A team of Czech seismologists led by Pavla Hrubcová has analysed two earthquakes that struck near Mirotice, one of magnitude 3.5 in March 2024 and a smaller magnitude 3.1 event in April 2025, combining conventional seismological data with more than 1,500 citizen-filled macroseismic questionnaires for each event. Their study, published in the journal Solid Earth, demonstrates that ordinary people describing their experiences can effectively constrain deep geological structures in a region where earthquakes, and therefore seismic instruments’ chances of illuminating the subsurface, are few and far between.

What made these earthquakes scientifically remarkable was not their size but their depth. Both events originated at approximately 24.5 kilometres beneath the surface, an unusually large focal depth for the Bohemian Massif, where present-day seismicity is low and typically confined to the upper crust at depths of less than 12 kilometres. The 2024 event was followed by four minor aftershocks within an hour and further weak events over the following weeks; the 2025 earthquake produced a similar modest aftershock sequence. A small magnitude 1.7 earthquake in 2012, located at nearly the same position and at 19 kilometres depth, had already hinted that something unusual was happening deep beneath this quiet corner of Bohemia.

Pinpointing the events required assembling waveforms from a patchwork of seismic networks spanning the Czech Regional Seismic Network, Saxon and German networks, local Czech networks, temporary stations from the ongoing AdriaArray experiment, and even industrial monitoring arrays. Hypocentres were computed using a three-dimensional location program with a one-dimensional gradient velocity model calibrated against travel times from deep seismic refraction experiments and quarry blasts. The depth estimates were cross-checked using the sP depth phase recorded at epicentral distances of 140 to 210 kilometres, and by verifying that independent focal mechanism inversions yielded consistent solutions. Both earthquakes showed normal faulting on a steeply dipping, northwest-southeast striking fault plane, derived from full waveform moment tensor inversion using the Bayesian ISOLA method and a velocity model drawn from the nearby CEL09 refraction profile of the CELEBRATION 2000 experiment.

But the real revelation came from the public. Despite the epicentral area being rural and sparsely populated, with roughly 20 to 50 people per square kilometre, the Institute of Geophysics collected 1,512 questionnaires for the 2024 earthquake and 1,679 for the 2025 event, submitted through online forms and a smartphone application. Media interviews following unusual events in Czechia regularly encourage citizens to report their experiences, and the response rate was so reliable that only about 2 percent of questionnaires had to be discarded for inconsistent timing, duplication, or effects not clearly attributable to the earthquake. Notably, the weaker 2025 earthquake actually generated more reports than the stronger 2024 event, a testament to both public engagement and unusually efficient transmission of seismic energy from depth.

Respondents described shaking felt across distances of 70 to 100 kilometres from the epicentre, reaching the towns of Písek, Strakonice, and Tábor and the Šumava Mountains. For the 2024 event, 66 percent of respondents reported light shaking and 17 percent strong shaking, while around 90 percent of reports in both years included acoustic phenomena such as rumbling or booming sounds. Individual responses were converted into intensities on the European Macroseismic Scale EMS-98, with qualitative terms like few, many, and most quantified into percentage ranges. Reports that the earthquake was heard but not felt were treated separately as heard-only observations, since acoustic effects alone are not diagnostic for intensity assignment. Maximum intensities reached IV to V for the 2024 event and IV for the 2025 event, with occasional hairline cracks in plaster reported near the epicentre.

The striking feature, however, was not how strong the shaking was but where it was felt. Rather than spreading in a roughly circular pattern around the epicentre, as would be expected in a geologically homogeneous region, the observations formed a distinctly trapezoidal patch, sharply bounded on the west by the contact between the Teplá-Barrandian Unit and the Moldanubian Domain, and on the east by the Rödl-Blanice Fault System. Crucially, this asymmetry could not be explained by population distribution. The more densely populated region around Pilsen to the west produced almost no reports, while the sparsely settled epicentral area yielded dense coverage. Normalising the observations by population density would only sharpen the contrast.

The geological explanation lies in the contrasting rocks on either side of those boundaries. The trapezoidal zone of strong reports is dominated by the felsic granites and granodiorites of the Central Bohemian Plutonic Complex and the high-grade metamorphic rocks of the Moldanubian Domain, crystalline rocks that attenuate seismic energy weakly and allow waves to propagate efficiently, possibly with resonant effects. West of the pluton, the early Palaeozoic sedimentary sequences of the Teplá-Barrandian Unit, with their steeply oriented, highly anisotropic deformation structures near the contact, appear to absorb seismic energy far more effectively, and observations drop off sharply. To the east, the attenuation coincides with the Rödl-Blanice Fault System, a crustal-scale discontinuity reactivated during the Cretaceous and Miocene. These subvertical structures, the authors argue, are not clearly resolved by conventional seismic investigations, yet the citizen reports delineate them with remarkable clarity.

