Welcome to this week’s issue of At the Intersections. This edition brings together investigations into hidden signals and structured systems, exploring how standard cameras register physiological markers of deception, how the brain tracks sound that exists only in the imagination, and how computational frameworks combine theory and clinical data to forecast disruptions in reality. In parallel, the focus turns to how institutions manage uncertainty, examining the role of intellectual humility in business education, statistical methods for identifying the mechanisms behind workplace policies, and the transition of global strategies for mitigating flood risks.
Sources
- Learning Video-Derived Behavioral and Physiological Representations for Automated Deception Detection
- Why being right can be wrong: intellectual humility as a key driver of responsible management education
- Registered Report: Replication and Extension of Nozaradan et al. (2011)
- Uncovering the Mechanisms of Policy Interventions: A Difference‐In‐Differences Framework for Mediation Analysis
- Flood management in transition: A systems-based review of traditional practices, modern approaches, and future strategic directions
- Integrating theory-driven and data-driven computational psychiatry
- Read the full issue
Full transcript
Everyday video cameras can identify internal physiological shifts that reveal when someone is being deceptive. Examining how hidden dynamics shape diverse disciplines is what we do on ComplexityPod. Here is this week's edition of At the Intersections.
On ComplexityPod, the focus turns toward extracting hidden signals from human behavior, beginning with video feeds in remote settings like legal hearings and job interviews.
Where the technical hurdle has always been the equipment. You either rely on webcams and accept low predictive accuracy from surface movements, or you require specialized hardware like thermal cameras that simply do not scale.
Which is why the DeceptSense framework designed with Boyu Wang attempts to extract those physiological signals directly from standard color video.
It pulls three distinct signals simultaneously: physical motion using dense trajectories, heart rate via remote photoplethysmography, and skin temperature changes estimated from the visible color spectrum.
And because heart rate, temperature shifts, and physical gestures operate on completely different time scales, the framework uses a cross-modal attention mechanism to align them in time.
Though the empirical performance varies considerably across benchmarks. On the Real-Life Trial dataset, the area under the receiver operating characteristic curve reached 89.6 percent, but dropped to 64.5 percent on BOL and 65.1 percent on MU3D.
That discrepancy shows how sensitive these multimodal models remain to the specific conditions of the recording.
Internal states present an even harder measurement problem when the signal is purely mental, as in Jessica Grahn's research on whether the brain's tracking of rhythm reflects physical sound or internal imagination.
This revisits the 2011 experiment by Nozaradan, where electroencephalography recorded heightened activity at frequencies matching an imagined binary or ternary meter over an unaccented beat.
Except that original finding came from a sample of only eight individuals. When a registered report replicated the experiment across thirteen independent sites and 152 participants, the raw effect size collapsed to 0.03 microvolts for both meters, with confidence intervals spanning zero.
Down from the 0.12 and 0.20 microvolt effects originally reported. And when they added a behavioral task to verify whether participants actually perceived the imagined meter, the neural activity at the imagined frequencies failed to predict performance.
Only the frequency of the physical sound stimulus predicted whether participants stayed on task. That outcome directly challenges the reliability of frequency tagging for measuring imagined auditory structure.
It also demonstrates why neural theories require rigorous empirical verification, a core tension in Lena Palaniyappan's work on computational psychiatry.
Where theory-driven cognitive simulations and large-scale data-driven predictions typically run on separate tracks. One explains mechanisms without clinical utility, and the other predicts outcomes without explaining why.
The proposed framework tests psychosis onset by feeding mechanistic parameters directly into predictive algorithms as computational features.
So the machine learning model does not just look for statistical correlations across patient records; it evaluates parameters derived from formal models of neural processing, keeping the predictions interpretable for clinicians.
Moving from clinical interventions to workplace policies, measuring cause and effect runs into the exact same problem of unobserved variables, which Anders Holm addresses in organizational mediation analysis.
Most human resource studies rely on sequential ignorability, assuming that hidden factors do not skew the path from the policy change to the mediator and down to the employee outcome.
To circumvent that assumption, Holm developed a two-period difference-in-differences design. Instead of assuming baseline independence, it requires parallel untreated trends and exchangeability based on changes over time.
Which allows researchers to mathematically separate the overall policy effect from the controlled direct and natural indirect paths, even when implementations occur across staggered schedules.
Yet analyzing organizational outcomes requires questioning how decisions are taught in the first place, which Pratima Bansal examines in business school curricula.
Business education routinely conditions students to isolate a single optimal answer for an operational scenario, ignoring how policies shift when applied across different scales, geographies, or timeframes.
And institutional incentives like school rankings, standardized grading, and faculty tenure metrics reinforce that search for uniform answers.
Bansal points to initiatives like the United Nations Principles for Responsible Management Education to advocate for intellectual humility—teaching students that management decisions carry conditional boundaries rather than universal certainty.
That shift away from static assumptions mirrors the historical transformation in flood management documented by Slobodan Simonovic.
Floods account for 35 to 40 percent of all weather-related natural disasters, impacting roughly two billion people. Yet prior to 2000, management depended almost entirely on concrete channels and levees, calculating risk on the assumption that historical hydrology was stationary.
Over the past two decades, modeling moved toward integrated, probabilistic frameworks that incorporate remote sensing, high-resolution topography, and non-structural adaptations.
Yet the systemic vulnerabilities remain severe: aging physical defenses, administrative fragmentation across shared river basins, and non-stationary climate conditions that break historical hydrological limits.
Which points toward Simonovic's framework for future policy: combining adaptive physical infrastructure, community resilience strategies, and basin-wide governance that crosses jurisdictional boundaries.
More research summaries exploring complex systems will return next week. From ComplexityPod, thanks for listening.
From our community
At the Intersections — latest issue
Welcome to this week’s issue of At the Intersections. This edition brings together investigations into hidden signals and structured systems, exploring how standard cameras register physiological markers of deception, how the brain tracks s