Skip to content
Contact
☰
Color scheme: Dark.
Back to Systems

Observatory

Research system / Measurement environment

The Observatory makes long-term changes in human–AI interaction, emergent patterns and technical system states open to study.

It is the measurement and reconstruction layer of EIL: it makes longitudinal human–AI interaction readable across six measurement dimensions and records what was observed and what was inferred.

In developmentPrototypeThis page describes the measurement model and shows no live data.

In development. Individual views run on real trajectory data; the full range of dimensions is not yet implemented. The Observatory gives no direct access to mental states.

Six measurement dimensions

The same interaction, seen from six directions.

  • Human

    Prototype

    Observable behavior of a person across many sessions: what they do, how they repeat it and how they change it.

    First behavioral views run in the admin panel on real interaction data. Cognitive proxies are behavioral indicators.

  • AI

    Prototype

    Observable behavior of the AI system in the same interaction, with a focus on change and continuity.

    Response and information views exist as a prototype. Semantic drift and contradiction detection are not yet validated.

  • Interaction

    Prototype

    The relationship between both sides as an object of study in its own right: who initiates, who follows, who corrects whom.

    A view on interaction and information dynamics is running. Role dynamics and dependence are still open as measures.

  • Emergence

    Exploratory

    Candidate patterns and the test of whether they recur, hold across contexts and stabilize.

    Patterns count as candidates here until recurrence across independent contexts is evidenced. After every completed exchange in research conversations, an LLM-based signal detector runs automatically. It delivers an interpretation by a language model, checks recurrence across separate conversations by embedding similarity and draws on CCET values. This detection is not validated.

  • System

    Prototype

    The state of the technical environment in which the interaction takes place, so that measurement artifacts stay recognizable.

    System state, agent activity, kernel ledger and time-scrubbable state snapshots run in the admin panel. The provenance chain is not yet continuous.

  • Neuroadaptive / Human State

    Planned

    Estimation of human state. Today exclusively from interaction behavior; physiological signals are planned.

    Planned. No physiological data is collected. Every state estimate is inferred and is never presented as an observed state.

Observables and channels

Human

Observables

  • behavior
  • interaction style
  • strategy shifts
  • persistence
  • correction behavior
  • exploration
  • repetition
  • decision dynamics
  • temporal patterns
  • cognitive proxies

AI

Observables

  • response behavior
  • semantic drift
  • contradiction
  • adaptation
  • state continuity
  • uncertainty
  • recovery

Interaction

Observables

  • reciprocity
  • coordination
  • convergence
  • divergence
  • repair
  • initiative
  • dependence
  • role dynamics
  • feedback loops

Emergence

Observables

  • candidate patterns
  • recurrence
  • cross-context recurrence
  • stabilization
  • new strategies
  • new roles
  • externalization

System

Observables

  • state
  • memory
  • runtime
  • handoffs
  • failures
  • retries
  • repair
  • agent activity
  • provenance

Neuroadaptive / Human State

Observables

  • behavioral proxies
  • temporal patterns
  • interaction-derived state estimation
  • physiological signals (planned)
  • wearables (planned)
  • EEG / BCI (planned)

Channels in use todaybehavioral proxies from interaction data, temporal patterns across sessions, interaction-derived state estimation

Planned channels, not implementedphysiological signals (planned), wearables (planned), EEG / BCI (planned)

15 views, each with its question

The status shows how far a view has come. Concept means the view does not exist yet. Prototype means a version runs in the admin panel without the research claim being confirmed.

ConceptApplies to every view without a label of its own.

  1. Temporal trajectories

    Prototype

    How do behavioral measures of a person develop over weeks and years?

    • Human
    • Interaction

  2. Recurrence matrices

    Does a pattern recur in independent time windows and contexts, or does it belong to one context only?

    • Emergence
    • Human

  3. State transitions

    Which transitions between interaction states occur, and which are more frequent than expected?

    • Interaction
    • Neuroadaptive / Human State

  4. Interaction graphs

    Who initiates, who follows and who repairs, and how does that shift over time?

    • Interaction
    • Human
    • AI

  5. Emergence timelines

    In what order does a signal become a pattern and a pattern become a stable structure?

    • Emergence

  6. Behavioral clusters

    Prototype

    Which behavioral regimes can be told apart, and does the person switch between them?

    • Human
    • Neuroadaptive / Human State

  7. Semantic drift maps

    Does the meaning of central terms shift on one side or on both?

