Agent Tomography
Agent Tomography
System Identification Across an Entire Life
Position note · multimodal reconstruction, embodiment, lifelogging, digital twins, and continuity
Version 1 · July 2026
Abstract
This position note introduces agent tomography: the idea that a particular embodied agent may be understood as a hidden, temporally extended generative process through the many traces it leaves across a life. A brain scan is one projection; others include body structure and physiology, genome, language, choices, sensorimotor habits, social relations, and the recorded environment. These sources are not equally important, and inference cannot recover distinctions absent from every measured channel. The narrower claim is that partially independent views can reduce ambiguity about the agent that produced them, allowing one modality to constrain another. The note develops that conceptual frame, places it beside personality capture, lifelogging, digital phenotyping, human digital twins, embodied cognition, and whole-brain emulation, and distinguishes reconstruction fidelity from personal continuity. It does not present an engineering design or a research program. It names a possible direction that others may choose to pursue.
STANDING A conceptual position, not an empirical result, engineering proposal, or personal research agenda. Its bounded claim is that several unequal and partially independent projections may constrain a particular agent more strongly together than any one view alone.
1. The object: a hidden generative process
The object to be reconstructed is not an image, a transcript, a genome, a connectome, or a trait vector. It is the organized process that generated them: a particular agent with a body, a developmental history, an environment, and a characteristic way of transforming perception into prediction, regulation, learning, and action. The term agent is intentionally species-neutral. The target may be a human, a parrot, another animal, or a future synthetic organism; the available priors and relevant modalities will differ, but the inverse problem has the same form.
The motivating analogy is the difference between a photograph and a three-dimensional model. A photograph is not false; it is a projection from one angle. Multiple views constrain a model that can explain them together. In an agent, however, the views are not merely spatial. They occur across scale, modality, and time. Neural structure, physiology, genome, movement, language, relationships, and lifetime behavior are different transformations of the same developing system. Their joint value lies not in accumulation by itself but in the intersections among them.
This does not imply democratic weighting of evidence. Some variables are far more load-bearing than others for reconstructing particular memories, dispositions, and cognitive dynamics. Fine-grained neural organization is likely to dominate many such questions. Genome is a strong developmental prior but is not a record of acquired memories. A body scan captures consequences of development and coupling but not every neural state. Behavioral history may reveal stable policies while leaving their implementation underdetermined. Agent tomography is therefore hierarchical: it asks what each projection constrains, what it leaves free, and whether another projection closes the remaining degrees of freedom.
2. Why the body belongs in the problem
A brain-only program risks treating the body as replaceable packaging around cognition. Yet the body is the first suit of the mind: the coupled system the agent learned to regulate and operate, often for decades. Proprioception, interoception, endocrine and immune dynamics, motor morphology, sensory surfaces, injury, habit, and metabolic history participate in the agent's learned control loops. Even where these variables are not constitutive of a memory, they can constrain the state and dynamics in which that memory was formed and expressed. Embodied and extended approaches to cognition supply the philosophical precedent for taking these couplings seriously (Varela, Thompson, & Rosch, 1991; Clark & Chalmers, 1998).
Moving from brain coverage to whole-body coverage need not mean lowering every measurement to one uniform resolution. There is no single scalar called resolution here. Spatial, temporal, chemical, electrical, and molecular resolution trade differently. A plausible architecture is multiscale and multimodal: exceptionally high resolution where causal sensitivity is greatest, coarser continuous measurement elsewhere, and targeted refinement when anomalies or uncertainty demand it. The full-body extension is therefore not one enormous scanner. It is an aligned stack of instruments and models operating at different scales.
Can computation reverse the apparent loss created by greater scale? Sometimes, conditionally. If several views constrain the same hidden variable, a learned prior and complementary measurements can reduce uncertainty below what any one instrument allows. But no algorithm can guarantee recovery of distinctions that leave no trace in any measured channel. AI can solve an underdetermined inverse problem only to the extent that lawful regularities and additional evidence make it less underdetermined. Plausibility is not recovery, and a population-typical completion is not automatically the particular person.
