Predictive Intelligence for Human–AI Interaction

Understand the interaction.
See where it is heading next.

Enactive Dynamics models human–AI interaction as an evolving trajectory—not a sequence of isolated messages. We make interaction state, historical organization, emergent motifs, and future trajectory structure observable and actionable.

A system can score well turn by turn and still produce a deteriorating interaction.Enactive Dynamics detects that difference—surfacing trajectory-level failures that conventional turn-level evaluation can miss.
What can Enactive Dynamics do for your organization?
Evaluate conversational AI systems Detect interaction breakdown Compare agents Forecast trajectory change
For AI teams, research labs, product organizations, and high-stakes conversational systems.
HDMIC Core Interaction Dynamics Observatory Temporal Scope
The missing layer

AI analytics can measure systems and outputs. They rarely model the co-creative and interaction dynamics that emerge between a human and AI over time.

A conversation can look successful turn by turn while gradually losing continuity, failing to preserve constraints, repeating unresolved repair, or drifting into a different interactional regime.

Traditional AI observabilityLatency · tokens · cost · tool calls · errorsSystem
Output evaluationCorrectness · safety · relevance · task completionResponses
Enactive DynamicsHistorical organization · recurrence · coupling · drift · motifs · future trajectoryInteraction
Platform architecture

From conversation events to predictive interaction intelligence.

The platform combines a lightweight causal engine, an operational observatory, and a cross-corpus knowledge layer.

Layer 01 · Infrastructure

HDMIC Core

A streaming interaction-state engine that reads the conversation causally and returns trajectory signals that product systems can consume.

  • Interaction & historical organization
  • Recurrence, coupling, drift
  • Trajectory and motif inference
  • Prospective forecasts
→
Layer 02 · Operations

Interaction Dynamics Observatory

Live and batch analysis for product teams, researchers, and evaluation groups.

  • Trajectory visualization
  • Emergent motif discovery
  • Transition monitoring
  • Model / prompt comparison
  • Reports & alerts
→
Layer 03 · Organizational knowledge

Temporal Scope

Cross-corpus replication and motif intelligence that turns repeated observations into reusable organizational knowledge.

  • Cross-dataset motif comparison
  • Replication & canonicalization
  • Provenance and stability
  • Interaction Motif Library
↗
Product offerings

Turn interaction intelligence into concrete evaluation, monitoring, and adaptation services.

These offerings package Enactive Dynamics’ research systems into customer-facing engagements. Each can begin as a bounded pilot using existing interaction records and expand into recurring evaluation, monitoring, or adaptive-system integration.

↔
Model & Agent Evaluation

Compare not only final outcomes, but the trajectories through which different systems reach them.

⌁
Interaction Forecasting

Estimate when an interaction is likely to depart from its current organizational pattern.

◌
Interaction Pattern Monitoring

Recognize recurring interaction regimes and estimate likely transitions or exit destinations.

↺
Long-Horizon Regression Testing

Detect releases that improve local response quality while degrading extended coordination.

✦
Failure & Repair Diagnostics

Identify recurrent structures that precede breakdown, successful repair, or abandonment.

⌘
Adaptive Interaction Systems

Longer term, use validated trajectory forecasts as inputs to interaction policy and agent behavior.

How an engagement starts

Begin with a focused pilot, then scale the offering that proves useful.

Find failures that turn-level evaluation misses before they become deployment problems.
01 · BringInteraction records

Conversation logs, evaluation corpora, model-comparison runs, or repeated human–AI workflows with a decision your team already cares about.

02 · AnalyzeTrajectory structure

HDMIC estimates interaction state, historical organization, recurrence, coupling, drift, motifs, transitions, and prospective trajectory signals.

03 · ReceiveInterpretable diagnostics

A comparative report with trajectory visualizations, failure and repair patterns, model differences, forecast tests, and candidate monitoring signals.

04 · DecideAct on interaction quality

Use the results to compare systems, diagnose regressions, refine prompts or policies, define alerts, or identify the next validation experiment.

A pilot can begin with historical data; deployment integration is not required to test whether trajectory-level analysis adds useful information.

