yellow and orange to red for extreme risk. All experimental materials were identical in content, length, and style; only the source cue varied. Pilot testing confirmed that participants could accurately identify who was said to have issued each warning and perceived the scenarios as realistic.

The first experiment employed a three-group between-subjects design comparing a government-only warning, an AI-only warning, and a consistent joint warning issued by both. Three hundred valid participants were randomly assigned, 101 to the government condition, 100 to the AI condition, and 99 to the joint condition, and balance checks confirmed no pretreatment differences across groups on gender, age, education, income, occupation, or disaster experience. A one-way analysis of variance revealed a substantial effect of warning source on information trust, with a partial eta squared of 0.215, a medium-to-large effect. Post hoc comparisons using Tukey’s honestly significant difference test showed that trust in government warnings significantly exceeded trust in AI warnings, with a mean difference of 0.29 points on the five-point scale and a Cohen’s d of roughly 0.45.

More striking still was the performance of the joint condition. Warnings attributed jointly to the meteorological station and an AI-driven warning platform produced the highest trust of all, exceeding the government-only condition by 0.26 points and the AI-only condition by 0.55 points, the latter corresponding to a Cohen’s d of approximately 0.85, a large effect. The authors interpret this as evidence that credibility emerges not from any single source but from the combination of institutional authority and technological channels. When AI alerts appear alongside official government communication, they are more likely to be perceived as complementary rather than autonomous, easing concerns about algorithmic opacity and unclear responsibility. The finding echoes the long-standing ‘speak with one voice’ principle in risk communication, which emphasizes coherent messaging across sources to sustain public confidence.

The experiment also confirmed that trust translates into action. In a hierarchical regression controlling for gender, age, education, prior disaster experience, AI usage, institutional trust, and risk perception, adding information trust to the model raised the explained variance in protective intentions from a nonsignificant 2.9 percent to 12.7 percent. Information trust was the strongest predictor of intentions to monitor hazard information, move vehicles and belongings to safety, identify emergency shelters, and stockpile supplies, with a standardized coefficient of 0.416. Risk perception retained a smaller but significant effect. Protective intention items showed acceptable internal consistency, with Cronbach’s alpha of 0.72, while the trust and anticipated regret composites reached 0.82 and 0.84 respectively.

The second experiment turned to the harder question: what happens when government and AI disagree? Four conditions were constructed. In one, both sources issued a low-level blue warning; in a second, both issued an orange warning, serving as a reference and replicating the joint condition from the first study. In the two conflict conditions, one source issued orange while the other issued blue, with the order of severity reversed between the two versions. After excluding invalid responses, 387 participants were distributed almost evenly across the four groups, and randomization checks again showed no meaningful imbalances. A pretest had verified that a one-level discrepancy was perceived as a genuine conflict and that orange was consistently read as more severe than blue.

Conflict, it turned out, produced a powerful emotional response. A one-way analysis of variance revealed a large effect of warning consistency on anticipated regret, with a partial eta squared of 0.349, and a nonparametric Kruskal-Wallis test corroborated the result. Participants in both conflict conditions reported significantly higher anticipated regret than those in either consistent condition, exceeding the consistent-low group by roughly 0.85 to 0.91 points and the consistent-high group by 0.33 to 0.39 points, all differences highly significant. Crucially, the two conflict groups did not differ from each other, meaning it did not matter whether the government or the AI issued the more severe alert. When sources diverged, people temporarily set aside questions of institutional authority and focused instead on avoiding the cost of underestimating the threat, a pattern consistent with worst-case thinking described in regret theory and loss aversion research.

Anticipated regret also proved to be a formidable driver of behavior. In a hierarchical regression, adding anticipated regret to the controls raised explained variance in protective intentions from a nonsignificant 1.7 percent to 26.8 percent, with a standardized coefficient of 0.502, meaning a one-unit increase in expected regret corresponded to roughly half a unit more protective intention, all else held constant. Together, the two experiments support a dual-pathway framework of decision making in multi-source warning environments: under consistent messaging, responses are structured by cognitive trust grounded in institutional authority and technological support; under conflicting messages, responses are instead guided by the emotional motivation to avoid future regret.