3. The projection set
If the idea were pursued, each modality could be treated as a view with a particular domain of sensitivity, characteristic noise, sampling cost, and failure mode. The following map is illustrative rather than exhaustive.
| Projection | What it may constrain | Principal limitation |
|---|---|---|
| Neural structure and dynamics | Connectivity, cell types, synaptic and molecular state, activity patterns | Potentially most load-bearing for memory and cognition; currently incomplete across spatial, temporal, and chemical scales |
| Whole-body structure and physiology | Sensorimotor morphology, interoception, endocrine, immune and metabolic coupling | Large state space; relevance varies by function; many variables change rapidly |
| Genome and molecular history | Developmental constraints, inherited variation, some regulatory priors | Does not encode acquired life state; epigenetic and somatic history are only partly recoverable |
| Naturalistic behavior | Policies, habits, preferences, skills, response regularities | Confounded by circumstance; unobserved capacities may never be expressed |
| Language and symbolic products | Self-model, concepts, commitments, memories, style and social reasoning | Selective and strategic; fluent reconstruction can imitate surface form without causal depth |
| Social and relational record | Roles, attachments, reputation, reciprocal expectations and external memory | Other people are noisy observers and rights-bearing participants, not passive sensors |
| Environmental and egocentric record | Inputs received, opportunities available, events to which responses were made | Recording is incomplete and perspective-bound; reconstructing context requires inference |
| Active probes | Responses to standardized or adaptively selected tasks and questions | High information potential but costs time, effort, ecological validity, and sometimes dignity |
The list is deliberately broader than conventional imaging. The outside record matters not because everything outside the skin is necessarily part of the person, but because effects can identify causes. External data may be epistemically useful even when it is not constitutive of the agent. Agent tomography should therefore remain neutral between a strongly extended ontology of mind and the weaker claim that environmental and social traces provide additional projections of an embodied system.
4. Observation, self-report, and active probing
Bainbridge's program of personality capture includes massive questionnaires, cognitive testing, autobiographical memories, text analysis, recommender histories, mobile and ubiquitous capture, and activity in virtual worlds (Bainbridge, 2014). The breadth is right. The modalities should not, however, be treated as equivalent in burden or epistemic status. A life recorded during ordinary activity is different from a life repeatedly interrupted to answer an instrument.
4.1 Passive naturalistic capture
Passive capture observes the agent in the course of living: egocentric audio and video, location, movement, device interaction, language, physiological signals, environmental state, and the responses of other agents. Its advantages are longitudinal coverage, low repeated effort, and ecological validity. Its disadvantages are equally serious: privacy invasion, selection effects, uncontrolled confounding, sensor blind spots, and the possibility that a capacity important to the person simply never appears in the record. MyLifeBits established the lifetime-archive side of this problem, pursuing a searchable store of nearly everything that could be captured (Gemmell, Bell, & Lueder, 2006). Its objective was principally storage and retrieval, not inversion of the archive into the generator.
4.2 Self-report
Questionnaires and interviews are effortful and partly artificial. They are also not dispensable merely because they are subjective. A self-report is evidence about what the person believes, remembers, notices, conceals, or can formulate about themselves. Its distortions are themselves outputs of the system, though they must not be naively interpreted as transparent measurements of an inner fact. Rich interviews can be unusually information-dense: recent generative-agent work built cross-domain behavioral simulations of 1,052 individuals from two-hour interviews and surveys, reaching 86 percent of participants' own two-week test–retest consistency on held-out survey items in the combined condition (Park et al., 2024, rev. 2026). That result is a narrow behavioral benchmark, not 86 percent reconstruction of a person.
4.3 Controlled probes
Cognitive tasks and deliberately constructed situations have a different role. In system identification, passive observations may fail to excite parts of a system sufficiently for their dynamics to become identifiable. A controlled input can reveal a response function ordinary history never sampled. Tests can therefore be valuable not as ceremonial batteries administered to everyone, but as experiments selected to discriminate among remaining candidate models. They also introduce fatigue, practice effects, demand characteristics, and unequal accessibility. The system must model those costs instead of hiding them.
MEASUREMENT PRINCIPLE Observe passively by default. Maintain a posterior over what remains unknown. Request an active measurement only when its expected reduction in uncertainty justifies the burden, risk, and interference it imposes on the person. Refusal is a boundary to respect, not merely another signal to exploit.
In principle, an adaptive system could choose a probe by its expected information gain about the latent agent, minus explicit penalties for effort, intrusion, and ethical risk. The important point is not an exact objective function but the ordering it makes visible: active effort from the person being imaged is a real measurement cost. Any future instrument would have to spend that cost deliberately.