Discuss a Pilot →
Research systems

Working systems for interaction, regulation, and temporal intelligence.

These systems are the experimental and computational infrastructure behind Enactive Dynamics. They demonstrate how the underlying research program is operationalized across conversation analysis, co-creative regulation, temporal motif discovery, and continual adaptation.

Human–AI conversation dynamics Conversation Observatory

Prospective HDMIC analysis of human–LLM conversations, including causal playback, interaction-state estimation, motif discovery, batch experiments, and future-trajectory forecasting.

Open tool ↗
Co-creative regulation Creative Regulation Instrument

A multimodal laboratory for studying creative regulatory dynamics, clamp/unclamp behavior, coupling, drawing interaction, co-agent action, and multiscale HDMIC dynamics.

Open tool ↗
Temporal knowledge Temporal Scope

An open platform for discovering, interpreting, comparing, and sharing temporal motifs across time-indexed data, with annotation, motif libraries, and multiscale regime analysis.

Open tool ↗
Continual adaptation Emergence Machine

A lightweight adaptive system that continuously updates from streaming signals, tracks drift and plasticity, and compares adaptive forecast skill against a persistence baseline.

Open tool ↗
Explore the applications

From interaction science to working research environments.

Explore eight complementary research environments spanning conversation observability, embodied enactive agents, participatory animal communication, temporal motif discovery, multimodal interaction dynamics, sense-making, co-creation, and instrumented drawing. These are research prototypes and experimental systems, not claims of validated animal-language translation or commercially available hardware.

HDMIC Conversation Observatory interface showing experimental controls and interaction analysis
01 · Human–AI interaction · Public prototype

Human-AI Conversation Observatory

Analyze conversation trajectories, interaction dynamics, recurrence, breakdown and repair, and emerging Local / Regional / Global organization. Experiment with causal playback, prospective forecasting, batch evaluation, and reusable interaction landscapes.

Interaction dynamicsProspective evaluationMultiscale HDMIC
Open the Conversation Observatory ↗
Public build: WildChat retrieval is disabled; local JSON and available example datasets are supported.
Unified Enactive Agent Living World with two agents and a temporal influence observatory
02 · Embodied multi-agent research · Experimental

Enactive Agents · Living World

Watch situated agents coordinate inside a changing environment, negotiate joint construction, form predictive memories, and adapt to hazards and resources. An interaction observatory studies how dyadic organization develops over time.

Dyadic coordinationAdaptive regulationEmergent language
View Interactive Demo ↗
A new type of agent inspired by the Cognitive Science theory of Enaction. This environment may help study participatory sense-making.
Canine Semiosis website showing companion app, home sensing, and research observatory concepts
03 · Participatory human–animal communication · Research preview

Canine Semiosis

Explore a participatory canine communication laboratory: detect recurring vocal patterns, bring in human observations, refine tentative interpretations, and investigate how shared meaning develops through interaction.

Acoustic patternsOwner-assisted groundingLiving meaningscapes
Explore Canine Semiosis ↗
The interactive research preview is available; sensing hardware and other product concepts remain proposed.
Enactive Translation Animal Kingdom experimental bioacoustic interface showing speaker differentiation, acoustic events, multiscale hypotheses, semantic attractors, and prospective interaction forecasts
04 · Cross-species semiotic research · Experimental

Enactive Translator · Animal Kingdom Semiosis

Investigate animal vocal communication through acoustic motifs, temporal sequences, interactional events, and Local / Regional / Global attractors. Candidate interpretations are revised through contextual evidence and adequacy regulation, rather than treated as established translations.

Animal-native L/R/GRegulatory adequacyInteractive grounding
Explore the Enactive Translator ↗
Dedicated Enactive Translator deployment link.
Temporal Scope interface with temporal motif resolution, computation settings, and options for statistical and semantic analysis
05 · Collaborative temporal science · Public platform

Temporal Scope

Discover recurring temporal motifs in dynamic signals, interpret patterns across multiple scales, and collaboratively develop reusable knowledge through human annotation and semantic interpretation.