The implications for warning system design are substantial. Because joint government-AI releases enhance confidence, the authors suggest that policymakers develop coordination mechanisms between meteorological agencies and AI service providers, aligning update schedules, severity thresholds, and release timing to reduce discrepancies before they reach the public. At the same time, they caution against exploiting regret as a persuasive tool, which could breed anxiety or distrust; instead, communication strategies should explain why different systems produce different alerts and offer graded recommendations for how to respond. The study acknowledges limitations: the simulated scenarios cannot fully reproduce the urgency of real disasters, the sample of urban online users limits generalizability, and the two experiments examined trust and regret in separate randomized conditions rather than estimating a unified causal pathway. Nonetheless, as AI becomes an increasingly visible participant in public warning dissemination, the research makes clear that institutional credibility remains the anchor of public confidence, and that when machines and governments speak in harmony, people listen most closely of all.

Subject of Research: Public trust and behavioral responses to disaster warnings issued by governments versus artificial intelligence systems.

Article Title: Public Responses to Disaster Warnings from Government Versus AI: Evidence from Behavioral Experiments

Article References: Lin, L., Tan, J., & Zheng, D. (2026). Public Responses to Disaster Warnings from Government Versus AI: Evidence from Behavioral Experiments. International Journal of Disaster Risk Science. https://doi.org/10.1007/s13753-026-00766-2

Image Credits: AI Generated

DOI: 10.1007/s13753-026-00766-2

Keywords: disaster warnings, artificial intelligence, government trust, anticipated regret, risk communication, behavioral experiments, early warning systems, protective behavior, algorithm aversion, flood warnings, China, decision making

The research, conducted by Yanqing Wang, Xiao Gu, and Yibao Wang of the China University of Mining and Technology in Xuzhou, tackles a question that has received surprisingly little attention: how does emergency knowledge actually move through informal social networks, the ones that operate beneath the radar of official campaigns and school curricula? Previous work on emergency science communication has focused largely on formal channels—government messaging, structured training programs, and, more recently, generative AI tools that can package safety information for broad audiences. But real-world learning often happens in conversations between people who already trust each other, and it is precisely these informal networks that the new study seeks to understand.

To do so, the team built their model on a foundation from network science: the small-world network, a structure famously characterized in 1998 by Duncan Watts and Steven Strogatz. Small-world networks combine two seemingly contradictory properties—they exhibit high clustering, meaning people form tight-knit groups, while also having short average path lengths, meaning any two individuals are connected through only a few intermediaries. This architecture mirrors real human societies well, and previous research has shown that small-world structures accelerate the spread of information and innovation. But a limitation of the classic model is that it treats every relationship as binary: either two people are connected or they are not. Human relationships are richer than that. Some bonds are strong and trusted; others are weak and infrequent.

The researchers therefore adopted a weighted small-world network, drawing on the framework introduced by Latora and Marchiori, which assigns each edge in the network a weight representing the intensity or closeness of the relationship between two nodes. In their formulation, the public’s emergency knowledge diffusion network consists of a set of nodes, a set of edges, and a matrix of relationship strengths ranging from zero to one. Each of the 1,000 simulated individuals carries six distinct types of emergency knowledge—disaster risk knowledge, emergency common sense, escape skills, first aid knowledge, knowledge of public participation, and knowledge of emergency policies and regulations. A node’s overall knowledge level is a weighted average of these six categories, and diffusion occurs whenever one node holds more of a given knowledge type than an adjacent node, creating a knowledge gap that can be closed through interaction.

The elegance of the model lies in its growth mechanics. When knowledge passes from a sender to a receiver, the receiver’s knowledge increases in proportion to their absorption capability, the strength of the relationship between them, and the size of the knowledge gap itself. Meanwhile, the sender’s knowledge remains unchanged—knowledge, unlike a physical resource, is not depleted by sharing. The researchers also incorporated a social reality: because personal time and energy are limited, strengthening one relationship necessarily weakens another. After each exchange, the bond between sender and receiver grows by a fixed increment, while a randomly selected alternative relationship of the receiver is weakened by the same amount. This dynamic captures the constant renegotiation of social closeness that characterizes human networks.

Perhaps the most interesting design choice concerns how a receiver selects a sender. The team tested three strategies. In random selection, a receiver simply picks any eligible neighbor. In relation priority, the receiver chooses the neighbor with whom they have the strongest bond. In knowledge priority, the receiver seeks out the neighbor with the largest knowledge gap—the person who knows the most. Running half a million simulation steps in MATLAB R2023a across networks of 1,000 nodes with an average of eight connections each, the team compared outcomes across all three strategies, measured both by the average knowledge level of the network and by the variance in knowledge across nodes, which indicates how equitably knowledge has spread.