5. Agent tomography as an inverse problem
A useful way to frame agent tomography is as an inverse problem. The agent is the hidden process; neural measurements, physiology, language, behavior, and environmental records are different projections produced by that process. The aim would be to infer what organization could have generated the views together.
The central question is not how much data can be accumulated, but which ambiguities each additional view removes. A genome constrains development without recording acquired memories. Behavior reveals policies without uniquely specifying their implementation. A brain measurement may be the most load-bearing view while still leaving uncertainty that body and lifetime evidence can sometimes reduce.
Any reconstruction should preserve uncertainty and provenance: what was measured, what was inferred from the particular person's other traces, what came mainly from population priors, and what remains unknown. System-identification research supplies vocabulary for this framing (Schoukens & Ljung, 2019), and individualized health models show that relevant dynamics can differ between people (Hekler et al., 2020). Extending that framing to an entire agent is the position advanced here, not a result established by those precedents.
6. System identification across an entire life
A cross-sectional scan attempts to infer a dynamical system from a small temporal slice. A lifetime record adds trajectories: what inputs arrived, what adaptations followed, which capacities emerged or disappeared, and how the same system behaved under changing conditions. Time is not merely more data. It separates stable organization from transient state and reveals learning rules that a static measurement cannot show.
The ideal of literal total capture is incoherent if it means recording every physically relevant variable at arbitrary precision. Measurement has bandwidth, energy, interference, and storage costs, and the boundary of relevance depends on the reconstruction target. One conceivable target is an indexed version: capturing nearly all functionally relevant activity at sufficient spatial, temporal, and modal resolution. That is not obviously impossible, but it is not one specification. Sufficient for voice and preference prediction is not sufficient for autobiographical memory, motor skill, neural emulation, or organism-level reconstruction.
A lifetime observation system would also be part of the life it records. Cameras change behavior; prompts redirect attention; a trusted archive becomes external memory; a prediction system can alter the decisions it predicts. Agent tomography is consequently an identification problem with feedback, not observation from nowhere. The model must represent the measurement apparatus as part of the input history and distinguish the unobserved agent from the agent adapted to being observed.
7. What AI can and cannot contribute
AI would likely be central to any attempt because the observation set is too heterogeneous and incomplete for manual integration. A population model could learn regularities connecting neural organization, morphology, language, action, development, and environment. For a particular agent, those regularities could serve as priors through which one view constrains another. Several functions are plausible.
Cross-modal alignment: place brain, body, behavior, language, and environment on a shared temporal and causal index.
Environmental reconstruction: infer three-dimensional scenes and event structure from partial egocentric and external recordings, while retaining uncertainty about occluded regions.
Adaptive sensing: direct scarce resolution or request an active probe where competing reconstructions disagree most.
Compression by learned sufficient structure: preserve information needed for target predictions without retaining every raw bit at equal fidelity.
Counterfactual testing: ask whether the reconstructed agent generalizes to situations and tasks withheld from the reconstruction process.
Multiscale completion: use biological and behavioral priors to infer unmeasured structure, clearly separating such inference from direct observation.
The danger is prior domination. A powerful model can generate a coherent person-shaped completion even when the individual evidence is weak. Such a result may be culturally and statistically plausible while being wrong about the particular agent. Any serious attempt would therefore need to distinguish held-out prediction, intervention response, cross-modal consistency, and mechanistic correspondence from conversational persuasiveness alone.
8. Position within the literature
Agent tomography is best presented as a synthesis with a new target and inverse-problem formulation, not as the discovery that people leave reconstructable traces. Several literatures already own important pieces of the program.