The tectonic story behind these boundaries stretches back hundreds of millions of years. The Bohemian Massif, the largest exposure of the European Variscan orogenic belt, formed between roughly 500 and 250 million years ago during the convergence of the ancient continents Gondwana and Baltica. The closure of the Saxothuringian ocean and subsequent continental underthrusting created the Central Bohemian Plutonic Complex as a magmatic arc at the boundary between the Teplá-Barrandian and Moldanubian units. Deep seismic profiling has revealed that the Moldanubian crust is cold and mechanically strong, with a sharp Moho discontinuity reaching depths of up to 39 kilometres and surface heat flow below 50 milliwatts per square metre. That low geothermal gradient, the study suggests, is what allows brittle failure, and therefore earthquakes, to occur in the lower crust at depths where rocks elsewhere would deform plastically. The hypocentres of the Mirotice events coincide with a reflective interface identified along the CEL09 profile, which may represent a structural or rheological discontinuity controlling earthquake nucleation.

The implications reach well beyond Czechia. In regions of sparse seismicity and limited instrumentation, crowdsourced macroseismic observations offer a cost-effective way to gather large volumes of data that reflect geological controls on seismic wave propagation, complementing instrumental networks that may record only a handful of events per decade. The Mirotice study shows that the pattern of who felt an earthquake, and who did not, can trace deep, steeply dipping tectonic boundaries rooted in the lower crust, structures that pre-Permian thrusts in the Variscan belt are known to follow. Two modest earthquakes, felt by thousands of ordinary people and dutifully reported through a web questionnaire, have drawn a portrait of the deep crust that no amount of waiting for the next big event could have produced. In seismology’s quiet corners, it seems, the public itself has become the most sensitive array of all.

Subject of Research: Crowdsourced macroseismic observations of two rare deep earthquakes constrain crustal structure in the south Bohemian Massif

Article Title: Public response to two rare earthquakes: crowdsourcing constraints on crustal structure in south Bohemian Massif

Article References: Hrubcová, P., Machek, M., Vackář, J., Pohořalá, A., Kampfová Exnerová, H., Špaček, P., & Zedník, J. (2026). Public response to two rare earthquakes: crowdsourcing constraints on crustal structure in south Bohemian Massif. Solid Earth, 17(8), 979-989. https://doi.org/10.5194/se-17-979-2026

Image Credits: AI Generated

DOI: 10.5194/se-17-979-2026

Keywords: earthquakes, Bohemian Massif, crowdsourcing, macroseismic intensity, crustal structure, seismology, Czechia, Mirotice, intraplate seismicity, seismic attenuation, Variscan orogeny, citizen science

A team of researchers led by Gunnar P. Epping and Jennifer S. Trueblood of Indiana University, working with Andrew Caplin of New York University, Erik Duhaime of Centaur Labs, and Daniel Martin of the University of California, Santa Barbara, has now demonstrated a strikingly simple fix. In a study published in Behavior Research Methods, they introduce what they call cognitive-inspired data engineering: the idea that empirical findings and models from cognitive science can be applied directly to crowdsourced judgments to correct systematic biases before the data ever reaches a machine learning algorithm. Rather than de-biasing models after training or modifying training objectives, the team intervenes upstream, at the level of the individual annotation, and shows that this single correction improves both the quality of crowdsourced datasets and the accuracy of the convolutional neural networks trained on them.

The core insight concerns how annotators express uncertainty. The industry-standard approach asks workers to provide a single binary label for each image, a method that discards valuable information whenever an annotator is unsure. An alternative is to elicit subjective probability judgments, asking annotators how likely they think it is that an image belongs to a given category. Probability judgments convey far more information, but they also open the door to more cognitive biases. Beyond simple response biases, where a rater favors one side of the scale, probability judgments are vulnerable to overconfidence and underconfidence, phenomena cognitive scientists have documented for decades. The hard-easy effect, for instance, leads people to be overconfident on difficult classification tasks and underconfident on easy ones. As a result, subjective probabilities rarely match the actual likelihood of being correct, a mismatch known as poor calibration. Crucially, however, these judgments remain correlated with correctness, which means the information they carry can be rescued if the biases are stripped away.

The researchers used a well-established cognitive model called the linear in log odds, or LLO, function to perform that rescue. The function transforms each annotator’s raw probability judgments into log-odds, fits a logistic regression against ground-truth labels from a small calibration set, and then maps the annotator’s remaining judgments onto recalibrated probabilities. The two free parameters of the function carry direct psychological interpretations. The slope parameter corrects for overconfidence or underconfidence: when its magnitude exceeds one, the function spreads judgments toward the extremes of the scale, indicating the annotator was underconfident, while a magnitude below one compresses judgments toward the center, correcting overconfidence. The intercept parameter corrects overall response tendency, shifting judgments up or down the probability scale to undo a systematic leaning toward one class. Because biases differ substantially across individuals, the researchers fit a unique recalibration function to each annotator separately, effectively normalizing every worker’s judgments before combining them into crowd labels.

To test the approach, the team chose a task with real clinical stakes: classifying white blood cells as blast cells or non-blast cells, a critical step in diagnosing malignant blood diseases such as leukemia and lymphoma. The stimuli were 549 digital images of Wright-stained cells from anonymized patient blood smears at Vanderbilt University Medical Center, captured with an automated cell morphology instrument and ground-truthed by three hematopathology faculty members who had to agree unanimously on each classification. The task is an ideal sandbox for studying high-skill perceptual decision-making: novices can reach roughly 65 percent accuracy with minimal training, the task remains challenging even for experts, and machine learning models can be trained on only a few hundred images rather than the thousands required for comparable problems.

