Skip to content
Contact
☰
Color scheme: Dark.
Back to research
RESEARCH SPACE D

Reconstruction, Perspective & World Models

How can hidden states, perspectives, relationships, constraints, causal structures and future developments be reconstructed from incomplete evidence?

What this space studies

What incomplete traces reveal about actors, relations and states. EIL reconstructs because hidden structure cannot be observed directly. Individual systems such as DINGIR support the space.

Central questions

  • Which perspectives, goals and assumptions can be reconstructed from traces?
  • How do a world model and an actor model differ?
  • Which causal structures are possible, and which counter-hypotheses stay open?

Research object

Hidden structure behind incomplete traces: states, perspectives, actors, relationships, constraints, causal chains and possible future developments. This includes missing intermediate steps and changes of state over time.

Prediction is critical

Reconstruction does not end in the past. Once state, actor structure, constraints, dynamics and causal relations are reconstructed, testable predictions about possible future states or behavioral consequences can be derived. The space must not be reduced to retrospective analysis or mere summarizing.

World model as a working model

A world model is kept as a working model. It holds relationships, dependencies, hypotheses and possible developments and is tested against new evidence. Whether a reconstructed connection is causal shows only when its predictions and counter-hypotheses are tested.

Research programs

2 programs of its own.

Core phenomena

  • Reconstruction
  • Perspective modeling
  • Actor models
  • World models
  • Causal models
  • Uncertainty
  • Prediction (central)
  • Further topics

Core phenomena

Reconstruction

  • Hidden State
  • State
  • Relationship
  • Constraint
  • Context
  • Temporal Reconstruction
  • missing intermediate steps
  • reconstruction of state changes
  • relational and temporal chains of emergence

Perspective modeling

  • actor perspectives
  • knowledge states
  • assumptions
  • goals
  • constraints
  • priorities
  • possible motive structures
  • competing perspective models
  • perspective dependence
  • intention and effect divergence

Actor models

  • agents
  • people
  • organizations
  • institutions
  • system components
  • dynamic actor models

World models

  • entities
  • relations
  • events
  • causal dependencies
  • state
  • history
  • possible futures
  • relationships and dependencies as a working model
  • hypotheses about possible developments

Causal models

  • causal chains
  • competing hypotheses
  • counterfactual reasoning
  • constraint propagation
  • causal hypotheses
  • counterfactual alternatives

Uncertainty

  • Observed
  • Inferred
  • Hypothesis
  • Unknown
  • Contradicted
  • confidence

Predictioncentral

  • testable predictions about possible future states
  • behavioral consequences
  • prediction from structure
  • scenarios

Further topics

  • Cognitive Reconstruction
  • Multi-Hypothesis Reconstruction
  • entity and relation modeling

Keywords (9)

  • hidden-state reconstruction
  • relational reconstruction
  • constraint discovery
  • causal reconstruction
  • state transitions
  • prediction under uncertainty
  • alternative hypothesis comparison
  • competing explanatory models
  • prediction of possible system developments

How this space is investigated

Observation and relation

Reconstruction works from traces.

Observation and relation

What is specifically observed

Hidden states are not accessible to direct observation. Every statement about a hidden state, relationship or causal chain is at least inferred and is kept together with competing hypotheses and their counter-evidence. What the traces do not yield stays unknown. A reconstructed model is never equated with the actual internal state.

Relation to the other spaces

This space turns observations from the other spaces into testable models. It asks what can be said about hidden states and possible developments and keeps observed, inferred, hypothesized, simulated, derived, contradicted and unknown apart.

System Reconstruction

Reconstructing hidden states and mechanisms from incomplete, distributed or longitudinal evidence.

Central question

What structure, state or relationship most completely explains the available traces when direct observation is not possible?

System Reconstruction

Scope
  • Hidden-state reconstruction
  • Temporal reconstruction
  • Relationship reconstruction
  • Constraint discovery
  • Causal chains
  • State transitions
  • Uncertainty and incomplete evidence
  • Perspective modeling
  • Separation of observation and interpretation
  • Contested hypotheses
Approach

Bring together fragmented, distributed or time-shifted traces, test competing reconstructions against each other, infer implicit constraints from breakpoints and failure states, and keep uncertainty explicit rather than collapsing it into a single explanation.

Reconstructed states are consistently separated by evidence status (observed, derived, inferred, hypothesized, contradicted or unknown), and the status of every statement stays visible, so a reconstructed model is never mistaken for the actual underlying state.

Observation and interpretation are kept in separate layers: what was actually said or done never sits in the same structure as a possible explanation for it.

