Data Visceralization: Embodied Data Insights

• Data visceralization is a mode of representation that transforms abstract data into multisensory, embodied experiences, fostering visceral and intuitive understanding.
• It leverages immersive technologies like VR, MR, and on-body displays to reconnect data metrics with physical and emotional reality, enhancing qualitative insights.
• Practical systems employ precise mapping of data attributes into spatial, haptic, and affective cues, improving empathy, sensemaking, and situational awareness.

Searching arXiv for recent and foundational papers on data visceralization and closely related immersive analytics work. arXiv search query: "data visceralization virtual reality immersive analytics humane data representations" Data visceralization is a mode of data representation in which data are made experientially present rather than only symbolically legible. Foundationally, it has been defined as “a data-driven experience which evokes visceral feelings within a user to facilitate intuitive understanding of physical measurements and quantities,” and later broadened to the practice of turning abstract datasets into multisensory, embodied experiences that people can feel, navigate, and relate to at a human level (Lee et al., 2020, Kondo et al., 2023). Rather than treating abstraction as the sole route to understanding, it seeks to restore the “ground-truth” of what measures, units, risks, and personal attributes correspond to in lived reality. Across VR, MR, tangible interfaces, on-body displays, and multisensory systems, data visceralization is consistently positioned as complementary to conventional visualization: visualization supports analytical and quantitative reasoning, whereas visceralization supports qualitative, experiential, situated, and affective understanding (Lee et al., 2020, Kam-Kwai et al., 10 Apr 2026, Sturdevant et al., 2022).

1. Definition, scope, and conceptual boundaries

In the literature, data visceralization is defined by its attempt to make data felt bodily, emotionally, and experientially. Conventional visualization is described as effective for revealing trends, patterns, and comparisons, but also as prone to abstracting away the real-world referents of measures or the people behind aggregates. Data visceralization addresses that loss by emphasizing embodiment, presence, situatedness, and multisensory encounter (Lee et al., 2020, Kondo et al., 2023).

The concept is closely related to, but distinct from, several neighboring traditions. Relative to conventional data visualization, it does not reject charts, maps, or dashboards; instead, it supplements them with experiential forms that convey what units, quantities, and consequences correspond to in reality. Relative to data physicalization, it often replaces fabricated artifacts with virtual simulation or hybrid physical-virtual environments, thereby retaining experiential and haptic benefits while avoiding some fabrication constraints; however, physicalization can itself become visceral when one-to-one mappings are preserved (Lee et al., 2020, Kam-Kwai et al., 10 Apr 2026). Relative to immersive analytics, data visceralization uses VR, AR, or MR not only to place data in 3D space but to connect spatial, kinesthetic, and temporal experience to the underlying phenomenon or process. Relative to affective visualization and speculative design, it is narrower in aim: affect and speculation may host visceralization, but the central objective remains to restore felt correspondence between representation and lived or physical reality (Bhardwaj et al., 25 Jul 2025, Brehmer, 2023).

A recurring boundary condition concerns representational fidelity. The 2020 VR design-probe paper argues that visceralization is strongest when there is a one-to-one mapping between data and representation, with transformations such as scaling weakening the connection to ground truth even when they preserve relative comparison (Lee et al., 2020). Later work generalizes this concern to humane encodings, policy communication, and simulation-based systems: metaphor, embodiment, and affect are treated as epistemically useful only when their mappings remain decodable, contextualized, and transparent (Kondo et al., 2023, Bhardwaj et al., 25 Jul 2025).

2. Foundational principles of visceral understanding

A central theoretical principle is presence: the “being there” sensation produced by immersive media and embodied interaction. In the foundational VR account, presence allows users to viscerally “believe” they are experiencing a phenomenon rather than merely inspecting its representation, which helps restore intuitive understanding of physical measures and quantities (Lee et al., 2020). Subsequent systems extend this principle beyond pure VR. LandSAR, for example, combines AR overlays, VR viewpoints, and passive haptics from a 3D-printed terrain model so that slope, flow, friction, and barrier placement can be understood through perceptual–motor coupling rather than detached map reading (Kam-Kwai et al., 10 Apr 2026).

A second principle is that mapping choices should preserve unit sensemaking. The 2020 work identifies perceptual sweet-spots, contextual frames of reference, optional annotations, and guided viewpoints as key factors. Human-scale phenomena such as sprint speed or jump length are described as particularly suitable, while extreme scales and abstract quantities often require scaffolding and are more vulnerable to distortion from miniaturization or proxy mappings (Lee et al., 2020). The 2025 policy paper extends this point from physical measures to decision contexts, arguing that embodied and multisensory design can reduce psychological distance along spatial, temporal, social, and hypothetical dimensions, thereby changing how policy outcomes are mentally represented (Bhardwaj et al., 25 Jul 2025).