In the first experiment, 400 Amazon Mechanical Turk workers recruited through CloudResearch judged the probability that each image contained a blast cell. After training with labeled example images and practice trials with feedback, participants completed testing blocks in which a small subset of images with known labels served as the calibration set for the LLO function. The results were revealing. Recalibration barely changed individual accuracy, which hovered around 65 percent before and after the transformation. But it dramatically improved calibration. Before correction, participants clustered their judgments at the extremes of the scale, and those extremes were badly miscalibrated: images judged to have a zero percent chance of being a blast cell were in fact blast cells roughly 28 percent of the time, while images judged to be certain blasts were truly blasts only 68 percent of the time. After recalibration, the response distribution spread across the scale and the calibration curve moved much closer to the ideal identity line. Most fitted slope parameters fell between negative one and one, confirming that the typical annotator was overconfident, and a few negative slopes even flipped the judgment ordering of participants who performed below chance.

The payoff appeared when individual judgments were aggregated into crowd labels. Using all available judgments, recalibration lifted crowd-label accuracy from 81.6 percent to 85.1 percent, even though no individual annotator had become more accurate. The researchers attribute this to an implicit variance-weighting mechanism: by compressing the judgments of overconfident annotators toward 50 percent, recalibration reduces their outsized influence after binarization, while stretching the judgments of underconfident annotators toward the extremes increases theirs. Recalibrated labels were also more efficient, reaching any given accuracy level with fewer judgments per image, and the benefit of calibration leveled off surprisingly quickly. Just six judgments on calibration images, three per class, captured most of the improvement, raising accuracy by 2.2 percentage points, with only a further 0.4 percentage point gain from extending to ten judgments.

The second experiment moved the study into a real-world setting using DiagnosUs, a crowdsourcing platform specializing in medical data annotation where skilled users, many of them medical students and healthcare professionals, compete in contests for prize money. The platform’s existing structure, in which gold-standard images with known labels are shuffled into unlabeled sets to provide feedback, supplied a natural calibration set at no extra cost. A total of 175 participants took part, split between a probability-judgment condition and an industry-standard binary-choice condition. Individual accuracy was higher than in the novice experiment, around 69 percent, and the two elicitation methods produced crowd labels of nearly identical accuracy, 88.3 percent each. Recalibration, however, pushed the probability-based crowd labels to 96.7 percent, an 8.4 percentage point gain that exceeded the 3.5 point gain seen with novices. The researchers attribute the larger benefit to the more accurate judgments available for fitting the recalibration function, not to the larger number of judgments, since gains again plateaued at roughly ten calibration judgments per annotator. Notably, the fitted intercept parameters skewed negative in this experiment, suggesting that medically trained annotators erred toward diagnosing cells as cancerous when unsure, consistent with the clinical convention that false positives are less costly than false negatives.

The improvements propagated directly to the machine learning models. The team trained GoogLeNet convolutional neural networks via transfer learning on crowd-labeled datasets, repeating the full training and evaluation procedure across 100 independently resampled datasets with five-fold cross-validation. At every number of judgments per image, models trained on recalibrated labels outperformed those trained on raw probability judgments or binary choices with respect to the true expert labels. Remarkably, models trained on just five recalibrated judgments per image matched the accuracy of models trained on the full set of uncalibrated judgments, a result with immediate economic implications. Because annotation cost scales with the number of judgments collected, recalibrated datasets achieve any target accuracy at a fraction of the annotations, directly attacking what the authors identify as the primary bottleneck in developing medical AI: the shortage of large, accurately labeled datasets in high-skill domains.

The authors are careful to frame the work as a proof of concept rather than a universal solution. The experiments used a single set of stimuli sampled with balanced class prevalence, so future work must test whether recalibration can also correct prevalence-induced errors that arise when target categories are rare, a notorious source of mistakes in visual search. The team also binarized crowd judgments before training, and training directly on continuous probability estimates may preserve even more of the uncertainty information that recalibration cleans up. Still, the broader message is hard to ignore. Cognitive scientists have spent decades cataloguing the systematic distortions in human probability judgment, and this study shows that knowledge can be converted into a simple, label-efficient transformation applicable to virtually any crowdsourcing task involving subjective probabilities. As the authors put it, the approach requires no changes to model architectures or training procedures; it simply fixes the data. In an era when the quality of AI systems is increasingly limited by the quality of human labels, teaching machines to think may begin with correcting how humans report what they think.

Subject of Research: Using cognitive-science-based recalibration of crowdsourced probability judgments to improve AI training data quality

Article Title: Improving crowdsourcing for AI through cognitive-inspired data engineering

Article References: Epping, G. P., Caplin, A., Duhaime, E., Holmes, W. R., Martin, D., & Trueblood, J. S. (2026). Improving crowdsourcing for AI through cognitive-inspired data engineering. Behavior Research Methods, 58(10), Article 283. https://doi.org/10.3758/s13428-026-03160-4

Image Credits: AI Generated

DOI: 10.3758/s13428-026-03160-4

Keywords: crowdsourcing, machine learning, cognitive bias, recalibration, medical imaging, data annotation, probability judgments, wisdom of crowds, neural networks, training data, human judgment, AI

The study, conducted by independent researcher Vera Wilde, introduces what she calls a capability-gap test. The logic is deceptively simple. Certain problems, such as calculating a positive predictive value from base-rate information, have been studied so extensively in humans that scientists know, with meta-analytic precision, exactly how well people can perform. When participants in an online study dramatically exceed those well-established human ceilings, the most plausible explanation is not superhuman cognition but machine assistance — a canary in the data coalmine, signaling that the sample may be contaminated by large language models.

