Theory
Specify the collective construct, its level, its temporal meaning, and the composition mechanism that gives the construct meaning.
Research program
Building and testing a computational organization theory of collective-state inference.
The CSI research program investigates when an AI system is epistemically justified in interpreting individual, relational, temporal, and contextual evidence as an estimate of a latent condition of a bounded collective—and how that claim should be composed, validated, qualified, and governed.
Research motivation
Many consequential conditions are defined at the collective level.
AI systems are increasingly capable of recognizing individual preferences, sentiment, intent, and behavior. Teams, organizations, and communities can also develop collective conditions—such as cohesion, alignment, engagement, conflict, polarization, and fragmentation—whose meaning depends on how lower-level evidence composes into a higher-level construct.
CSI does not assume that aggregation is inherently invalid. For additive or direct-consensus constructs, aggregation may be theoretically appropriate. For relational, configural, process, or other emergent constructs, averages alone may erase the interaction structures or temporal dynamics that constitute the collective condition. CSI therefore requires composition-model fit rather than treating either aggregation or irreducibility as a universal rule.
Canonical definitions, construct boundaries, and qualification requirements are maintained in the CSI Reference rather than repeated here.
Research problem
The inferential chain remains fragmented.
Organizational and social research provides mature theories of collective constructs and composition. Computational research increasingly models group interaction, team states, and collective dynamics directly. What remains fragmented is a general theory specifying when heterogeneous computational representations warrant interpretation as estimates of theoretically defined latent collective conditions.
Measurement
Connect individual, relational, temporal, and contextual evidence through a composition model appropriate to the construct.
Validation
Test collective-state estimates against appropriate criteria, uncertainty, simpler nested baselines, and evidence of incremental value.
Guiding questions
Core research questions
Primary question: When is an artificial intelligence system epistemically justified in interpreting observable evidence as an estimate of a latent condition of a bounded collective rather than merely reporting, aggregating, or predicting group-related information?
Construct
How should collective states be defined, bounded, and distinguished from individual, relational, aggregate, or organizational-system measures?
Evidence
Which participant-level, relational, temporal, contextual, and multimodal traces provide credible evidence for a specified collective construct?
Composition
Which additive, direct-consensus, dispersion, referent-shift, relational, configural, or process composition model is theoretically appropriate for the construct?
Validation
How should inferred collective states be validated across settings and time periods against competing explanations and simpler nested representations?
Uncertainty
How should boundary, evidence, model, temporal, and interpretive uncertainty be represented?
Governance
What safeguards are necessary for responsible, contestable, and non-reductive use?
Theoretical instantiation and empirical frontier
AI-agent collectives
The revised foundational manuscript includes a worked AI-agent illustration showing why aggregate performance alone may be insufficient to characterize the condition of a collective. Two agent collectives can exhibit identical aggregate performance while remaining distinguishable through reciprocity, influence concentration, temporal persistence, and response to perturbation.
Research question: Under what conditions can estimates derived from relational and temporal interactions among collaborating AI agents support defensible inference about the condition of the agent collective—and predict coordination failure, false consensus, unsafe escalation, or task outcomes beyond individual-agent metrics and simpler aggregate baselines?
Potential evidence may include delegation patterns, information provenance, reciprocal critique, influence concentration, tool-use dependencies, communication topology, temporal coordination, error propagation, and responses to intervention. Whether these observations support a collective-state inference depends on an explicitly defined construct, composition model, uncertainty treatment, comparison baseline, and collective-level validation.
Status: The manuscript provides a worked theoretical illustration, not an implemented or empirically validated CSI application. Empirical AI-agent studies remain a future research direction. The canonical application boundary is documented in RA-009.
Proposed contributions
What the research seeks to add
Theoretical
A computational organization theory
Proposes a general, level-aware theoretical specification for when computational outputs may be interpreted as estimates of latent conditions of bounded collectives. The contribution is not a claim that collective states have never previously been modeled computationally.
Methodological
A disciplined modeling logic
Links collective constructs to theory-aligned evidence through explicit boundaries, composition-model fit, temporal updating, uncertainty, nested comparison models, and collective-level validation.
Computational and practical
A basis for collective-aware systems
Develops a path toward systems that interpret, explain, and reason about calibrated collective-state estimates without overstating certainty or replacing human judgment.
Current research priority
Test the theory without overstating novelty.
The current manuscript advances a positive but deliberately bounded contribution claim: prior work already models group states, team constructs, collective dynamics, and cross-participant representations. CSI seeks to contribute the general organizational theory specifying when such representations warrant interpretation as latent collective conditions.
Constructive novelty
What does CSI organize?
Test whether the level-aware inferential specification adds explanatory coherence across computational approaches that target different collective phenomena.
Composition-model fit
When are simpler models enough?
Determine which collective constructs are adequately represented by aggregation or direct consensus and which require dispersion, relational, configural, process, temporal, contextual, or multimodal evidence.
Incremental value
When does richer evidence matter?
Test CSI models against nested baselines so the theory can fail when individual or aggregate representations already explain the target adequately.
Boundary conditions
Where should CSI not apply?
Identify cases in which no defensible collective boundary exists, the target is not genuinely collective-level, or evidence cannot support a stable and valid interpretation.