Temporal motif discoveryHuman annotationMultiscale interpretation
Explore Temporal Scope ↗
Open research platform for exploring and sharing temporal knowledge.
Interaction Dynamics Observatory dashboard with live computational coupling metrics, acoustic features, an intentional action channel, and a relational drawing canvas
06 · Multimodal human–AI interaction · Experimental instrument

Interaction Dynamics Observatory

Observe coupling, adaptation, and co-creativity as they unfold across computational, acoustic, intentional-action, and relational drawing channels. Explore how human–AI interaction develops through real-time multimodal evidence.

Multimodal couplingRelational drawingIntentional action
Open Interaction Dynamics Observatory ↗
Experimental research instrument; some channels depend on enabled inputs and device permissions.
Participatory Emergence Machine showing separate User and Agent interaction-dynamics streams and an independent observation clock with activity timelines
07 · Participatory sense-making · Experimental observatory

Participatory Emergence Machine

Observe sense-making and co-creation as an interaction unfolds. Separate online HDMIC engines trace User and Agent dynamics on a shared event clock, maintaining independent Local / Regional / Global attractors and adaptive histories for comparative analysis.

Sense-making dynamicsDual HDMIC streamsCo-creation observatory
Explore Participatory Emergence Machine ↗
Online experimental instrument. Composite interaction scores are exploratory heuristics, not independently validated consciousness measures.
Cognitive Trajectory Laboratory instrumented drawing canvas with a live metrics dashboard for engagement, flexibility, persistence, emotional regulation, and integration
08 · Enactive art research · Interactive drawing laboratory

Cognitive Trajectory Laboratory

Explore creativity through an instrumented drawing canvas that records stroke dynamics, local and structural events, and evolving cognitive trajectories. Live process indicators, replay, and export tools support investigating organization and regulation during drawing.

Instrumented drawingCreative trajectoriesProcess indicators
Open Cognitive Trajectory Laboratory ↗
Research environment inspired by enactive art therapy. Process indicators are exploratory and are not clinical diagnoses.

One research direction connects these systems: intelligence is studied not merely as isolated recognition or prediction, but as historically situated, adaptive organization emerging across agents, environments, and interactions.

Lightweight interaction intelligence

Intelligence without the heavyweight model.

Lightweight by design. Local by default. Learning through interaction.

HDMIC does not depend on a massive pretrained model to understand the evolving structure of an interaction. Its core architecture is designed to learn online from the interaction itself, continuously updating its representation of trajectory, recurrence, coupling, historical organization, and emergent motifs as the conversation unfolds.

Because the core system is lightweight, it can operate directly inside a modern web browser. In browser-local deployments, interaction data can remain on the user’s device rather than being sent to an external inference service. State estimation, trajectory modeling, motif learning, and prediction can occur locally as the interaction develops.

No massive inference model  →  No required cloud round-trip  →  Online learning  →  Browser-local analysis
⌂
Browser-Local

Analysis can run where the interaction occurs, keeping conversation data on-device in local deployments.

↻
Online Learning

HDMIC updates from the unfolding interaction rather than requiring an expensive retraining cycle.

◫
Lightweight Compute

The structural HDMIC engine is small enough to operate within contemporary browser environments.

→
Causal by Design

Predictions are generated from interaction history available at that moment—not from future turns.

The goal is not to reconstruct another foundation model. It is to provide a lightweight intelligence layer for understanding the dynamics that emerge between a human and an AI through time.
Evidence

Accumulated interaction history appears to carry prospective information.

The primary evidence comes from controlled longitudinal experiments designed to test whether accumulated interaction history can forecast how interaction develops through time. A separate naturalistic benchmark is retained as external research validation, showing that trajectory-level effects also appear in publicly available human–AI conversations outside the controlled synthetic corpora.

Instrument validity
Controlled positive control · 1,000 synthetic conversations

When coherent temporal structure is deliberately present, HDMIC detects and forecasts it strongly.

A controlled corpus of 1,000 coherently constructed human–AI conversations produced 13,500 prospective transitions. The corpus was designed around structured forms of progressive elaboration, repair and reframing, delayed return, and branching followed by synthesis. Because the data are synthetic and intentionally organized, this section is best interpreted as an instrument check / positive control: it asks whether HDMIC responds strongly when the kinds of temporal organization it is designed to measure are known to be present.