The empirical basis for the proposal comes from two preregistered pilot studies of a Bayesian reasoning training tool, with a combined sample of 148 participants recruited through the Prolific platform. The tool was designed to teach people how to solve Bayesian inference problems, the kind of task exemplified by medical diagnosis questions: given a disease with a certain prevalence, a test with a certain sensitivity and false-positive rate, what is the probability that a person who tests positive actually has the disease? Decades of research, dating back to classic work by Daniel Kahneman and Amos Tversky and extended by Gerd Gigerenzer and Ulrich Hoffrage, have shown that most people fail such problems, even when the numbers are presented in natural frequency formats that make the underlying logic easier to grasp.

That failure is precisely what makes the task useful as a detector. A meta-analysis by McDowell and Jacobs found that only about 24 percent of people can solve a single Bayesian reasoning problem presented in natural frequency format — and that figure represents a ceiling, a level at which achieving a perfect score on a battery of five such problems is effectively unattainable for genuine human respondents. Yet in Wilde’s pilots, participants’ accuracy on positive predictive value calculation problems reached roughly three times the established human performance ceiling. In the second pilot, 57 percent of participants achieved perfect 5-for-5 scores, a result that should be extraordinarily rare in an uncontaminated human sample.

The technical heart of the approach lies in distinguishing two outcome measures: accuracy and algorithm use. Accuracy refers simply to whether the participant produced the correct numerical answer. Algorithm use, by contrast, refers to evidence in the participant’s response that they actually followed the Bayesian reasoning process — for example, constructing a frequency tree, counting cases, or showing the intermediate steps of the calculation. This distinction matters because a training intervention designed to improve Bayesian reasoning should, if it works, change both measures in tandem. A large language model, however, can produce correct answers without any visible reasoning process, or with a reasoning process that does not respond to the training manipulation in the way human learning does.

By tracking both measures simultaneously, researchers can separate two rival explanations for suspiciously high performance. If accuracy spikes but algorithm use does not, contamination is the likelier culprit, because the model supplies answers without the participant acquiring the underlying skill. If both accuracy and algorithm use rise together, the pattern is consistent with authentic learning effects, and the treatment signal can be preserved rather than discarded. In this way, the capability-gap test does not merely flag bad data; it helps researchers decide which parts of their dataset reflect genuine psychological phenomena and which parts reflect machine-generated noise.

Wilde frames the detection problem itself as a signal detection problem, structurally analogous to the mass screenings for low-prevalence conditions — such as disease screening — around which the Bayesian reasoning literature was originally developed. Just as a medical test must balance hits against false alarms, a contamination detector must catch bot-driven responses without wrongly excluding honest participants who happen to be statistically savvy. The known reference distributions from the Bayesian reasoning literature make this calibration possible in a way that ad hoc attention checks cannot. Rather than relying on generic screening questions, researchers can compare observed performance against quantified human benchmarks and estimate the probability that a given response pattern arose from machine assistance.

The approach offers four practical advantages over existing data-quality tools. First, the human performance ceilings are grounded in meta-analyses rather than informal intuition, giving researchers a defensible threshold for suspicion. Second, the human–large language model performance gap on these problems is large, which increases the sensitivity of the test. Third, the known reference distributions allow nuanced assessment rather than crude pass–fail judgments. Fourth, Bayesian reasoning problems are easy to embed in existing surveys, requiring no special software or platform cooperation. Together, these properties make the method deployable at scale across the many fields — psychology, marketing, political science, epidemiology — that increasingly depend on online samples.

The stakes are considerable. Prior research on crowd work has documented substantial and growing use of large language models by online workers, and studies of data contamination in machine learning itself show how memorized content can masquerade as genuine capability. If a meaningful fraction of respondents in an online study are completing tasks with chatbot help, effect sizes may be distorted, replication attempts may fail for reasons that have nothing to do with the underlying science, and the credibility of entire literatures built on crowdsourced data could be undermined. The problem echoes an older statistical concern: John Tukey’s foundational work on sampling from contaminated distributions warned that even small amounts of contamination can seriously mislead inference drawn from nominally clean data.

Wilde is careful to note the provenance of the idea: neither pilot study was originally designed to validate a contamination detection method, and the proposal emerged from post hoc analysis of unexpectedly strong results. That origin makes the capability-gap test a promising hypothesis rather than a fully validated diagnostic, and the author provides practical recommendations for researchers who wish to use these problems as data-quality diagnostics while the validation literature matures. Both studies were preregistered on the Open Science Framework, and all data, materials, and analysis code are publicly available, allowing other teams to scrutinize and extend the approach. If the method holds up under broader testing, Bayesian reasoning problems — long a symbol of human statistical frailty — may find a second career as guardians of scientific integrity, ensuring that the data feeding behavioral science come from human minds rather than the machines trained on those minds’ collective output.