A third principle is that affect is not incidental but structured. DataGarden operationalizes humane interpretation through a “garden” metaphor in which each flower or tree is an individual and the whole scene is a community portrait; the metaphor is intended to cultivate empathy, engagement, and reflection while legends, tooltips, and grouping preserve sensemaking (Kondo et al., 2023). Brehmer’s essay on the weird and the eerie extends affective design further by arguing that outsideness, absence, silence, and subtle cues of agency can be deliberately orchestrated so that viewers feel uncertainty or suspicion of unseen forces in ecological and economic data (Brehmer, 2023). This suggests that visceralization includes not only realism and one-to-one scale, but also carefully bounded affective rhetoric.

A fourth principle is situatedness. On-body clinical visualization leverages the existing visual tradition of anatomy by mapping data to a virtual patient body at the loci where symptoms manifest; LandSAR grounds overlays on a physical terrain proxy; policy-oriented work recommends place-based walkthroughs, concrete incident narratives, and material artifacts tied to infrastructures that decision-makers know and inhabit (Presnov et al., 2019, Kam-Kwai et al., 10 Apr 2026, Bhardwaj et al., 25 Jul 2025). Across these cases, visceral understanding is not treated as generic immersion but as embodied reasoning anchored to a canonical spatial frame.

3. From abstract variables to experiential form

Operationally, data visceralization depends on explicit translation functions from data attributes to experiential features. DataGarden provides a direct example. Its pipeline is Survey → spreadsheet → mapping module → VR scene; each respondent becomes a discrete object in a virtual garden, faculty are mapped to trees, students to flowers, MBTI personality type determines primary color, and plastic usage level is encoded by the number of clouds associated with each object (Kondo et al., 2023). Perspective modes—ground-level, bird’s-eye, and top-down—connect individual and community scales, while interactive groupings reorganize objects into axis-aligned layouts for counting, comparison, and trend summaries. The resulting scene is static in composition, but locomotion, touch-triggered tooltips, legends, and regrouping create an embodied analytic loop.

A more computationally explicit variant appears in the VR tool for deep learning transparency. There, each MNIST image is mapped by an encoder to a 3D latent vector $z \in \mathbb{R}^3$, rendered as a cube in a navigable point cloud. The network comprises an encoder $f_\phi$, decoder $g_\theta$, and classifier $h_\psi$, optimized with a combined VAE-plus-classifier objective, while the visible “force” on a point is interpreted as the negative gradient of the loss with respect to its latent code, $\mathbf{F}_i \equiv -\nabla_{z_i}\mathcal{L}$ (Kath et al., 2023). Spatial position is thus mapped to latent representation, temporal updates to training iterations, and perceived attraction and repulsion to backpropagation gradients. The paper is explicit that these are metaphorical forces rather than a literal physics simulation, but the metaphor is designed to make learning dynamics tangible.

Clinical on-body visualization formalizes the mapping problem even further. The neurosurgical prototype introduces a three-level encoding with category mapping $M^c$, property-type-to-attribute mapping $M^t$, and property-value mapping $M^d_p$, alongside local injectivity, spatial injectivity, visibility constraints, and quantization detection (Presnov et al., 2019). In the spinal disc herniation case study, Radicular Pain is mapped to red, Paresis to purple, T-Reflex to green, Excretion Disorder to orange, and Sensory Disorder to cyan; intensity is encoded through saturation-brightness combinations, while textures encode trigger or paresthesia. Because anatomy predefines shape and position, the visualization uses hue, saturation, brightness, texture, transparency, and time rather than arbitrary plot geometry.

LandSAR shows how these mappings can be coupled to domain models and risk formulations. Its overlays use the simplified formulation $Risk = Hazard \times Vulnerability$, a barrier impact-force equation $F = \alpha \rho v^2 h_0 w$, and a velocity attenuation model $f_\phi$0 to color-code hazard and vulnerability layers rather than to produce certified engineering designs (Kam-Kwai et al., 10 Apr 2026). Velocity is then felt through motion speed and egocentric sweep, overtopping becomes a directly observed event, and barrier placement changes flow paths in real time through tracked tangible proxies.

Across these systems, the shared pattern is a translation from abstract variables to embodied cues such as locomotion, viewpoint, force-like motion, touch, passive haptics, narrative proximity, or humane metaphor. A plausible implication is that data visceralization is best understood not as a single modality but as a family of mapping strategies in which experiential variables become part of the representational grammar.

