PaperlayerView exact revision on arXiv

Source-linked AI summary

Epistemic Disturbance in the Graph Model for Conflict Resolution: State-Preserving Actions, Four-Valued Assessments, and the Distinction between Capability and Intention

Yukiko Kato

arXiv:2610.11690v1cs.GTcs.AIcs.MA

TL;DR

GMCR’s implicit treatment of unchanged states as inaction misses actions that alter beliefs without changing the physical conflict. The paper models these actions through epistemic updates and derives directional consequences for perceived moves and stability judgements.

  • Problem

    GMCR treats every unchanged-state event as doing nothing, although declarations and disclosures can change beliefs about available moves and what others would want to do.

    Source: §1 · p. 2

  • Method

    The paper augments states with epistemic states and models actions as generating evidence through observer-specific interpretation maps.

    Source: §4.2 · p. 7

  • Results

    Evidence for moves only enables and evidence against only disables perceived moves; capability affects all sanction-based concepts, while intention affects sequential stability alone.

    Source: §1 · p. 3

  • Takeaways & Limitations

    State-preserving actions can shift others’ stability judgements without changing preferences over physical states; their subsequent move choices remain outside this analysis.

    Source: §8 · p. 10

    The paper analyzes effects on stability judgements, not how DMs choose among moves or plan sequences of state-preserving actions.

    Source: §1 · p. 3

Abstract

from arXiv · show

In the graph model for conflict resolution (GMCR), a decision maker (DM) either moves the conflict to another state or does nothing. The basic definitions leave inaction implicit, so every action that leaves the state unchanged is treated as doing nothing. Yet announcements, exercises, leaks and selective disclosures are neither moves nor inaction: they leave the state unchanged but change what other DMs believe about which moves are available and which moves others would want to make. We introduce such state-preserving actions by augmenting states with the DMs' epistemic states: a physical move changes the physical state, a state-preserving action changes only the epistemic state, and inaction is the absence of a transition. Actions generate evidence through observer-specific interpretation maps. Building on a four-valued extension of GMCR from the author's earlier work, which separates evidence for and against, we show that evidence for a move can only enable perceived moves and evidence against can only disable them, that two of the four reduction operators ignore one kind of evidence, and that contradictory assessments are absorbing under monotone accumulation. With the monotonicity of stability in move sets, this fixes the direction in which any action moves a DM's stability judgements and characterizes when actions can enable provocation or deterrence. Capability assessments affect all sanction-based stability concepts, and on the DM's own side also Nash stability, whereas intention assessments affect only sequential stability. Hedging between two candidate types weakly expands or shrinks an observer's sequentially stable set according to how it reads contradiction. In the 1995 DVD format negotiation, general metarationality cannot distinguish its phases, since the computer industry group could always sanction; sequential stability, which asks whether it would, can.

1 Introduction

The paper distinguishes state-preserving actions from inaction by adding epistemic states to GMCR, then traces how evidence and assessments affect stability judgements. It positions this analysis against work on misperception, manipulation, and evidence representation.

1 Introduction

  • GMCR’s implicit treatment collapses everything that leaves the physical state unchanged into inaction; the paper instead models such actions as epistemic changes without physical moves.Inaction is the absence of a transition, while a state-preserving action changes only the epistemic state; physical moves change the physical state.Source: §1 · p. 2
  • Evidence direction, reduction rules, and move-set monotonicity determine how state-preserving actions alter stability judgements and when they can enable provocation or deterrence.Evidence for can only enable perceived moves and evidence against can only disable them; capability assessments affect sanction-based concepts and own-side Nash stability, while intention affects sequential stability. Hedging between candidate types weakly expands or shrinks sequentially stable sets according to how contradiction is resolved; the DVD case contrasts sanction capability with willingness.Source: §1 · p. 3

2.1 Moves and inaction in the graph model

  • Earlier GMCR work treats moves as arcs between distinct states and excludes the current state from reachable lists used in stability analysis.The paper notes that work on status quo evolution also traces sequences of moves, without representing what a decision maker does while staying.Source: §2.1 · p. 3

2.2 Misperception and unawareness

  • Prior frameworks study misperception or unawareness as given structures, rather than actions that change perceptions while the physical state stays fixed.The cited frameworks include hypergames, perceptual graph models, and interactive unawareness.Source: §2.2 · p. 4

2.3 Manipulation and deception

  • Related manipulation research adjusts preferences or studies deception through frameworks commonly using cardinal payoffs or probabilistic beliefs.The paper distinguishes its focus on capability and intention assessments from preference manipulation.Source: §2.3 · p. 4, §2.5 · p. 4

2.4 Representations of evidence

  • The four-valued GMCR extension represents option status but had previously been used for given assessments, not changes produced by actions.Belnap’s logic separates truth order from information order.Source: §1 · p. 2, §2.4 · p. 4

2.5 Positioning

  • The paper analyzes observer-specific action effects on capability and intention assessments, with their stability consequences determined ordinally by evidence and reduction rules.Its direction-of-effect results do not require cardinal payoffs or probabilities.Source: §2.5 · p. 4

3 Preliminaries

The preliminaries define graph-model stability through improvement moves and sanctions, then introduce four-valued evidence and reduction operators for interpreting assessments.