Subject of Research: Detecting large language model contamination in online behavioral research samples using Bayesian reasoning problems as capability-gap tests

Article Title: An LLM canary in the online data coalmine: Bayesian reasoning problems as a capability-gap test for LLM contamination in online samples

Article References: Wilde, V. (2026). An LLM canary in the online data coalmine: Bayesian reasoning problems as a capability-gap test for LLM contamination in online samples. Behavior Research Methods, 58(11), Article 301. https://doi.org/10.3758/s13428-026-03184-w

Image Credits: AI Generated

DOI: 10.3758/s13428-026-03184-w

Keywords: large language models, Bayesian reasoning, data contamination, online research, data quality, signal detection, natural frequencies, positive predictive value, crowdsourcing, behavioral research methods, capability-gap test, Prolific

The research, conducted by Wei Zong, Huize Feng, and Zongyao Nie of the School of Economics and Management at Xidian University in Xi’an, China, introduces a method called GETI-WR, short for Graph Embedding for Truth Inference with Worker Reliability. The work addresses what computer scientists call truth inference: the problem of determining the correct answer, or ground truth, for a task when the only information available is a set of conflicting labels contributed by workers of unknown ability. Because obtaining expert-verified answers for every task is expensive or impossible, algorithms must somehow distill reliable signals from unreliable sources.

Most existing truth-inference methods follow a familiar recipe. They begin by assigning every worker an initial reliability value, often a uniform guess, and then iteratively refine those values as labels accumulate and preliminary truths emerge. The trouble, the authors argue, is that this methodology fundamentally mischaracterizes how human labeling actually works. A worker may be excellent at classifying photographs of animals but poor at distinguishing fine-grained sentiment in product reviews. A single global reliability number flattens that nuance, and the resulting misestimates cascade through the aggregation process, ultimately reducing the accuracy of inferred truths across the entire dataset.

The Xidian team’s alternative is to measure label quality at the level of individual instances, and to do so using two complementary statistical quantities: bias and variance. Bias, in their formulation, captures the discrepancy between a worker’s label on an instance and the labels that other workers assign to similar instances. A worker whose answers consistently diverge from the crowd on comparable tasks accumulates high bias, signaling a systematic departure from consensus. Variance, by contrast, measures the inconsistency of a worker’s own labels across different instances, exposing erratic behavior such as random clicking, fluctuating attention, or careless speed. Neither metric alone tells the full story: a worker can be consistently wrong, which is a bias problem, or inconsistently right, which is a variance problem.

By fusing these two components, the method computes a fine-grained estimate of each worker’s label quality for every inference instance, allowing the algorithm to assess actual labeling capability in context rather than in the aggregate. This per-instance reliability then becomes the foundation for the second half of the framework. GETI-WR constructs a graph that encodes the relationships among tasks, workers, and labels, the so-called task-worker-label graph, and maps it into a low-dimensional continuous vector space through graph embedding. In this embedded space, tasks and workers with similar labeling patterns end up close together, preserving the structural information of the crowdsourcing process in a form that machine learning models can exploit.

The crucial computational step comes next. The researchers employ a graph neural network, a class of deep learning models designed to operate directly on graph-structured data by passing and aggregating information between connected nodes, to transform the crowdsourcing problem into a graph-node prediction task. Instead of treating truth discovery as an iterative statistical estimation problem, the network learns to classify nodes in the embedded graph, effectively predicting the true label of each task from the learned representations of the tasks, the workers who labeled them, and their reliability-weighted relationships. This reframing is what gives the method its efficiency: once trained, the network can propagate information across the graph and enhance truth-discovery performance in ways that pairwise voting schemes cannot.

The empirical case for the approach rests on experiments with two real-world crowdsourced datasets. The authors benchmarked GETI-WR against six state-of-the-art baseline methods for truth inference and report superior performance on both accuracy and F1-score, the harmonic mean of precision and recall that is widely used to evaluate classification quality when class distributions are imbalanced. The consistency of the improvement across datasets and metrics suggests that the bias-variance decomposition of worker reliability is capturing genuine signal about label quality that global reliability scores miss, and that the graph neural network is able to translate that signal into better final predictions.

The significance of the work extends beyond a single benchmark. Truth inference sits at the foundation of the data supply chain for artificial intelligence, and the field has a long history stretching back to early truth-discovery methods for resolving conflicts in heterogeneous data sources and to influential surveys asking whether the problem had already been solved. Prior research has explored worker similarity, label similarity, label confidence clustering, meta-knowledge transfer, deep clustering-based aggregation, and reliability-driven multi-view graph embedding, each capturing a different facet of the aggregation challenge. The new study’s contribution is to combine a statistically motivated, instance-level reliability measure with the representational power of graph learning, addressing a documented weakness in the iterative reliability estimation paradigm that underpins many earlier systems.

There are practical implications as well. Crowdsourcing platforms could, in principle, integrate per-instance reliability estimation into their quality control pipelines, down-weighting suspicious labels in real time and routing tasks toward workers whose bias and variance profiles match the task type. For machine learning teams building training sets, better truth inference means cleaner labels without additional annotation cost, which in turn can improve downstream model performance. The method’s reliance on graph neural networks also aligns with a broader trend in the field: heterogeneous graph neural networks and graph contrastive learning have both been applied to label aggregation in recent years, indicating that graph-based representations are becoming the dominant lens through which crowdsourcing quality is analyzed.