1,000synthetic conversations13,500 prospective transitions
87.5%Recurrence MAE reduction vs causal persistence at +10 turns
84.0%Historical Organization MAE reduction at +10 turns
72.9%Interactional Organization MAE reduction at +10 turns

Strict leave-one-conversation-out trajectory forecasting

Global +10-turn forecasts compared with causal persistence. Every row is clickable for the metric definition, baseline, error values, replication count, and interpretation.

Recurrence
87.5%
Historical Organization
84.0%
Interactional Organization
72.9%
Coupling
64.5%
Drift
32.7%
Conversation-level replication: Recurrence, Historical Organization, Interactional Organization, and Coupling improved in 1,000/1,000 held-out conversations. Drift improved in 868/1,000. The bootstrap 95% intervals for mean conversation-level gain were positive for all five dimensions.

Early warning & future-state tests

The controlled corpus also supports tests that deliberately resist trivial adjacent-window persistence. These ask whether HDMIC can anticipate reorganization, predict its destination, and identify a future motif reconstructed from turns that do not overlap the source window.

MOTIF CHANGE · +3
100%
balanced accuracy · 2,880 held-out events · +50.0 pp over 50% benchmark
MOTIF CHANGE · +5
100%
balanced accuracy · 2,380 held-out events · +50.0 pp over 50% benchmark
EXIT DESTINATION
100%
next different motif · majority destination 50.0% · N = 500 exits
NON-OVERLAP · 5 TURN
+79.4 pp
100% model accuracy vs 20.6% stay-current · N = 315
FUTURE BLOCK · t+1…t+5
100%
accuracy vs 63.8% stay-current · +36.2 pp · N = 1,380
FUTURE BLOCK · t+6…t+10
100%
accuracy vs 0.0% stay-current · +100 pp · N = 130
Click any card to see exactly what the measure means, how it is computed, and what can—and cannot—be concluded from a controlled synthetic corpus.
What else the run reveals · multiscale diagnostics

The same batch yields measures of state, historical influence, model architecture, learned episodes, and predictability.

These measures show the breadth of information available beyond forecast accuracy. They are especially useful for understanding what kind of interaction the corpus produced and which parts of the HDMIC representation carry predictive value.

DELAYED INFLUENCE
7,014
detected events in which earlier interaction structure becomes consequential again after a lag
FINAL HISTORICAL ORG.
68
mean final Historical Organization score across 1,000 conversations
FINAL INTERACTIONAL ORG.
63
mean final Interactional Organization score across 1,000 conversations
MEAN COUPLING
0.52
mean human–AI coupling across pooled interaction states
PRIMARY IMMEDIATE MODEL
49.9%
HDMIC-C strict held-out accuracy · +7.9 pp vs Joint History · +2.1 pp vs Dual-HDMIC
LEARNED EPISODE TYPES
3
multi-turn dynamical episode classes discovered after motif learning, without legacy labels
EPISODE VALIDATION COVERAGE
89.9%
held-out temporal windows within the discovery-set episode distance envelope
COHERENCE → MODEL ADVANTAGE
r = −0.634
Recurrence: higher early coherence is associated with smaller +10 gain over persistence—not lower absolute predictability
Current Strict Model CV hierarchy: HDMIC-C (49.9%) > Dual-HDMIC (~47.8%) > Joint History (42.0%). HDMIC-C is the primary immediate predictor; Dual-HDMIC is retained as an adaptive/history-learning ablation.

Interpretation boundary: these 1,000 conversations were deliberately constructed to contain coherent temporal organization. Near-perfect motif-transition results therefore demonstrate that the instrument can recover known structured dynamics; they should not be presented as estimates of real-world production accuracy. The naturalistic WildChat runs below remain the more appropriate evidence for ecological generalization.
Longitudinal prediction
Controlled longitudinal evaluation · 1,000 conversations · 30–60 turns

Long-Horizon Trajectory Prediction

In a controlled experiment across 1,000 longitudinal conversations containing 44,788 prospective transitions, HDMIC predicted future interaction organization substantially better than causal persistence baselines. Predictive gains remained strong—and for Recurrence increased—at horizons extending to 20 turns into the future. The corpus averages 45.8 turns per conversation and was designed to contain extended motif formation, return, repair, transformation, branching, reconsolidation, and other long-range interaction dynamics.