3.1 Graph models, sanctions and stability

  • Nash stability requires no available improvement, while GMR, SMR, and SEQ require each improvement to have the corresponding sanction.SEQ sanctions use a reachable opponent-improvement state that is no better for the focal DM than the status quo.Source: §3.1 · p. 5

3.2 Four values as evidence pairs

  • Four values encode evidence for and against as pairs: Y=(1,0), N=(0,1), U=(0,0), and YN=(1,1).The knowledge order adds evidence componentwise; the truth order orders the same pairs differently.Source: §3.2 · p. 5

3.3 Reduction operators

  • A reduction operator maps four-valued assessments to Y or N while preserving Y and N; four operators differ in their treatment of U and YN.δNN treats only uncontested evidence for as Y, whereas δYY treats anything other than uncontested evidence against as Y.Source: §3.3 · p. 5

4 State-preserving actions

The model augments physical states with epistemic profiles so it can distinguish physical moves, state-preserving actions, and inaction. Observer-specific interpretations update capability and intention assessments, and disturbances can add evidence to either or both.

4.1 Epistemic states and perceived models

  • Each DM’s epistemic state records evidence about option availability and whether other DMs’ moves would improve their controllers’ outcomes.The perceived model uses these assessments to determine perceived arcs and credible arcs, from which the DM judges stability.Source: §4.1 · p. 6

4.2 Three kinds of transition

  • A physical move changes the physical state, a state-preserving action changes only the epistemic profile, and inaction is no transition.Physical moves leave epistemic states unchanged in this model; observed physical moves are not modeled as causing epistemic updates.Source: §4.2 · p. 6
  • An action generates evidence through observer-specific interpretation maps, so different observers may interpret the same action differently or not observe it.The actor’s own epistemic state is unchanged; the model does not pursue the extension in which reactions to an action update the actor’s beliefs.Source: §4.2 · p. 7

4.3 Kinds of disturbance

  • A disturbance can raise evidence-for components, evidence-against components, or both, across capability and intention assessments and own-side or opponent-side options.These component updates commute in the final assessment, though intermediate perceived models may depend on their order; accumulation excludes retraction, correction, forgetting, and source discrediting.Source: §4.3 · p. 7, §4.2 · p. 7

5 How reduction operators respond to evidence

The section establishes how evidence changes reduced move assessments and identifies which assessments or operators are insensitive to particular evidence updates.

Direction of evidence

  • Raising evidence for can change a reduced assessment only from N to Y, while raising evidence against can change it only from Y to N.This direction holds for every reduction operator.Source: §5 · p. 7

Sensitivity

  • δNY depends only on evidence for, δYN only on evidence against, and δNN and δYY depend on both components.As evidence accumulates, δNY is knowledge-monotone, δYN antitone, and δNN and δYY are neither.Source: §5 · pp. 7–8

Immune assessments

  • U is never immune to further evidence, whereas YN is immune for every operator; Y and N are immune only when contradiction resolves in their direction.Under monotone accumulation, contradiction is absorbing, while definite assessments resist change only under the stated operator condition.Source: §5 · p. 8

6 Stability and move sets

The section relates changes in move sets to reachability and stability, distinguishing opponent move additions from changes to the focal DM’s own moves.

Reachability and stability

  • When opponents’ move sets expand, existing legal sequences remain available and the corresponding reachable-state sets can only grow.The stated inclusion also applies to reachability along designated arc sets when opponents’ designated sets expand.Source: §6 · p. 8

Stability in move sets

  • Expanded opponent move sets preserve existing sanctions, while Nash stability depends only on the focal DM’s own moves.The reason given is that the sets in which sanctions are sought can grow, so every existing sanction persists.Source: §6 · p. 8
  • Expanding the focal DM’s own moves makes sanction-based stability conditions harder to satisfy because improvements and escape sets can grow while sanction-search sets stay unchanged.SMR also requires the specified return set to remain within the relevant sanction set.Source: §6 · p. 8

7 The direction of disturbance

Capability disturbances change perceived move sets, whose effects on stability follow the monotonicity of stability in those sets. Their effects vary by which DM's capability is disturbed, the stability concept, and the reduction operator.

7 The direction of disturbance

  • Opponent-side capability for-disturbances weakly expand stability judgements, while against-disturbances weakly contract them; own-side disturbances reverse these directions.For Nash stability, opponent-side capability disturbances leave judgements unchanged.Source: §7 · p. 9
  • For arbitrary opponent-side capability disturbances, δNY weakly expands stable-state sets and δYN weakly shrinks them; own-side disturbances reverse the inclusions.For δNN and δYY, no general direction follows.Source: §7 · p. 9
  • Capability affects all sanction-based concepts; on the focal DM’s own side it also affects Nash stability, but on opponents’ side it does not affect Nash stability.Intention affects only sequential stability.Source: §7 · p. 9
  • A judgement is immune to every state-preserving action when relevant own-option capability and opponents’ move assessments are immune under the DM’s reduction operator.The relevant options and arcs are those whose tails lie in the specified reachable set.Source: §7 · p. 10

8 Actions that enable provocation or deterrence

State-preserving actions leave the physical state unchanged but can alter observers’ stability judgements. The paper characterizes when such actions enable provocation or deterrence through changes in perceived capabilities and intentions.