| Tradition | Contribution | Where it stops |
|---|---|---|
| Personality capture and digital immortality | Bainbridge systematized questionnaires, ubiquitous capture, memories, text, cognitive abilities, recommender histories, and virtual worlds; virtual-human work explores posthumous personas. | Usually targets personality or conversational presence rather than the embodied causal generator. |
| Lifelogging | MyLifeBits and related work pursue lifetime stores of documents, media, communications, and experience. | Archive, search, and memory support do not by themselves identify the process that produced the archive. |
| Digital traces and phenotyping | Likes, smartphones, mobility, communication, and device use predict traits and psychological variables. | Predictions concern sparse labels and are vulnerable to context, confounding, and population bias. |
| Generative personal agents | Interview- and survey-grounded LLM agents predict held-out attitudes, traits, and experimental behavior across tasks. | Present validation is thin relative to a person and does not establish biological or mechanistic fidelity. |
| Human digital twins | Personalized computational models integrate clinical, sensor, imaging, lifestyle, and sometimes molecular data. | The field is predominantly medical and often organ- or disease-specific; many claimed twins are static models or shadows. |
| Whole-brain emulation | A detailed scan of a particular brain anchors an attempt to reproduce its internal causal dynamics. | The program privileges brain measurement and generally treats body and lifetime evidence as secondary or environmental. |
| Embodied and extended cognition | Body and environment can be causally—and on some accounts constitutively—integrated into cognition. | Provides ontology and motivation, not a reconstruction methodology. |
| System identification | Infers dynamical organization from input–output histories and designs informative perturbations. | Usually applies to bounded mechanisms or behaviors, not a whole agent across a life. |
The closest direct predecessor is Bainbridge's Personality Capture and Emulation (2014). Agent tomography inherits the proposal to combine heterogeneous personal evidence and explicitly attributes that inheritance. It broadens the target from personality to the embodied generator and gives the integration a system-identification form. It also treats measurement burden as part of experimental design: massive questionnaires and cognitive batteries are possible views, not mandatory rituals.
Digital-footprint studies establish a weak empirical premise: latent attributes can be predicted from ordinary records of behavior (Kosinski, Stillwell, & Graepel, 2013; Youyou, Kosinski, & Stillwell, 2015; Stachl et al., 2020). Generative personal agents extend this toward cross-task simulation (Park et al., 2024, rev. 2026). Human digital twins provide a framework for personalized, dynamically updated models linked to a physical counterpart (National Academies, 2024), but a 2025 scoping review found that only 18 of 149 included healthcare studies fully met those criteria and only two mentioned verification, validation, and uncertainty quantification (Tudor et al., 2025). The result is a field with serious components but no established method for reconstructing a whole agent.
Whole-brain emulation approaches from the mechanistic pole. Sandberg and Bostrom define it as detailed scanning of a particular brain followed by construction of a software model faithful enough to reproduce its causal dynamics and behavior (2008). Agent tomography asks a complementary question: when a brain measurement is incomplete, how much can the remaining ambiguity be reduced by the organism and life that same brain generated? The framework does not deny the brain's likely centrality. It denies that a single privileged view should be assumed sufficient before the identifiability question is tested.
NOVELTY CLAIM The components have substantial precedents. Among the literature surveyed, what appears to be missing is their unification as one inverse problem: reconstructing an embodied, developing generative agent from the heterogeneous projections it produced across an entire life.
9. Reconstruction is not mere imitation
A model can imitate surface behavior while implementing a different generator. Conversational resemblance is therefore a useful test but a poor definition of success. At least five increasingly demanding objects should be distinguished: an archive of the person; a descriptive profile; a predictor over familiar outputs; a causal model that generalizes under intervention; and a mechanistically faithful model capable of reproducing the relevant internal dynamics. These objects may coexist, but progress at one level should not be relabeled as success at another.
This distinction would matter in any future attempt. Holding out random survey answers is weaker evidence than holding out entire life domains. Reproducing familiar language is weaker than predicting a novel decision under known inputs. Matching an output is weaker than matching how that output changes after a perturbation. Agreement across modalities not used to construct the model would provide stronger evidence than resemblance alone.
A reconstruction can also be more complete in one dimension and less complete in another. A behavioral twin may capture social style while missing physiology. A mechanistic neural emulation may preserve learned dynamics while lacking the body's familiar feedback. There is no honest one-dimensional percentage of person recovered until the target functions and their weights are specified.
10. Reconstruction fidelity is not continuity
Agent tomography concerns qualitative and causal fidelity: how much of the hidden generator can be reconstructed from its projections. It does not by itself establish that a later implementation is numerically the same individual. The companion position note What You Actually Are argues that a person is a particular boundary-maintaining work-process; a scan followed by re-instantiation begins a second token, even if the pattern is faithfully inherited (Đurič, 2026). On that account, the result is a twin, not a transfer.