    • AI
    • Human
    • Interaction

  8. Role transition maps

    When and how do roles switch between human and AI, and when does a role stabilize?

    • Interaction
    • Emergence

  9. Human–AI co-change plots

    Do human and artificial behavior change together, one after the other or independently of each other?

    • Human
    • AI
    • Interaction

  10. Evidence chains

    Which evidence does a claim rest on, and how much of it is observed rather than inferred?

    • System
    • Emergence

  11. Reconstruction confidence views

    How reliable is a reconstructed state, and where does the observed end and the inferred begin?

    • Neuroadaptive / Human State
    • System

  12. Causal hypothesis graphs

    Which causal explanations compete with each other, and which counter-evidence stands against each?

    • Emergence
    • Interaction

  13. System event timelines

    Prototype

    What did the system do at which point in time, and does a system event explain an observed behavioral change?

    • System

  14. Agent handoff traces

    Prototype

    How do tasks and responsibility move between agents, and where do failures, retries and repair arise?

    • System

  15. Externalization lineage views

    From which recurring thinking operation did which method, role or agent structure derive, and how has it changed since?

    • Emergence
    • System

Research questions of the views

Temporal trajectories

Research questions (2)
  • Does a human–AI interaction improve, preserve or deteriorate over months and years?
  • How do roles, agency and trust between human and AI change across sustained interaction, and when does adaptation tip into dependence?

Recurrence matrices

Research questions (2)
  • Does a pattern recur across independent contexts, or does it belong to one context only?
  • When does a single signal become a pattern, and when does a pattern become a stable structure with a functional effect?

State transitions

Research questions (2)
  • Which causal chain and which counterfactual paths best explain a state transition?
  • How can inferred human or system state remain distinguishable from direct observation?

Interaction graphs

Research questions (2)
  • How do roles, agency and trust between human and AI change across sustained interaction, and when does adaptation tip into dependence?
  • How does repeated interaction change both the behavior and the structure through which later interaction takes place?

Emergence timelines

Research questions (2)
  • When does a single signal become a pattern, and when does a pattern become a stable structure with a functional effect?
  • When does a recurring interaction pattern become a stable structure?

Behavioral clusters

Research questions (2)
  • What recurring thinking and working habits can be reconstructed from a multi-year interaction history without confusing observation with direct access to mental states?
  • How do roles, agency and trust between human and AI change across sustained interaction, and when does adaptation tip into dependence?

Semantic drift maps

Research questions (2)
  • How does repeated interaction alter both human and artificial behavior?
  • How does repeated interaction change both the behavior and the structure through which later interaction takes place?

Role transition maps

Research questions (2)
  • How do roles, agency and trust between human and AI change across sustained interaction, and when does adaptation tip into dependence?
  • When does a recurring interaction pattern become a stable structure?

Human–AI co-change plots

Research questions (2)
  • How does repeated interaction alter both human and artificial behavior?
  • Does a human–AI interaction improve, preserve or deteriorate over months and years?

Evidence chains

Research questions (2)
  • How can inferred human or system state remain distinguishable from direct observation?
  • When does a recurring habit become a method, and what has to be evidenced for that?

Reconstruction confidence views

Research questions (2)
  • How can inferred human or system state remain distinguishable from direct observation?
  • What structure, state or relationship most completely explains the available traces when direct observation is not possible?

Causal hypothesis graphs

Research questions (2)
  • Which causal chain and which counterfactual paths best explain a state transition?
  • How does a reconstructed actor model change over time, and how do competing hypotheses stay side by side with their counter-evidence?

System event timelines

Research questions (2)
  • How can longitudinal changes be measured without confusing tool drift with real behavioral change?
  • How can a measured trajectory be checked as genuine and not the product of instrument drift or a false state?

Agent handoff traces

Research questions (2)
  • How does distributed intelligence alter responsibility, authority and failure?
  • Which collective phenomena arise unplanned, such as spontaneous specialization or hidden centralization?

Externalization lineage views

Research questions (3)
  • What cognitive functions externalize?
  • How do they change after externalization?
  • Which cognitive operations become externalized?

7 modules of the Observatory

The modules describe what the Observatory is meant to make accessible. Prototype means a version runs internally. Planned means there is no dedicated module yet.

  • Experiments

    Prototype

    Which experiments are currently running, which are finished, and with what result.

