The authors are careful to stress that they are not advocating blind automation. Rights- and process-based constraints such as contestability, nondiscrimination, recourse, and accountable oversight can legitimately limit any decision system, whether human or machine. Their claim is instead one of symmetry. When audited evidence shows AI outperforming humans on endorsed criteria and the machine can satisfy core procedural safeguards, the justificatory burden shifts. At that point, they argue, asking only whether AI is safe enough to deploy leaves the harder question unasked: whether human control is safe enough to keep.

The argument rests on a substantial body of empirical research showing that human judgment is often less reliable than we assume. Decades of work dating back to Paul Meehl’s 1954 analysis of clinical versus statistical prediction, and the 2000 meta-analysis by William Grove and colleagues in Psychological Assessment, have repeatedly found that simple mechanical rules match or beat expert clinical judgment. Daniel Kahneman and colleagues’ work on noise documents how inconsistent human judgments are, even among trained professionals, while research on time-of-day effects shows that memory performance and other cognitive functions fluctuate with factors as mundane as the hour of the day. Ego depletion research further suggests that self-control, a resource central to careful decision-making, is limited and unreliable.

Aeon and Gruda trace the human gold standard fallacy to four recurring psychological drivers. The first is omission bias, the well-documented tendency to judge harms caused by action more harshly than equal or greater harms caused by inaction. When an algorithm makes a bad decision, the harm feels like something we did; when a human makes the same or worse decision, the harm is reframed as something that simply happened. Meta-analytic work by Sek Yeung, Timur Yay, and Gilad Feldman in 2022 confirmed robust omission-commission asymmetries in moral judgment, and philosophical analysis by Pinky Willemsen and Katia Reuter questions whether the omission effect is even coherent.

The second driver is agent-relative ethics, the philosophical position that what one is permitted to do differs from what one would prefer to see happen, making the identity of the decider morally salient. Research by David McNaughton and Piers Rawling on agent-relativity and the doing-happening distinction underpins this tradition, and the authors argue that such considerations make status-quo human authority feel morally safer even when it performs worse by our stated standards. The third driver they call intuitive exceptionalism, the deep-seated intuition that human minds occupy a special category, related to findings by Yoel Inbar and colleagues on people’s aversion to machines making moral decisions, and to taboo trade-off research by Philip Tetlock and colleagues showing that people react strongly when sacred values are weighed against mere utilities.

The fourth driver is the control premium: the value people place on retaining control for its own sake. Ellen Langer’s classic 1975 work on the illusion of control demonstrated that people systematically overestimate their influence over outcomes, and studies of decision-making under uncertainty show a propensity to under-delegate even when delegation would yield gains. Recent work by Anna Grundke on status threat suggests that outperforming machines evoke a threatened sense of human standing, while research by Nicolas Spatola and Antoine Normand documents the psychological costs of being compared to a superior artificial agent. Together, these four forces, the authors argue, make human authority feel better than machine authority even when evidence favors the machine.

The paper situates this argument within the current regulatory landscape, particularly the European Union’s Artificial Intelligence Act, which mandates human oversight for high-risk systems. The authors cite work by Jakob Laux on institutionalized distrust and democratic AI governance, by Leander Enqvist on what human oversight under the AI Act actually requires and by whom, and by Ben Green on the flaws of policies requiring human oversight of government algorithms. Legal scholarship on hybrid human-algorithmic decision-making by Tomas Enarsson, Leonhard Enqvist, and Marko Naarttijärvi, together with critiques of moral crumple zones by Madeleine Clare Elish, suggests that human-in-the-loop requirements often place humans in positions where they rubber-stamp automated outputs or absorb blame without exercising genuine control.

The empirical evidence on performance gaps is, in some domains, already striking. Generative AI has been shown to increase worker productivity in field experiments published in Science and the Quarterly Journal of Economics. Randomized controlled trials have found that AI tutoring outperforms in-class active learning in authentic educational settings. Large language models have been reported to outperform experts on challenging biology benchmarks, and work on AI-assisted prostate cancer MRI diagnosis examines whether domain experts can rely on AI appropriately. Studies of perceived fairness by Christoph Kern and colleagues and of public attitudes by Kevin Bansak and Elizabeth Paulson show that public acceptance of algorithmic decision-makers often depends on perceived performance, suggesting that intuitions may shift as evidence accumulates.

The authors also engage with the classic objection that delegating decisions to machines creates a responsibility gap. Andreas Matthias’s influential 2004 analysis of learning automata, subsequent work by Filip Santoni De Sio and Jero van den Hoven on meaningful human control, and Jilles Davidovic’s 2023 argument on the purpose of meaningful human control all grapple with the problem that machines may act in ways no one intended. Aeon and Gruda do not dismiss this concern, but they note that accountability research by Claude Novelli, Mariarosaria Taddeo, and Luciano Floridi, and by Atoosa Kasirzadeh and colleagues in the fairness, accountability, and transparency literature, suggests that institutional structures can attribute responsibility across human and machine actors. Meanwhile, legal scholars such as Diane Desierto and Frank Pasquale have long argued for due process rights in scored, automated societies, indicating that procedural protections need not presuppose human superiority.

The article’s central provocation is ultimately about epistemic fairness in policy design. If a society claims to care about accuracy, fairness, and welfare, then its governance frameworks should be symmetric in how they treat human and machine decision-makers: both should be audited, both should be subject to procedural constraints, and neither should enjoy an unearned presumption of legitimacy. The authors acknowledge that some decisions may remain legitimately reserved for humans on the basis of rights and dignity, but they insist that such reservations must be articulated and defended rather than smuggled in as defaults. Their closing challenge is likely to reverberate through AI governance debates for years: the burden of proof does not rest permanently on the side of automation. Once machines demonstrably serve our stated values better than we do, the burden falls on those who would keep control in human hands to explain why.

Subject of Research: The ethics of human versus AI decision-making authority and the burden of proof in AI governance

Article Title: When AI outperforms humans, who bears the burden of proof?

Article References: Aeon, B., & Gruda, D. (2026). When AI outperforms humans, who bears the burden of proof?. AI & SOCIETY. https://doi.org/10.1007/s00146-026-03375-1

Image Credits: AI Generated

DOI: 10.1007/s00146-026-03375-1

Keywords: artificial intelligence, AI governance, algorithm aversion, human oversight, decision-making, ethics of technology, omission bias, algorithmic accountability, AI Act, burden of proof, automation, moral psychology

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