This distinction should be built into the terminology. A high-fidelity reconstruction may preserve memories, values, dispositions, skills, and a causal profile while still failing to continue the original process. Conversely, a continuously operating agent may change its material and pattern substantially while remaining the same token. Agent tomography can improve what a reconstruction inherits. Whether any downstream bioprinting, emulation, or substrate migration preserves the original is a separate question about the mechanism and continuity of the work, not about the quality of the resulting product (Corabi & Schneider, 2012; Schneider, 2019).
11. A possible direction
If someone chose to develop this idea, the most modest first question would be whether adding genuinely independent kinds of evidence reduces uncertainty about a particular agent, rather than merely making a model more generally human-like. That question could be approached in narrow domains long before any attempt at whole-person reconstruction.
Synthetic agents, constrained behavioral tasks, or nonhuman organisms might make parts of the question tractable. Dense whole-life capture, whole-body sensing, brain imaging, and bioprinting remain much more distant possibilities. They are mentioned here to locate the direction, not as a proposed engineering plan.
This note does not claim ownership of the larger undertaking. It supplies a name and a conceptual relation: a person is a hidden generative process; a brain scan is one view; a life leaves many views. Others may decide whether that relation is scientifically productive.
12. Conclusion
A brain scan is an extraordinarily important view of a person, but it is still a view. A life leaves others: a body shaped by development, a genome that constrained it, an environment that supplied inputs, a history of actions and language, and the models of the person carried by other agents and institutions. Agent tomography proposes that these be treated as coordinated projections of one hidden, embodied, changing process.
The proposal is holistic without claiming that everything matters equally. It is ambitious without calling inference measurement. It accepts active questionnaires and cognitive tasks as potentially powerful probes while refusing to make the person's labor disappear from the cost function. It uses AI to align, complete, and test many views, but requires the result to expose uncertainty and prior dependence. And it refuses a final conceptual shortcut: reconstructing the generator with high fidelity is not the same as continuing the token that generated the evidence.
The literature already contains many of the components: personality capture, lifetime archives, digital phenotyping, generative replicas, human digital twins, whole-brain emulation, embodied cognition, and system identification. Agent tomography is offered as a name for the conceptual object that appears when an entire life is treated as a set of projections rather than a pile of records. Whether that framing becomes scientifically productive is a question for those who choose to pursue it.
Standing of this paper
This is a position note, not a research-program note, engineering roadmap, or completed scientific result.
The strongest claim is the bounded one: different projections of one embodied agent can, when they carry partially independent information, constrain a reconstruction more strongly together than separately. The claim is conditional on identifiability and must fail where no measured channel carries the distinction at issue.
The whole-agent formulation is a conceptual horizon rather than a present engineering specification. Voice prediction, preference modeling, autobiographical memory, motor skill, physiological regulation, and mechanistic emulation would require different targets and measurements.
The literature claim is synthetic and provisional. Personality capture, lifelogging, digital phenotyping, generative agents, human digital twins, embodied cognition, and whole-brain emulation each contain substantial parts of the idea. The contribution claimed here is the relation drawn among them: a burden-aware, uncertainty-preserving inverse problem over a particular agent across a life.
The continuity claim is separable. Even a causally faithful reconstruction may be a new token rather than the continuation of the agent reconstructed. Agent tomography concerns what a successor can inherit; the companion note What You Actually Are addresses whether the original process survives.
The author offers this framing for others who may find the direction useful; it is not presented as a research agenda the author intends to undertake.