The results carry a nuanced message about timing and strategy. Regardless of how strong the relationships in a network are, the knowledge priority strategy produces the fastest short-term growth in collective emergency knowledge, because people learn the most from those who know the most. But this advantage fades over time. As the network’s average knowledge rises, the knowledge gaps that made the strategy effective diminish, and outcomes converge. In strong relationship networks, the relation priority strategy ultimately catches up and, in some conditions, surpasses knowledge priority, while also producing the lowest variance—meaning knowledge spreads most evenly and the network reaches equilibrium fastest. This makes intuitive sense: in tightly bonded communities, repeated contact and high trust reinforce the diffusion process, consistent with prior research showing that trust amplifies knowledge sharing.

The study also examined how external institutional factors shape these dynamics. Four variables were introduced: public emergency participation policies, legal protection for participants, information technology, and the cost of knowledge diffusion, each assigned an intervention coefficient derived from earlier survey-based research. Policies, legal safeguards, and information technology all showed significant positive effects on how quickly emergency knowledge spreads through informal networks. Diffusion cost, by contrast, exerted a significant negative influence—a reminder that participation in emergency management demands human, material, and financial resources that ordinary people may be unwilling to spend without incentives. The researchers argue this points to the need for material rewards, moral recognition, or tax incentives to stimulate public engagement.

A further finding concerns the difference between strong and weak relationship networks. Knowledge diffuses more efficiently through strong-tie networks, but weak-tie networks proved more sensitive to external interventions. Because individuals in weak networks interact less frequently and hold fewer overlapping connections, policy changes or technological improvements have more room to reshape their behavior. In other words, if governments want to influence how people in loosely connected communities learn about emergencies, external interventions—clear policies, legal protection, accessible technology—may be the most effective lever.

The practical implications extend well beyond China. The authors recommend building community-based emergency participation networks by cultivating a sense of shared belonging, organizing risk-screening exercises, and training neighborhood “grid” workers to act as connectors between residents and professional emergency services. They also highlight the growing role of virtual communities, suggesting that online opinion leaders could package emergency knowledge into engaging formats such as short videos—without compromising accuracy—to reach audiences that formal channels miss. The study draws a sobering lesson from the 2021 Zhengzhou flood disaster, in which some trapped individuals lacked the escape knowledge to act before the critical window closed, underscoring the stakes of getting this right.

The work is not without limitations, which the authors acknowledge candidly. The model assumes everyone wants to learn emergency knowledge, ignores the tendency of knowledge to decay over time without reinforcement, and treats participation willingness as uniform. Future iterations could introduce thresholds of motivation and knowledge attenuation, and field experiments could validate whether the suggested interventions work in practice. Still, by fusing network science with emergency management, the study opens a new paradigm for understanding how the public becomes prepared—not through top-down instruction alone, but through the quiet, cumulative exchange of knowledge between people who know and trust one another. In an era of intensifying disasters, that quiet exchange may prove to be one of the most powerful tools available.

The research, published in the International Journal of Disaster Risk Science, was conducted by Sara Bonati of the University of Genoa, Monia Del Pinto of Loughborough University, and Giuseppe Forino of the University of Salford. The team combined place branding theory with critical disaster studies to examine how the projects of internationally celebrated architects, so-called starchitects, have been deployed in Italian disaster zones over the past fifteen years. Their central claim is provocative: starchitecture in the aftermath of catastrophe operates less as reconstruction than as communication, and that communication serves powerful political and economic interests.

The authors build on the concept of “branding disaster” introduced by consumer researchers Humphreys and Thompson, who described how private companies use market communication to restore trust after industrial accidents. Bonati and colleagues expand the idea from corporate reputation to whole territories, arguing that in Italy the practice has become a recurring, structural feature of the reconstruction system, serving both private donors and public institutions. When a local administration recruits a Pritzker Prize winner in the weeks following an earthquake, the study suggests, it is not simply commissioning a building. It is purchasing a seal of guarantee, a globally recognized brand mark stamped onto a recovery process that may be slow, contested, or underfunded.