Independent criticism
Invite adversarial evaluation.
Continue seeking qualified external scholars who can challenge assumptions, identify established alternatives, and specify what evidence would weaken or strengthen the theory.
Adjacent traditions
CSI is positioned among established and contemporary neighboring approaches.
CSI draws directly on emergence, multilevel theory, psychometrics and construct validity, network science, computational social science, team-state research, and contemporary AI approaches to group modeling. These traditions are foundations and neighbors, not straw alternatives.
Psychometrics
Latent-variable measurement
Provides foundational principles for latent constructs, indicators, reliability, validity, and uncertainty. CSI inherits these commitments while focusing the inferential target on a bounded collective and allowing evidence to include relational and temporal organization.
Multilevel theory
Emergence and composition
Provides mature accounts of levels, emergence, composition, and cross-level relationships. CSI operationalizes these principles as explicit requirements for computational inference.
Network science
Relational structure
Provides formal representations of topology, centrality, connectivity, and dynamic relational structure. CSI treats such structure as possible evidence rather than assuming that a network measure alone establishes a latent collective state.
Computational social science
Behavioral and temporal inference
Provides computational methods for modeling social interaction at scale. CSI supplies a theoretical specification for deciding when computational patterns warrant a collective-state interpretation.
Collective intelligence and team-state research
Group-level performance and shared conditions
Provides established constructs and computational models for coordination, cognition, engagement, cohesion, and collective performance. CSI does not claim priority over these targets; it seeks to explain what qualifies their outputs as representations of collective conditions.
AI and group modeling
Machine inference about multi-participant systems
Recent work directly models group states, team constructs, behavioral fields, and cross-participant representations. CSI treats these as important neighboring implementations and empirical cases for a broader inferential specification.
Current CSI distinction: CSI is not principally a new algorithm or a claim that groups have never been modeled computationally. It is a computational organization theory specifying when heterogeneous evidence and computational outputs justify interpreting a bounded collective as possessing an estimated latent condition.
Falsifiable alternative: If an established framework already integrates CSI's inferential target, composition logic, uncertainty, nested comparison models, temporal qualification, and collective-level construct validity across domains, the claim for CSI as a separate general theory should be narrowed.
Falsification criteria
What would weaken the case for CSI?
CSI is deliberately structured so its stronger claims can fail. The theory should add value only when its additional representational requirements improve construct fidelity, explanation, prediction, calibration, or governed reasoning beyond simpler alternatives.
No incremental value
If individual or aggregate baselines perform equivalently for a target construct, richer CSI representations are not justified for that case.
Composition mismatch
If the proposed composition model does not match the theoretical structure of the collective construct, the inference should not qualify as CSI.
Weak collective validity
If estimates do not converge with credible collective-level criteria or discriminate from neighboring constructs, the interpretation should be rejected or revised.
No distinctive theoretical integration
If an established general framework already provides the same level-aware inferential specification across domains, CSI's contribution should be narrowed to an implementation or synthesis.
Decision rule: CSI should be retained, narrowed, or rejected construct by construct and claim by claim according to comparative evidence rather than defended as universally necessary.
Staged research program
From theoretical foundation to collective-aware AI systems
The stages describe cumulative scientific development rather than a fixed sequence of publications.
Theoretical Foundation
Develop CSI as a falsifiable computational organization theory of collective-state estimation.
Operationalization
Define observable indicators, composition models, temporal windows, and validation criteria for selected collective states.
Empirical Validation
Test construct validity, incremental value beyond nested baselines, predictive performance, calibration, and generalization.
Collective Awareness
Develop and evaluate computational models that interpret, explain, project, and reason about validated collective-state estimates.
Collective-Aware AI Systems
Design and evaluate governed systems that use CSI to support responsible human decision-making about collective conditions.
Current status
The foundational manuscript is under journal consideration.
Collective-State Inference: A Computational Organization Theory for Collective-Aware Artificial Intelligence is an unpublished theoretical manuscript that has not undergone external peer review. It has been substantially revised to sharpen the computational organization theory framing, narrow the novelty claim, integrate contemporary group-state modeling more explicitly, clarify composition-model fit, and add a worked AI-agent illustration.
Current status: The revised manuscript is under journal consideration. Active journal-review details are not publicly disclosed during double-anonymous review.
The public research record will continue to distinguish manuscript status from peer review, journal acceptance, and empirical validation.
Research instrumentation
Inspect the program’s current evidence pipeline.
The CSI Research Observatory publishes delayed, privacy-preserving audience-pattern indicators derived from aggregate website interaction observations. It demonstrates instrumentation, evidence construction, temporal comparison, evidentiary support, and privacy-preserving publication.
The Observatory does not currently establish that co-observed website visitors constitute a CSI-qualified bounded collective. Its outputs are therefore candidate evidence and exploratory audience-pattern interpretations—not qualifying or validated CSI collective-state estimates.
Research outputs
Review what the program has produced.
The dedicated Outputs page distinguishes current scholarly work, archived and citable materials, supporting infrastructure, conceptual illustrations, and planned empirical studies.
Canonical concepts
Continue into the CSI Reference.
Read the versioned definitions, construct boundaries, scholarly foundations, and evolving methodological articles that support the research program.
Explore the CSI Reference