The study, published in Knowledge and Information Systems as volume 68, article 251, was supported by the National Natural Science Foundation of China and the Fundamental Research Funds for the Central Universities, with additional support from the Innovation Capability Support Program of Shaanxi. The authors report no competing interests. As artificial intelligence systems increasingly depend on human-labeled data at massive scale, methods like GETI-WR point toward a future in which the wisdom of the crowd is not taken at face value but carefully audited, one label at a time, with statistical rigor and learned representations working in concert to separate reliable signal from human noise.

Subject of Research: Truth inference in crowdsourcing via bias-variance fusion graph embedding

Article Title: Truth inference for crowdsourcing via bias-variance fusion graph embedding

Article References: Truth inference for crowdsourcing via bias-variance fusion graph embedding. (n.d.). https://doi.org/10.1007/s10115-026-02868-1

Image Credits: AI Generated

DOI: 10.1007/s10115-026-02868-1

Keywords: crowdsourcing, truth inference, worker reliability, graph embedding, graph neural networks, bias, variance, label aggregation, data labeling, machine learning, Truth, inference

Nature-based interventions, often abbreviated as NBIs, have proliferated in recent years as policymakers and communities grapple with the twin crises of mental ill-health and environmental degradation. These programmes range from structured green prescriptions issued by health services to participatory science initiatives and community gardening schemes. Despite their wildly different formats, most share two common ambitions: promoting psychological wellbeing and strengthening what researchers call nature connectedness, the subjective sense of closeness to the natural world. Yet the sector has been repeatedly criticised for lacking a consistent evidence base, because countless projects evaluate themselves with different methods shaped by individual funder requirements, producing scattered datasets that cannot be compared, combined, or analysed with any statistical power.

The new study, led by Michael L. Lengieza of Durham University together with Elizabeth Richardson of Natural England and Miles Richardson of the University of Derby, tackles this fragmentation head-on. The team presents a centralised evaluation tool, developed by Natural England, that allows individual projects to administer a standardised survey, enter key features of their programme, and feed participant data into a shared dashboard. Larger organisations gain the ability to combine data across many initiatives, while smaller organisations, which often lack the expertise to design their own evaluations, receive a ready-made solution that removes the burden of survey design, ethics applications, and basic analysis.

As a proof of concept, the researchers pooled data from 75 members of the public who participated in one of eight real-world nature projects and completed both baseline and follow-up surveys. Nature connectedness was measured using an adapted version of the inclusion of nature in self scale alongside the statement-based item measuring how much people feel part of nature, while wellbeing was captured with the Office for National Statistics personal wellbeing questions. Project coordinators also completed a survey describing their programme’s features, including whether they used a recognised delivery framework such as the pathways to nature connectedness.

The results were striking. Using multi-level statistical models with random intercepts for both participants and projects, the team found that nature connectedness increased significantly between the start and end of the programmes across all eight initiatives. But the more revealing discovery emerged when the researchers tested whether the size of that increase depended on programme characteristics. Projects that explicitly employed the pathways to nature connectedness framework, which identifies five routes to connection through sensory contact, emotion, beauty, meaning, and compassion, showed a robust and statistically significant rise in participants’ nature connectedness. Projects that did not use the framework showed only a small, non-significant change, suggesting they might not have been effective at all.

Crucially, the analysis showed that pathways-based and non-pathways-based projects attracted participants with similar baseline levels of nature connectedness, meaning the difference in outcomes could not be explained by one type of programme simply recruiting more devoted nature lovers. The type of engagement, not the type of audience, appeared to make the difference. This pattern aligns closely with the existing academic literature, which has repeatedly shown that actively noticing nature, attending to its beauty, and engaging the senses produce stronger gains than mere time spent outdoors. Analysing any single one of the eight projects in isolation would have been statistically impossible; only the pooled data made the comparison feasible.

The authors are careful to stress what the data cannot show. Because the study lacked experimental control, causality cannot be established. People who seek out nature-connection programmes might be making other lifestyle changes that drive the effect, and programmes using the pathways framework might simply be more enthusiastic or rigorous in ways that matter more than the framework itself. Nevertheless, the researchers note that pre-post time-series designs have a long and respected history in public health evaluation, and that the absence of a change would have more definitively ruled out a causal link. The observed increase, combined with the moderation by framework type, signals exactly where the sector should direct its next wave of controlled experiments and its resources.

The study also offers practical guidance for scaling the approach. Data must include at least one baseline and one follow-up measurement, ideally collected at similar intervals relative to each project’s start and end. Because crowdsourced data will inevitably be noisier than laboratory data, larger samples are essential, and null effects should be interpreted with caution since they may reflect insufficient statistical power rather than a true absence of effect. The tool’s live dashboard, built on Power BI, already gives project managers simple benchmarked summaries of their own performance against the sector, while allowing policymakers and funders to view aggregated results and export datasets for formal analysis.

Looking ahead, the authors envision a modular future for the tool, with optional modules capturing additional outcomes such as physical activity and loneliness, demographic variables, and psychological factors like mindfulness and emotion. They also argue that national and regional bodies, including environmental funders, should move beyond publishing evaluation frameworks and instead provide practical tools that support reporting and data combination. By reducing the evaluation burden on small organisations, the approach frees resources for delivery while building the scalable, policy-relevant evidence base that nature-based interventions need to influence decision-makers at national and international levels, from the Dasgupta Review to the Kunming-Montreal global biodiversity framework.