  • State & Runtime

    Prototype

    How the state of selected research systems changes over time, as a trajectory.

  • Reconstructions

    Planned

    Making reconstructed states, relationships and transitions visible, each carrying its evidence status (observed, derived, inferred or hypothesis), so that a smooth text report cannot hide that distinction.

  • Research Streams

    Planned

    Which research threads actually connect across multiple investigations.

  • Data

    Planned

    Where a single finding comes from: which source, which point in time and which dataset stand behind it, traceable to its origin.

  • Scientific Visualization

    Prototype

    Representing trajectories, system dynamics and causal structure so a pattern becomes recognizable.

  • Emergence Registry

    Planned

    An observed pattern only counts as verified emergence once it recurs with high content similarity across several independent conversations.

Modules in detail

  • Experiments

    Prototype

    Which experiments are currently running, which are finished, and with what result. Already running internally in the Simulation Center against real data, not yet released publicly.

  • State & Runtime

    Prototype

    How the state of selected research systems changes over time, as a trajectory. Already captured internally through System State and Flugschreiber, not yet released publicly.

  • Reconstructions

    Planned

    Making reconstructed states, relationships and transitions visible, each carrying its evidence status (observed, derived, inferred or hypothesis), so that a smooth text report cannot hide that distinction. Not yet implemented as its own module.

  • Research Streams

    Planned

    Which research threads actually connect across multiple investigations. Not yet implemented as its own module.

  • Data

    Planned

    Where a single finding comes from: which source, which point in time and which dataset stand behind it, traceable to its origin. Not yet implemented as its own public module.

  • Scientific Visualization

    Prototype

    Representing trajectories, system dynamics and causal structure so a pattern becomes recognizable. Already available internally through several modules: Behavioral Landscape, Knowledge & System Map, Interaction+Information Dynamics. Not yet released publicly.

  • Emergence Registry

    Planned

    An observed pattern only counts as verified emergence once it recurs with high content similarity across several independent conversations. A single occurrence is not enough. The pattern is classified into one of four levels: human, AI, interaction or system. Not yet implemented as its own module.

Neuroadaptive direction

Neural and physiological data are not collected, the neuroadaptive dimension is so far a plan.

The roadmap with its stages, closed loop and adaptation levels sits on the research page Human & Cognitive Dynamics (Human Cognition & Behavior).

To the roadmap on the Research page

15 data layers of a longitudinal corpus

The layers run from the single conversation to the externalized structure.

PrivacyPrivate raw data is not published and not shown on the website. At most aggregated results are public, from which no single conversation can be reconstructed.

Data foundation

Datasets

  • Longitudinal interaction corpus

    Active research

    Multi-year human–AI conversation history, structured into fifteen data layers from conversation to externalized structure. Basis for recurrence, pattern lineage and trajectory analysis.

    Public in aggregate form only

  • Pattern lineage and externalization events

    Exploratory

    Derived layer above the corpus: candidate patterns, their recurrence across contexts and the events in which a pattern became a method, role or structure.

    Public in aggregate form only

Data layers

  1. Conversation
  2. Turn
  3. Session
  4. Timestamp
  5. Topic
  6. Interaction context
  7. Behavior
  8. Correction
  9. Conflict
  10. Strategy
  11. State
  12. Semantic change
  13. Outcome
  14. Emergent pattern
  15. Externalized structure

What longitudinal analysis can support

  • recurrence detection
  • pattern lineage
  • behavior trajectories
  • interaction regimes
  • semantic transitions
  • strategy evolution
  • correction loops
  • externalization events
  • longitudinal change

What the Observatory is for

Research purposeMakes longitudinal interaction data readable across six measurement dimensions, so that every visualization answers a concrete research question.

Public limitsShows only what was actually logged. Raw data stays private; only aggregate views are public.

Research context of the Observatory

Why it existsWithout a shared observation layer, longitudinal research stays tied to single sessions and one-off analyses, and a pattern cannot be told apart from a chance swing.

PlacementObservation environment for human, AI, interaction, emergence, system state and neuroadaptive signals.

Every visualization is bound to at least one research question.

Current and future neuroadaptive channels are kept apart.

OriginThe measurement dimensions and views trace back to questions that arose in longitudinal human-AI interaction. Origin of the systems →

Supported research programs

Research questions (1)
  • Which signal types can meaningfully be linked to interaction trajectories, and how would they be calibrated against established behavioral measures?