4. Representative systems and domains

The foundational 2020 paper demonstrates data visceralization through six VR design probes: Olympic sprint speeds on a real-scale track, long-jump distances, real-scale skyscraper heights, solar-system size comparisons at a fixed 1:40,000,000 scale, crowd quantities, and U.S. debt represented as stacks of \$100 bills (Lee et al., 2020). These probes establish the domain in which visceralization is most straightforward: physical measures with readily interpretable human or environmental referents such as speed, height, distance, density, and scale.

DataGarden extends the concept from physical measurement to social and community data. Its contribution is not merely immersive display but a modular pipeline for humane, anthropomorphic, community-scale portraiture. Flowers, trees, and clouds constitute a configurable visual vocabulary intended to reveal “the people behind the data,” while interaction design preserves analytic tasks such as count, compare, and summarize (Kondo et al., 2023). This work exemplifies a shift from unit sensemaking toward empathy-oriented, humanistic representation.

The VR deep learning tool applies visceralization to model transparency. Instead of visualizing model internals as static diagrams, it turns latent geometry, fine-tuning, and class separation into a navigable, temporally evolving environment. Users batch-label subsets through spherical gestural selection, and progressive labeling sharpens clusters over iterations (Kath et al., 2023). Here the object of visceralization is not an external physical phenomenon but the internal dynamics of representation learning.

LandSAR applies visceralization to environmental hazard analysis and situational awareness. It combines Digital Terrain Models, building footprints, historical landslide records, climate scenarios, real-time simulation, a Meta Quest Pro MR headset tethered to a desktop, SteamVR base stations, HTC VIVE trackers, and 3D-printed terrain models fabricated with HP Jet Fusion 540 and HP 3D High Reusability PA 12 (Kam-Kwai et al., 10 Apr 2026). The system’s distinctive feature is the synthesis of situated visualization, passive haptics, and computational steering: users touch terrain, teleport to first-person viewpoints, point for local statistics, and move tangible barriers to test mitigation strategies.

The clinical literature contributes a different form of visceralization: anatomically integrated in-place visualization for cooperative hospital work. Rather than immersion at room scale, it relies on a mobile Android prototype rendered with Vulkan, a refined anatomical avatar, projective proxies for small structures, image-space occlusion handling, and timeline navigation (Presnov et al., 2019). The key domain condition is that the data already possess stable anatomical referents, making the body itself the canonical spatial substrate.

Policy-oriented work enlarges the concept again. “Limits at a Distance” frames data visceralization as a design response to psychological distance in climate and planetary-boundary decisions, recommending multisensory cues, place-based scenarios, identifiable micro-stories, speculative artifacts, and hybrid media ecosystems that combine dashboards with VR/AR, installations, and story-based components (Bhardwaj et al., 25 Jul 2025). The essay on the weird and the eerie adds a communicative register in which missingness, silence, thresholds, anomalous hybrids, and delayed explanation become devices for making ecological or economic disruption felt (Brehmer, 2023). The multisensory “data vivification” agenda broadens the medium set further to sound, touch, muscular resistance, smell, balance, and voice dialogue, arguing that data can be mapped to the human Umwelt rather than to sight alone (Sturdevant et al., 2022).

5. Empirical findings, performance, and known limitations

The empirical base is heterogeneous and still relatively early-stage. The foundational VR study involved 12 university students, comparing desktop and VR versions of selected probes. All participants preferred VR for the sprint-speed and skyscraper examples, preferences were mixed for the debt example, and participants spent more time in VR by factors of 3.4×, 13.2×, and 1.8× for the three examples respectively (Lee et al., 2020). Reported benefits included believability, fear of heights, awe, and the feeling of being run past, while limitations included time spent hunting for good viewpoints, sparse annotations in some cases, and cybersickness for a minority when flying locomotion was used.

DataGarden reports a preliminary user study with five participants performing five tasks involving counting, comparing, analyzing relationships, and summarizing trends. All participants successfully completed the tasks, the average System Usability Scale score was 66.5 on a 0–100 scale, and interviews indicated engagement through immersion, more positive and meaningful experience through movement and aesthetics, and perceived potential for “humanizing data to encourage empathy and understanding” (Kondo et al., 2023). At the same time, several participants felt overwhelmed by the amount of information and found navigation, the legend, and the menu challenging; prior gaming or VR experience correlated with easier navigation and control mastery.