References
- Bainbridge, W. S. (2014). Personality Capture and Emulation. Springer. doi:10.1007/978-1-4471-5604-8 — The closest direct predecessor: a program for progressively higher-fidelity personality capture using questionnaires, ubiquitous recording, memories, text, cognitive measures, recommender histories, and virtual worlds. Source checked · July 2026
- Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19. doi:10.1093/analys/58.1.7 — The extended-mind thesis and the possibility that reliably coupled external structures participate in cognition rather than merely recording it. Standard
- Corabi, J., & Schneider, S. (2012). The metaphysics of uploading. Journal of Consciousness Studies, 19(7–8), 26–44. Source — The metaphysics of uploading and the argument that qualitative fidelity need not preserve numerical identity. Standard
- Đurič, A. (2026). What You Actually Are: What a Person Is, and Therefore What Survives Being Copied, Migrated, and Ended (Version 2). Position note. — The companion account of a person as a particular boundary-maintaining work-process, and therefore of reconstruction as twinning unless that work continues. Internal
- Gemmell, J., Bell, G., & Lueder, R. (2006). MyLifeBits: A personal database for everything. Communications of the ACM, 49(1), 88–95. doi:10.1145/1107458.1107460 — The canonical lifetime-store project: capture, organization, search, and retrieval of personal documents, communications, media, and experience. Source checked · July 2026
- Hekler, E. B., et al. (2020). Precision health: The role of the social and behavioral sciences in advancing the vision. Annals of Behavioral Medicine, 54(11), 805–826. doi:10.1093/abm/kaaa018 — Individualized dynamical modeling and system identification in precision health, demonstrating that predictors and tailoring variables can differ by person. Source checked · July 2026
- Kosinski, M., Stillwell, D., & Graepel, T. (2013). Private traits and attributes are predictable from digital records of human behavior. Proceedings of the National Academy of Sciences, 110(15), 5802–5805. doi:10.1073/pnas.1218772110 — Evidence that sensitive latent attributes can be predicted from ordinary digital records of behavior. Source checked · July 2026
- National Academies of Sciences, Engineering, and Medicine. (2024). Foundational Research Gaps and Future Directions for Digital Twins. National Academies Press. doi:10.17226/26894 — The current consensus definition of a digital twin and the associated requirements for dynamic updating, prediction, validation, and uncertainty quantification. Source checked · July 2026
- Park, J. S., Zou, C. Q., Kamphorst, J., et al. (2024; revised 2026). LLM agents grounded in self-reports enable general-purpose simulation of individuals. arXiv:2411.10109. doi:10.48550/arXiv.2411.10109 — A direct empirical precursor to behavioral reconstruction: interview- and survey-grounded LLM agents tested on held-out responses and cross-domain behavior. Source checked · July 2026
- Sandberg, A., & Bostrom, N. (2008). Whole Brain Emulation: A Roadmap. Future of Humanity Institute, Technical Report 2008-3. Source — The classic roadmap for scanning a particular brain and reproducing its internal causal dynamics in software. Source checked · July 2026
- Savin-Baden, M., & Burden, D. (2019). Digital immortality and virtual humans. Postdigital Science and Education, 1, 87–103. doi:10.1007/s42438-018-0007-6 — Virtual humans, personality capture, digital legacies, and the distinction between static memorials and adaptive posthumous personas. Source checked · July 2026
- Schneider, S. (2019). Artificial You: AI and the Future of Your Mind. Princeton University Press. Source — The philosophical and practical danger of treating a functionally convincing digital successor as automatic personal survival. Standard
- Schoukens, J., & Ljung, L. (2019). Nonlinear system identification: A user-oriented road map. IEEE Control Systems Magazine, 39(6), 28–99. doi:10.1109/MCS.2019.2938121 — The system-identification foundation: inference of nonlinear dynamics from input-output histories and the design of informative experiments. Standard
- Stachl, C., Au, Q., Schoedel, R., et al. (2020). Predicting personality from patterns of behavior collected with smartphones. Proceedings of the National Academy of Sciences, 117(30), 17680–17687. doi:10.1073/pnas.1920484117 — Prediction of Big Five traits from communication, mobility, activity, app-use, music, and other smartphone-derived behavioral patterns. Source checked · July 2026
- Tudor, B. H., Shargo, R., Gray, G. M., et al. (2025). A scoping review of human digital twins in healthcare applications and usage patterns. npj Digital Medicine, 8, 587. doi:10.1038/s41746-025-01910-w — A current maturity check on human digital twins: most claimed healthcare twins fail the full dynamic, predictive, individualized standard. Source checked · July 2026
- Varela, F. J., Thompson, E., & Rosch, E. (1991). The Embodied Mind: Cognitive Science and Human Experience. MIT Press. Source — The embodied-enactive precedent for treating cognition as organized through brain, body, action, and environment rather than brain in isolation. Standard
- Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. Proceedings of the National Academy of Sciences, 112(4), 1036–1040. doi:10.1073/pnas.1418680112 — Evidence that computer models of digital traces can sometimes predict personality judgments more accurately than human observers. Standard
Verification key: “Source checked” means the cited work was located and its bibliographic identity reviewed; “Standard” marks a broadly established source; “Internal” marks a claim or formulation originating in this note. These references support the conceptual framing and its component precedents; they do not by themselves establish successful whole-agent reconstruction.
Comments
Post a Comment