1,000longitudinal conversations30–60 turns · 44,788 transitions
77.9%Recurrence MAE reduction vs causal persistence at +10 turns
73.8%Interactional Organization MAE reduction at +10 turns
84.7%Recurrence MAE reduction at +20 turns

Longitudinal trajectory forecasting · +10 turns

Strict leave-one-conversation-out Global temporal forecasts compared with a causal persistence baseline. The held-out conversation contributes only its observed prefix as a query; sequence-memory examples come from other conversations.

Recurrence
77.9%
Interactional Organization
73.8%
Historical Organization
63.8%
Coupling
54.6%
Drift
50.5%
Conversation-level replication: the +10 Recurrence, Historical Organization, and Interactional Organization gains were positive in 1,000/1,000 conversations. The longitudinal result therefore reflects broad conversation-level replication within this controlled corpus, not a small number of unusually predictable runs.

The signal persists—and strengthens—at longer horizons

The strongest result is not limited to adjacent or short-range prediction. Recurrence forecasting improves as the horizon extends from +10 to +15 and +20 turns.

RECURRENCE · +15
80.1%
Global MAE reduction vs causal persistence
RECURRENCE · +20
84.7%
Global MAE reduction vs causal persistence
MOTIF CHANGE · +3
99.6%
balanced accuracy · +49.6 pp above the 50% benchmark
NON-OVERLAP MOTIF
100%
accuracy · +12.7 pp vs stay-current baseline

Longitudinal structure recovered

REGIONAL MOTIFS
3
learned motif types · 90.0% held-out validation coverage
EPISODE STRUCTURE
4
learned episode types · 90.1% held-out coverage
DELAYED INFLUENCE
29,694
detected events · 29.7 per conversation
IMMEDIATE PREDICTION
32.1%
HDMIC-C prequential accuracy · +1.7 pp vs Joint History
Controlled longitudinal evaluation. This corpus is structurally varied but synthetic and deliberately contains long-range interaction dynamics. It tests whether HDMIC can recover prospective temporal structure under controlled conditions; it is not presented as external validation on naturalistic human–AI conversations. The strongest current interpretation is therefore methodological: accumulated interaction history contains recoverable information about the future trajectory of interaction, including structure that remains detectable 15–20 turns later.
Enactive Dynamics thesis. The organization produced through prior interaction constrains what the interaction becomes next. The +20 result is especially informative because it is difficult to reduce to ordinary adjacent-turn persistence: the strongest Recurrence forecast reaches an 84.7% reduction in mean absolute error twenty turns into the future.
Naturalistic replication
External naturalistic benchmark · 924 conversations

Naturalistic research validation

HDMIC has also been evaluated on publicly available naturalistic human–AI conversations, where trajectory forecasting effects replicated outside the controlled synthetic corpora.

924conversations13,686 transitions
Research context. This benchmark uses publicly available naturalistic human–AI conversations and is included as ecological corroboration that trajectory-level effects also appear outside the controlled synthetic corpora. It is secondary to the site's primary commercial evidence, which comes from the controlled longitudinal evaluation.
72.8%Recurrence MAE reduction vs causal persistence at +10 turns
62.2%Historical Organization MAE reduction at +10 turns
53.5%Interactional Organization MAE reduction at +10 turns
Replication. Recurrence improved in 678/680 eligible conversations, Historical Organization in 679/680, and Interactional Organization in 672/680. These results support naturalistic generalization, but they are presented here as research validation rather than the core commercial benchmark.
Scope. The strongest current commercial-facing evidence remains the controlled longitudinal 1,000-conversation evaluation, which directly tests long-horizon trajectory prediction, motif structure, delayed influence, and +15/+20 forecasting. The naturalistic benchmark is retained to show that the signal is not confined to synthetic data.
Core architecture

HDMIC — Hellinger Drift Modulation Interactional Coherence.