To test this hypothesis, the researchers applied critical discourse analysis, a method rooted in linguistics that treats texts as exercises of power and influence. Rather than studying the buildings themselves, they analyzed the official project descriptions published on starchitects’ firm websites, documents that typically serve as the source text for all subsequent press releases. Using a framework derived from place branding and critical disaster theory, the team identified recurring macro-codes and sub-codes in the language, then examined how meanings were stratified and conveyed. The cases included an auditorium in L’Aquila after the 2009 earthquake, a “Taste Hub” in post-earthquake Amatrice, a civil protection center in Norcia, the replacement for the collapsed Morandi Bridge in Genoa, and even Renzo Piano’s flower-shaped vaccination pavilions designed during the Covid-19 pandemic.

Across all of the cases, certain keywords surfaced with striking regularity: rebirth, identity, authenticity, effectiveness, and safety. The language of recovery was ubiquitous. Projects were framed as urgent necessities, “accelerators” of reconstruction or spaces in which a wounded place could “start the healing process.” In Amatrice, famous as the birthplace of the Amatriciana pasta dish, the new food hub was justified as a way of letting the town be “born again” through the continuity of its economy and cuisine, with food tourism positioned as the real beneficiary. In Genoa, the new bridge was presented as the “driving force” behind the transformation of a “fragile ecosystem,” reasserting the city’s identity as a commercial harbor first and a community of grieving residents second.

The researchers show how authenticity itself becomes a commodity in these narratives. Local traditions, whether musical heritage or gastronomy, are distilled into marketable symbols, a process the authors describe as the manipulation of culture. Place branding literature has long noted that such selective storytelling removes the aspects of a place that are not appealing to the market and silences the voices that do not support the dominant narrative. In the post-disaster context, this means that systemic drivers of risk, such as depopulation, unemployment, and decades of unaddressed vulnerability, are carefully edited out of the story, replaced by romanticized images of local excellence that reassure outsiders and tourists alike.

Participation, when it appears in the project descriptions, also comes under scrutiny. The auditorium in L’Aquila included an educational laboratory for engineering and architecture students, yet only twenty-two students took part, a small fraction of the local student population. In Amatrice, the planning consultation involved restaurant owners, a niche of the community with direct market interests, even as the building was presented as being “aimed at the common good.” These limited participatory gestures, the study argues, function as branding devices that manufacture consensus rather than genuinely distribute decision-making power, creating what the researchers call shadow spaces in which marginalized residents and dissenting voices are excluded from the transformed place.

Perhaps the sharpest critique concerns the themes of speed, effectiveness, and safety. Project descriptions emphasized completion times with an almost competitive tone: “only 12 weeks” for the Amatrice food hub, “just over three months” for the Norcia civil protection center, and record-speed delivery for the Genoa bridge, explicitly offered as “a model” for future projects. Safety was dramatized through materials and structure, with timber celebrated for its elasticity, lightness, and anti-seismic properties, and reinforced concrete invoked for the bridge as a promise of robustness. The authors note that such claims of effectiveness are a classic instrument governments use to demonstrate competence during crises, and that in the Italian cases they amplified and consolidated the official message of an efficient reconstruction machine, even where residents complained bitterly about delays and unmet needs.

The historical sweep of the study gives the argument additional weight. The authors trace a lineage from the Renaissance master builders through the celebrification of Le Corbusier and Frank Lloyd Wright to the global starchitect boom of the 1980s and 1990s, when a handful of internationally consecrated designers began operating on a global circuit, driving capital and, in the authors’ words, serving as a product of neoliberal culture. Iconic buildings became urban totems of the “Bilbao Effect,” deployed by policymakers for consensus-building and media exposure. Post-disaster contexts, from Hurricane Katrina in New Orleans to the tsunami-stricken Tohoku region of Japan, have increasingly attracted such figures, sometimes for genuinely humanitarian work like Shigeru Ban’s paper architecture, but also for scenic one-off interventions that become symbols of recovery while remaining marginal to the actual rebuilding.

Italy offers a particularly revealing laboratory. The study contrasts earlier episodes, such as the deliberate invitation of artists and architects to the new town of Gibellina after the 1968 Belice earthquake, a collective civic experiment that nonetheless failed to bridge the divide between residents and their new surroundings, with the Foligno church won by Massimiliano Fuksas after the 1997 Umbria quake, a standalone “sacred box” whose international aesthetic marked a shift toward individuality and symbolic abstraction. Since the 2000s, the authors argue, Italian reconstruction has progressively opened to punctual interventions of authorial architecture, with signature designs acting as devices for strategic place resignification, the hallmark of branding.