For a sector long dismissed as fragmented and under-evidenced, the message of this proof of concept is quietly transformative: the collective data of many small initiatives, pooled intelligently, can illuminate what works, guide investment, and ultimately amplify the social and ecological benefits of humanity’s reconnection with nature.

Subject of Research: Crowdsourced evaluation of nature-based interventions using pooled nature connectedness and wellbeing data

Article Title: Amplifying social benefits from multiple nature-based initiatives by crowdsourcing nature connectedness and wellbeing data: A proof of concept

Article References: Lengieza, M. L., Richardson, E., & Richardson, M. (2026). Amplifying social benefits from multiple nature-based initiatives by crowdsourcing nature connectedness and wellbeing data: A proof of concept. Ambio. https://doi.org/10.1007/s13280-026-02472-1

Image Credits: AI Generated

DOI: 10.1007/s13280-026-02472-1

Keywords: nature connectedness, nature-based interventions, crowdsourcing, programme evaluation, wellbeing, pathways to nature connectedness, Natural England, green prescriptions, environmental psychology, evidence-based practice, biodiversity policy, public health

Pharmaceutical delivery differs from ordinary parcel logistics in critical ways. A late medicine can interrupt treatment, a temperature excursion can ruin the product, and a substituted or falsely confirmed delivery can harm a patient. Between a distributor and the final pharmacy or patient, traffic, varied vehicles, returns, and fluctuating demand make these obligations harder to satisfy. The Moroccan and wider North African context adds further complications, since small pharmacies and independent carriers may have uneven digital capacity, and any production system must comply with pharmaceutical, personal-data, and electronic-trust regulations.

The researchers’ answer is a ledger-grounded agent architecture in which the blockchain, not the language model, remains the authoritative source of operational truth. Smart contracts govern identity status, order assignments, capacity bookings, delivery evidence, and returns, while a real-time index and a Neo4j graph database organize contract events into retrievable relationships. An AI agent can only act on an order, carrier, booking, or delivery status if that record exists in verified platform data, substantially reducing the risk of decisions built on fabricated or unsupported information.

Four specialized agents divide the work. A Main Agent monitors the operation, retrieves similar past scenarios from graph memory, and coordinates specialist workflows. A Forecasting Agent estimates order arrivals, warehouse readiness, dispatch needs, and the risk of missing wave cutoffs. A vehicle-routing agent constructs feasible route plans and evaluates urgent insertions, while a Dispatch Agent carries out carrier checks, assignments, bids, bookings, and approved state changes in sequence. Numerical services compute forecasts and routes using portfolios of statistical and deep-learning forecasting methods and optimization solvers including ant colony optimization, particle swarm optimization, genetic algorithms, simulated annealing, proximal policy optimization, and deep Q-learning.

The routing logic applies lexicographic priorities rather than blending metrics through arbitrary weights: unassigned urgent orders come first, then unassigned standard orders, then minutes of lateness, and finally distance as a tiebreaker. Feasibility conditions enforce capacity limits, vehicle and temperature compatibility, valid carrier identity and eligibility, booking consistency, and locked route prefixes. Motorbikes carry either cold-chain or ambient products within a route but never both, while suitably equipped cars can mix both classes when packaging and capacity constraints permit. Urgent orders are prepared within 30 to 45 minutes and dispatched immediately after staging, while standard orders consolidate into waves at 10:00, 15:00, and 17:00.

Crowdsourced carriers introduce a trust problem that the framework tackles through verifiable credentials. Enrollment requires an authorized issuer to check official identity documents, driving licences, photographs with liveness checks, vehicle registration, insurance, and any cold-chain qualification. The approved carrier signs a one-time challenge to prove control of a wallet key, and the issuer issues a short-lived W3C verifiable credential binding the carrier, wallet, and capabilities. The blockchain stores only the credential fingerprint, issuer, validity period, and revocation status; personal documents and photographs remain encrypted off-chain. Pickup, delivery, and return events then link product scans, carrier signatures, time and location evidence, and pharmacy confirmations to the same digital identity.

Evaluation used five fixed-seed simulated operating days totaling 1600 orders, 40 carriers per day, 288 urgent orders, and 448 cold-chain orders, with each day replayed once through the human-operated baseline and once through the agent-optimized flow under identical conditions. The agent flow reduced route time by 25.0 percent and planning time by 78.6 percent, with consistent improvements in distance, urgent-order handling, assignment readiness, and proof completeness. A composite efficiency index reached 126.4 against the baseline’s 100, and the favorable direction remained stable across sensitivity analyses of the weighting scheme. Smart-contract functions were benchmarked separately on a local Ethereum virtual machine and the Sepolia testnet, quantifying gas, fees, and confirmation times for registration, order creation, booking, bidding, pickup, proof, and completion.

The authors caution that the evidence comes from simulation, not production. Five day-level replications with one distribution center and synthetic pharmacy profiles cannot capture real traffic, carrier misconduct, connectivity loss, sensor behavior, or user interaction, and the study did not conduct a dedicated hallucination test. Cryptographic credentials cannot guarantee document authenticity, prevent credential lending, stop a stolen key before revocation, or certify that the approved person physically performed every delivery. The next phase will involve the collaborating Moroccan distributor, first replaying live ERP and warehouse events in shadow mode, then running a monitored single-center pilot with vetted carriers and pharmacies, and finally scaling load across centers, carriers, and transactions.