LandSAR’s evaluation emphasizes situational awareness rather than conventional accuracy metrics. A formative workshop included 12 graduate students and junior researchers in civil engineering, with baseline SA means of 4.17 for Perception, 4.03 for Comprehension, 3.58 for Prediction, 3.53 for Preparedness, and 3.97 for Confidence on a 5-point Likert scale, plus UEQ and SART after the experience; expert interviews involved three senior geotechnical engineers (Kam-Kwai et al., 10 Apr 2026). The most impactful feature was reported to be the first-person perspective, which transformed participants from observers to stakeholders, while tangible terrain models improved depth perception and spatial accuracy. Reported challenges included coarse simulation mesh, passthrough-streaming smoothness, mild dizziness, and ethical concerns in stakeholder communication with high-fidelity local models.

The on-body clinical prototype was evaluated by 10 neurosurgeons. Purple in muscle was interpreted as paresis by 10/10 participants, red dermatome as pain by 8/10 and sensory disorder by 2/10, healing progress in paired visualizations was correctly interpreted by 10/10, and all 9 questionnaire respondents positively assessed the approach’s potential to enhance and accelerate information transfer in targeted cooperative situations (Presnov et al., 2019). However, tendon reflexes were often misinterpreted, most participants required hints to use filtering and zoom, and practical concerns remained regarding complexity growth and carrying an additional mobile device.

By contrast, the VR deep learning system is explicitly a demonstration rather than a user study. It reports qualitative observations that users can perceive topology before labeling and that clustering strengthens with progressive labeling, but it does not report controlled evaluation, latency figures, or dataset-size limits (Kath et al., 2023). More broadly, several papers note the absence of inferential statistics, effect sizes, or standardized measures of empathy, presence, and cognitive load. A recurring interpretation across the literature is that visceralization reliably improves engagement and qualitative appreciation, whereas its effects on exact quantitative comprehension, decision quality, or behavior change remain more contingent and methodologically underdetermined.

6. Ethics, misconceptions, and research directions

A persistent misconception is that visceralization is simply immersive visualization in 3D. The literature argues otherwise. What distinguishes it is not three-dimensional rendering per se, but the attempt to reconnect abstract representations to bodily understanding, human reality, physical dynamics, or situated consequences (Lee et al., 2020, Kam-Kwai et al., 10 Apr 2026). Another misconception is that stronger affect necessarily improves judgment. The climate-policy synthesis explicitly states that empirical results on psychological distance are mixed and that proximizing may change mental representation rather than reliably increasing action (Bhardwaj et al., 25 Jul 2025). Similarly, DataGarden notes that anthropomorphic encodings may nurture empathy but can also risk over-simplification or stereotyping, requiring transparent legends and cautious symbolism (Kondo et al., 2023).

Ethical concerns recur across domains. Humane or anthropographic representations raise issues of privacy, consent, stigmatization, and fairness when encoding personality, worldview, or sensitive community attributes (Kondo et al., 2023). Clinical on-body displays centralize highly legible patient data and therefore require strong authentication, encryption, role-based access, and protection against shoulder-surfing or public display (Presnov et al., 2019). Hazard and policy systems introduce another ethical axis: visceral experiences can be persuasive, potentially distressing, and easy to overread. LandSAR states that it is an evaluation probe for sensemaking and situational awareness rather than a design-certification tool, and the weird/eerie essay recommends eventual reintroduction of provenance, uncertainty, and explanatory context after affective engagement so that ambiguity does not become misinterpretation (Kam-Kwai et al., 10 Apr 2026, Brehmer, 2023).

Technical limitations are equally recurrent. One-to-one mappings are ideal but often difficult to preserve for extreme scales, abstract values, dense scenes, or high-cardinality domains (Lee et al., 2020, Presnov et al., 2019). VR and MR systems face occlusion, crowding, latency, control complexity, and comfort constraints; tangible systems face fabrication limits; multisensory systems face immature hardware, standardization problems, and accessibility trade-offs (Kam-Kwai et al., 10 Apr 2026, Sturdevant et al., 2022). Rich immersive scenes can increase cognitive load even when they also increase engagement.

The forward agenda described in the literature is therefore both methodological and design-oriented. It includes scalability and interpretability for larger communities and denser scenes; multisensory expansion through sound, haptics, and temporal encoding; rigorous empathy and presence evaluation; non-VR or hybrid embodiments for broader accessibility; collaborative or shared VR sensemaking; stronger workflow integration with dashboards and decision tools; and co-designed, reflexive practices that document values, assumptions, and uncertainty (Kondo et al., 2023, Bhardwaj et al., 25 Jul 2025, Sturdevant et al., 2022). A plausible implication is that the most durable future of data visceralization lies not in replacing analytical visualization, but in developing principled hybrid systems in which symbolic, embodied, affective, and situated forms of knowing are intentionally combined.