HDMIC is the causal interaction-intelligence engine behind Enactive Dynamics. It models human–AI interaction as a developing trajectory rather than a sequence of isolated messages, estimates the current interaction state from history available at that moment, discovers recurring organizational structure, and forecasts how the interaction is likely to develop next.

Measurement status. HDMIC variables are computational interaction constructs derived from observable temporal behavior. They are designed as testable estimates of interaction organization—not direct measurements of latent psychological states, consciousness, or creativity.
High-level HDMIC architecture showing the interaction stream, streaming features, state estimation, multiscale dynamics, emergent structure learning, forecasting, the Interaction Dynamics Observatory, Temporal Scope, and future adaptive systems.
Click any highlighted architecture region to explore how it functions inside HDMIC.
High-level HDMIC system architecture: observe → model → discover → predict → adapt.
01

Interaction Stream

HDMIC begins with the unfolding human–AI exchange: ordered turns, responses, pauses, recurrence, and changes in the interaction through time.

02

Streaming Features

The system derives lightweight causal features such as turn-taking, response length, lexical recurrence, delayed recurrence, historical similarity, change magnitude, and path dependence.

03

HDMIC State Engine

From history available at time t, HDMIC estimates interactional organization, historical organization, coupling, recurrence, drift, and related trajectory variables. Future turns are not used to construct the current state.

04

Multiscale Dynamics

Interaction is modeled across local, regional, and global temporal scales so the system can distinguish immediate changes from larger trajectory structure.

05

Emergent Structure Learning

Local attractors can compose into regional motifs and broader episode or trajectory patterns, allowing recurring forms of interaction to emerge from the data rather than being imposed as a fixed taxonomy.

06

Forecasting & Interaction Intelligence

The resulting trajectory representation supports future-state forecasting, motif-change prediction, transition and exit forecasting, and early warning for interactional reorganization.

The architecture connects measurement to prediction. The Interaction Dynamics Observatory operationalizes HDMIC for live and batch analysis, while Temporal Scope compares and validates recurring structures across corpora. Longer term, validated state and trajectory forecasts can become inputs to adaptive interaction policy.

Interaction stream → causal state → multiscale trajectory → emergent motifs → future-state forecast
Theory · linked enactive models

Co-creation unfolds through a historically structured field in which emergent affordances and constraints shape what becomes possible next.

The Enactive Model of Creativity and the Enactive Model of Consciousness describe the same regulatory movement from two perspectives. The creativity model emphasizes how awareness shifts between internal modeling and inspection of the world. The consciousness model emphasizes how that shift changes the breadth of feedback and feedforward loops connecting experiential history, mind, body, and environment.

Model 01

Enactive Model of Creativity

Creative regulation moves awareness across a continuum between thinking and inspecting. At center, cognition is clamped into stable, coherent engagement. Moving away from center opens the system to a wider field of alternatives.

Clamped
Enactive Model of Creativity showing awareness, intention, perceptual logic, perception, mental model, environment, and a clamped-to-unclamped continuum.
Unclamped Clamped Unclamped
Internal modeling
Clamped
World inspection
What changes as awareness moves?

Leftward movement privileges internal modeling, memory, conceptualization, and reflective integration. Rightward movement privileges sensory input, affordances, observation, and action. Either direction away from center represents unclamping: a widening of the cognitive search space rather than simple loss of organization.

Model 02

Enactive Model of Consciousness

Conscious experience is situated between experiential history and the environment. Feedback and feedforward loops continually connect remembered interaction, projected possibility, embodied action, and environmental exploration.

Balanced loops
Updated Enactive Model of Consciousness diagram showing experiential history, brain, mind, body, environment, feedback and feedforward loops.
Why the loops widen and narrow

When awareness moves toward reflective thinking, the upper experiential-history loops enlarge: cognition ranges more broadly across memory, integration, and projected possibilities. When awareness moves toward inspection, the lower body–environment loops enlarge: cognition ranges more broadly through action, exploration, and feasible environmental possibilities.