For deployment, the researchers propose a phased production architecture that keeps personal data off-chain, uses a permissioned ledger among authorized partners, isolates external model providers behind an API privacy gateway, and routes agent recommendations through a human approval boundary before any smart-contract commitment. Compliance reviews must cover Moroccan pharmaceutical law, personal-data protection under Law 09-08, and electronic trust services under Law 43-20, while WHO good distribution practices make calibrated sensors, excursion alerts, and documented corrective actions operational requirements rather than optional features.

Beyond the specific Moroccan case, the study offers a template for adding operational intelligence to blockchain-based traceability without sacrificing accountability: the ledger supplies durable facts, deterministic tools handle computation, the language model coordinates context and sequencing, and every consequential action passes through explicit rules, auditable evidence, and human oversight. As agentic AI systems move into safety-sensitive logistics, the authors argue that reliability must be measured, not assumed, and future work will include dedicated hallucination tests, model comparisons, and quantitative security evaluation of contracts, carriers, identity, and agent behavior.

The distinction between traceability and orchestration helps explain why earlier blockchain pharmaceutical systems, however robust their records, left dispatchers largely on their own. Prior platforms built on Ethereum and Hyperledger demonstrated that role-based contracts, distributed file storage, and IoT sensing could make missing or inconsistent custody events easier to detect, and recent systems have combined ledger provenance with counterfeit-detection classifiers and certificate validation. Yet none of these contributions centered on real-time dispatch under fluctuating crowdsourced capacity, which is precisely where the Casablanca team positions its work.

The choice of graph-based retrieval rather than conventional text retrieval reflects a deliberate design judgment. In a retrieval-augmented setup built on a property graph, an order can be linked to its products, required temperature class, booking, carrier, route, and delivery proofs as explicit relationships, so the agent retrieves structured context rather than isolated passages of text. This matters in logistics, where the meaning of a record depends on its position in a web of custody events, and it reduces the chance that an agent reasons from fragments disconnected from the operational state.

The architecture also acknowledges that tool interfaces themselves create an attack surface. Malicious or misleading tool schemas, compromised servers, and excessive permissions can all corrupt an agent’s behavior, which is why the framework relies on allow-lists, authentication, input validation, and trace logging for every external call. This defensive posture extends to the model’s own tendencies: recent evaluations of language-model factual recall penalize confident wrong answers and reward appropriate abstention, a principle the authors carry into their design by ensuring the agent can abstain or escalate rather than guess when verified data is absent.

The evaluation methodology deserves attention in light of the broader research landscape. Reviews of generative AI in logistics consistently find that empirical field evidence remains scarce and that simulation dominates the literature, a pattern this study follows knowingly. By replaying five fixed-seed days through both workflows under identical demand, carrier pools, and order mixes, the researchers created a controlled comparison in which differences in route time, planning time, and proof completeness can be attributed to the coordination method rather than to environmental noise. The separate benchmarking of contract functions on a local virtual machine and a public testnet similarly isolates the cost of on-chain accountability from the benefits of agent coordination.

The lexicographic routing priorities also connect to a long-standing debate in optimization practice. Weighted objective functions force planners to express trade-offs in commensurable units, and small weight changes can silently reorder priorities in ways operators cannot anticipate or audit. By ranking urgent assignments, standard assignments, lateness, and distance in strict order, the framework makes every routing decision explainable as a sequence of dominance checks, which aligns naturally with the accountability requirements of pharmaceutical custody and with the need for authorized staff to understand and contest agent recommendations.

The identity model likewise reflects emerging standards beyond the pharmaceutical domain. Binding a short-lived verifiable credential to a proven wallet key, storing only fingerprints and revocation status on-chain, and keeping documents encrypted off-chain follows the privacy principle that blockchain should attest to facts without becoming a repository of personal data. Expiry and revocation mechanisms acknowledge that carrier eligibility is a continuing state, not a one-time gate, addressing risks such as credential lending or stolen keys only partially, as the authors themselves note.

Ultimately, the study’s contribution lies less in any single technique than in the division of labor it enforces: durable facts from the ledger, deterministic computation from numerical services, contextual coordination from the language model, and consequential state changes gated by contracts and human approval. Whether that division survives contact with live traffic, real carriers, and regulatory inspection is the question the planned shadow-mode replay and monitored pilot are designed to answer.

Subject of Research: Integration of smart contracts and AI agents for secure, traceable crowdsourced last-mile pharmaceutical delivery

Article Title: Smart contracts and AI agents for secure last-mile pharmaceutical delivery through crowdsourcing

Article References: Nadime, K. L., Haidar, D., Benabbou, R., & Benhra, J. (2026). Smart contracts and AI agents for secure last-mile pharmaceutical delivery through crowdsourcing. Discover Artificial Intelligence, 6(1), Article 1128. https://doi.org/10.1007/s44163-026-02206-y

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02206-y

Keywords: blockchain, smart contracts, AI agents, last-mile delivery, pharmaceutical logistics, crowdsourcing, large language models, vehicle routing, cold chain, traceability, verifiable credentials, supply chain