A reconstructed actor or perspective model keeps competing hypotheses with their counterevidence explicitly in parallel, rather than collapsing early into one preferred reading. This follows the distinction between a world model, which represents the overall state of the system, and an actor model, which describes only one reconstructed perspective within it; the two levels are never silently merged.

Relationship

The core method behind DINGIR and the Research Knowledge Graph, and a precondition for defensible causality and prediction claims.

Causality, Prediction & DINGIR

Reconstructing causal structure and system state to reason about transitions and plausible future states.

Central question

Which causal chain and which counterfactual paths best explain a state transition, and when is that enough to call a future development plausible rather than merely possible?

Causality, Prediction & DINGIR

Scope
  • DINGIR
  • Hidden-state reconstruction
  • Causal chains
  • State transitions
  • Temporal reasoning
  • Counterfactual reasoning
  • Forecasting
  • Uncertainty
  • Forward, inverse and constraint reasoning
Approach

Derive causal chains from reconstructed states, explicitly test counterfactual alternatives against each other, separate temporal reasoning from mere correlation, and carry forecasts with an explicit uncertainty level rather than a single number.

Depending on the question, reasoning runs forward (what might happen next), inverse (what hidden state explains what was observed), through constraints (which states are impossible or inconsistent), or through temporal reconstruction (what sequence of states best explains the observations).

Relationship

DINGIR is the system represented for this space: it chains evidence, hidden state, causal structure, state transitions and possible futures together.

Perspective Modeling & Cognitive Reconstruction

Exploratory EIL investigates whether artificial intelligence can serve as a structured instrument for reconstructing how another person may perceive, interpret and reason about a situation.

To the chapter →

Full text in eleven sections

Central question

How far can a human-AI system reconstruct another person's likely internal model without confusing interpretation with observation?

  1. What is reconstructed from
  2. The central question
  3. What is reconstructed
  4. A possible analytical structure
  5. AI as a cognitive reconstruction instrument
  6. Perspective Modeling
  7. Multi-hypothesis reconstruction
  8. Theory of Mind and artificial systems
  9. From World Models to Models of Minds
  10. Epistemic boundary
  11. Long-term direction

What a reconstruction model must deliver and how it is checked

Planned The reconstruction needs especially strict controls, because it infers a plausible internal model from external evidence and does not observe thoughts directly.

Output and validation of the model

The check is planned and not yet carried out.

Output of the model (8)
  1. Known and observed facts.
  2. Likely beliefs with calibrated uncertainty.
  3. Candidate assumptions needed to explain the behavior.
  4. Possible goals and constraints.
  5. A model of how the person may interpret other actors.
  6. Contradictions within the reconstructed model.
  7. At least one plausible competing reconstruction.
  8. Unknowns that cannot be resolved from the available evidence, and evidence links for every nontrivial inference.
Validation (5)
  1. Blind reconstruction on held-out episodes.
  2. Alternative models are compared with later observable behavior without changing the earlier prediction afterwards.
  3. False positives and overconfident inferences are tracked.
  4. Descriptive prediction and psychological diagnosis are kept separate.
  5. Where evidence is weak, explicit uncertainty and abstention apply.

Methods, systems and axes

Methods (4)

All methods of this space

Methods and analysis procedures (4)

See all methods →

Systems (9)

All systems of this space

Research systems (2)

Multi-agent systems (1)

Shared technical foundation (1)

System state, memory, security & integrity (5)

See all systems →

Cross-cutting research axes (4)

How the axes act in this space

Cross-cutting condition Ethics / Human Agency / Human Rights →

Evidence & limits

Confidence and contradiction

Every reconstructed statement carries a confidence next to its status. Contradicted statements stay visible.

Plausible is not proven

A plausible reconstruction is not proven causality. A causal hypothesis remains inferred or hypothesized until competing explanations are excluded or its predictions are tested. Counterfactual alternatives show which assumption a conclusion depends on, but they do not establish causality by themselves.

Boundary of this space

Reconstructions are model-dependent approximations, not a complete restoration of hidden reality. Several models can be compatible with the same evidence.

How to read status and origin

Every statement carries its status (observed, inferred, hypothesis, planned or unknown), no program is externally validated, and origin statements about systems are candidates.

Evidence register → Origin of the systems →

Topics and publications

working formulation Formulated from the research program, not yet a fixed wording.

Topics with this space as main space (1)

Further topics that touch this space (1)

Guiding questions of the topics

  • 08 World Models

    How does a world model stay traceable while it changes with new observations?

  • 07 Knowledge, Evidence and Epistemic Integrity

    How does observed information stay distinguishable from inferred information when a system accumulates knowledge over a long time?

Publications (2)

Publications (2)