One regulatory movement · two representations

Centered awareness produces a tighter, clamped coordination.

At center, the two models converge on a focused state: intention, perceptual logic, mind, body, and environment are organized around a comparatively narrow set of relevant possibilities.

History-loop breadth 1.00×
Environment-loop breadth 1.00×
From theoretical construct to interaction instrument

The sense-making curve turns clamping and unclamping into a temporal interaction signal.

The two enactive models become operationally useful when treated as a time-varying regulatory process. Clamping describes moments when cognition stabilizes around a narrower field of relevance, while unclamping opens the field toward remembered experience, projected possibilities, alternative interpretations, environmental affordances, and exploratory action. Commercially, these dynamics can be represented as a sense-making curve: a longitudinal trace of how users regulate interaction with an AI, allowing Enactive Dynamics to characterize sense-making across the evolving interaction rather than evaluating isolated turns.

Live clamping stream

Move the Theory slider above. Each movement emits a streaming interaction code and updates both plots below.

Current code−1.00
Cumulative sum−1.00
Samples1
Current modeClamped
Raw interaction code through timecenter = −1 · either extreme = +1
Creative sense-making curverunning cumulative sum of raw interaction codes
Operational definition used in this demonstration. Slider position is converted continuously to an interaction code from −1 to +1. The center of the continuum is maximally clamped and receives −1. Either extreme is maximally unclamped and receives +1. Intermediate positions interpolate continuously between those values. The creative sense-making curve is the running cumulative sum of that raw stream, so it preserves the historical residue of prior clamping and unclamping behavior rather than treating each moment independently.
tsliderrawΣ raw
Conceptual status. These interactive diagrams visualize the theoretical relationship between creativity, consciousness, clamping, and feedback-loop breadth. The animation is an explanatory model, not a direct physiological measurement. Enactive Dynamics tests downstream interactional consequences of history, recurrence, coupling, drift, and trajectory formation.
Research lineage

Research Systems Behind Enactive Dynamics.

Enactive Dynamics grows out of a broader program of working systems for studying interaction, adaptation, co-creation, and temporal organization.

Research lineage: interaction-centered prototypes → trajectory instrumentation → online regulation and adaptation → HDMIC → Interaction Dynamics Observatory + Temporal Scope.
Foundational publications

A research lineage from enactive creativity to interaction-centered intelligence.

This research program has developed through an iterative process of building prototypes, observing interaction, and using those observations to construct new theories and models. Early creativity-support tools examined how computation could help individuals engage in creative work. Co-creative systems then shifted the focus toward creativity as an emergent process between human and computational partners.

Each prototype functioned as a technical probe, making interaction dynamics visible and generating empirical questions that existing theories could not fully explain. Across successive systems, the research moved from artistic computer colleagues and participatory sense-making to Creative Sense-Making, quantified co-creation, hybrid intelligence, and explainable co-creative AI. This trajectory ultimately led to Interaction-Centered Intelligence and Cognitive Trajectory Modeling, which understand creativity and intelligence as relational, temporal processes that emerge and reorganize through interaction.

Select a publication to view its abstract and online paper.

Enactive creativityCo-creative collaborationParticipatory sense-makingCreative Sense-MakingQuantified interactionFive pillars of enactionExplainable co-creative AIInteraction-Centered IntelligenceCognitive Trajectory ModelingEnactive Drift Regulation
Research & product ecosystem

Built from a broader program in interaction, temporal organization, and adaptive intelligence.

Enactive Dynamics sits at the commercial intersection of a larger research ecosystem: Interaction Science provides the field-level theory, Enactive AI develops interaction-centered systems, the Emergence Machine explores regulation under continual change, and Temporal Scope builds cross-corpus temporal knowledge.

Work with Enactive Dynamics

Bring us a repeated human–AI workflow and a question that matters.

A strong pilot begins with real interaction records and a concrete decision: compare two agents, diagnose a recurring breakdown, evaluate a new prompt, identify motif transitions, or test whether trajectory signals anticipate outcomes your team already cares about.