The consequences of disaster branding, the researchers warn, extend well beyond aesthetics. Because architecture exercises a form of soft power, reassuring and aesthetically compelling narratives can make the alteration of place identity acceptable even when it does not reflect the will of local communities, potentially facilitating dispossession and gentrification. In Amatrice, the branding of rebirth coexisted with public claims against a reconstruction that dragged on for years. The symbolic restoration of the pre-disaster status quo through a single landmark, an ideal long criticized in disaster research, risks trivializing a complex, contested process by reducing it to one photogenic building.

The study does not call for the exclusion of famous architects from disaster zones, but it does insist that their dual influence, physical and communicative, be critically assessed. Unpacking who benefits from disaster branding, the authors conclude, is essential to understanding the networks of power and consensus-building that shape post-disaster recovery. As climate change intensifies hazards worldwide, and as recovery increasingly offers a clean slate for neoliberal speculation, the spectacle of the starchitect’s ribbon-cutting deserves more skeptical attention. A building may symbolize rebirth, the researchers suggest, but the question worth asking is whose rebirth it actually serves.

What happens when weather forecasts do not match reality? How do the public emotionally respond when a disaster unfolds differently from what they expected? A research team led by Professor Jonghun Kam and Kiru Kim from the Department of Environmental Engineering at POSTECH investigated how forecast error types influenced public emotion during the landfall of Typhoon Khanun. Using an artificial intelligence (AI), Natural Language Processing (NPL), the researchers found that different types of forecast error (e.g., over/underestimation) triggered distinct emotional responses among the public. The study has been published in GeoHealth.

Weather forecasts are essential for disaster preparedness, but they are inherently uncertain. During extreme weather events such as typhoons, forecasts may either overestimate rainfall that never occurs or underestimate rainfall that turns out to be severe. The research team focused on how these mismatches between expectation and reality affect public perceptions and emotional responses before and after the landfall of Typhoon Khanun.

The researchers assessed the Korean Meteorological Administration’s prediction skill of Typhoon Khanun, which crossed the Korean Peninsula in August 2023 against observed rainfall records from 613 weather stations. They also analyzed more than 43,000 online posts from NAVER Report Talk using an AI-based natural language processing model.

The results revealed clear spatial differences in forecast performance. Forecasts tended to underestimate heavy rainfall in the eastern and southeastern regions of the Korean Peninusla while overestimating rainfall in the western and metropolitan areas. In regions where rainfall was overestimated, anxiety, worry, and fatigue became dominant emotion types in online discourses while the public from the regions with underestimated rainfall showed a high level of certain emotion types like confusion, embarrassment, and sadness.

The researchers also examined changes in regional emotion. In the eastern and southeastern regions, where actual rainfall exceeded forecast amounts, online discussions frequently expressed confusion, uncertainty, and anxiety as Typhoon Khanun passed by the Korean Peninsula. The western and metropolitan regions, where rainfall forecasts were higher than observed rainfall, showed dominant emotions of worry and concern, followed by increasing a level of emotion types, relief and reassurance, as the typhoon passed. These findings suggest that people respond not only to the disaster itself but also to the gap between what they expected and what they actually experienced.

Overall, approximately 55% of all online discourses expressed negative emotions, with anxiety and worry being the most common. The researchers also found that information-seeking activities peaked before the typhoon’s landfall, whereas online reporting and experience sharing surged during the event itself. This indicates that the public shift from passive information consumers to active information providers as disaster impacts become more immediate.

The findings demonstrate that forecast accuracy is not only a technical issue but also an important factor influencing public perception and emotional well-being. More importantly, the study highlights that the mismatch between anticipated risk and experienced reality plays a critical role in shaping public emotions and their perceived risk during disasters. The researchers suggest that communicating forecast uncertainty more effectively could improve public trust and reduce emotional distress during future extreme weather events.

Kiru Kim, the lead author, noted, “This study demonstrates that in disaster situations, it is important not only to improve forecast accuracy but also to develop risk communication strategies that effectively convey uncertainty to the public.”

Professor Jonghun Kam said, “This study demonstrates how AI can be used to analyze large-scale public discourse and monitor the emotional impacts of forecast errors. The findings provide new insights into how to develop effective risk communication strategies during landfalling typhoons and other natural disasters.”

Journal

GeoHealth

DOI

10.1029/2026GH001837