8 Actions that enable provocation or deterrence

  • Provocation changes a state from judged stable to unstable, where an improvement is welcome to the actor; deterrence changes it from unstable to stable, where the actor prefers the state to the contemplated improvements.These definitions concern the observer’s interpretation, regardless of true option availability or preferences; they do not assert that the observer takes an improvement.Source: §8 · p. 10
  • Provocation requires an effective opponent-side against-part or own-side capability for-part; deterrence requires an effective opponent-side for-part or own-side capability against-part.An effective part changes at least one perceived arc set or perceived credible arc set.Source: §8 · pp. 10–11
  • Under δNY, only evidence for own options can enable provocation, while evidence for opponents’ options or intentions can enable deterrence; δYN reverses these channels.Under δYN, provocation can be enabled by evidence against opponents’ options or intentions, and deterrence by evidence against own options.Source: §8 · p. 11
  • If an observer using δNY receives evidence only about others’ options and intentions, actions can enable deterrence but not provocation; for δYN, the reverse holds.The paper derives the action-level conditions by applying the direction results to effective disturbance parts.Source: §8 · p. 11
  • In the threat example, DM 1 reads DM 2’s exercise as evidence for the threat option and DM 3’s leak as evidence against it.The example uses a threat move from t to u that DM 1 assesses.Source: §8 · p. 11

9 Types and hedging

The paper derives intention assessments from an observer’s assessment of a DM’s candidate preference types. Hedging makes moves contested when the candidate types disagree, and the observer’s reading of contradiction determines how hedging changes sequential stability.

9 Types and hedging

  • An observer uses candidate types to assess intention: live types are those assessed as Y, and their preferences determine which moves count as improvements.The observer’s reduction operator is applied both to identify live types and to determine move credibility.Source: §9 · p. 12
  • When a DM hedges between two types, a move is assessed YN exactly when it is an improvement under one type but not the other; contested assessments are absorbing.Moves that are improvements under both types are Y, and those under neither are N.Source: §9 · p. 12
  • If contradiction is read as credibility, hedging weakly expands the observer’s sequentially stable set; if it is read as non-credibility, hedging weakly shrinks it.In the latter case, the hedged set is contained in the intersection of the sets under either revealed type.Source: §9 · p. 12
  • In the two-powers example, observers who read contradiction as credibility judge the status quo sequentially stable under hedging, though it is not so in the true model.In the true model, it is GMR-stable for both powers because GMR asks whether the third party can respond, not whether it would.Source: §9 · p. 13

10 Illustration: the 1995 DVD format negotiation

The DVD negotiation illustrates how perceived capability and intention can produce different stability judgements, while state-preserving communications alter what the camps believe about the TWG. The case is a reconstruction of the negotiation, not an account establishing what caused its outcome.

10.1 Background

  • The case traces the 1994 announcement of rival MMCD and SD formats through the September 1995 agreement and December announcement, amid conflicting accounts of summer signalling.The authors use the case to illustrate capability versus intention, not to establish the causes of the outcome; the framework accommodates either signalling account.Source: §10.1 · p. 13

10.2 Model

  • The model represents the MMCD camp, SD camp, and Technical Working Group, whose influence rested on members’ de facto market power.The TWG’s option B stylises refusal to support a format launched by one camp alone; the sources do not report a literal boycott.Source: §10.2 · p. 14
  • The model has 33 feasible states, from status quo state 1 to unified-format state 33, subject to constraints on acceptance, endorsement, and TWG statements.When both camps accept, their unified format is represented by one state with no TWG statement recorded.Source: §10.2 · p. 14
  • All transitions are irreversible: camps can launch or accept, and the TWG can make one statement while the format remains ununified.Figure 1 shows option-level transitions; Figure 2 integrates them into a graph with 88 arcs.Source: §10.2 · p. 14, Figure 1 · p. 14
  • Each camp prefers its endorsed sole format, while the analysis treats the TWG’s capability as certain and represents its intention with five candidate types.The analyst assigns the TWG the true preferences of θU; the candidate types include blockers, enforcers, and partisans.Source: §10.2 · p. 16

10.3 Capability and intention in the case

  • GMR judges the status quo stable for every assessment because the TWG can sanction; SEQ can distinguish negotiation phases because it asks whether the TWG would.Table 5 presents the status-quo judgements against unilateral launch.Source: Table 5 · p. 16, §10.3 · p. 16

10.4 State-preserving actions and their interpretations

  • May’s public call was read as evidence for a blocking TWG type, summer communications produced different readings for the camps, and September communication made an enforcer type live for S.After the summer reading of θS, S no longer judged the status quo SEQ-stable against its launch, while M continued to do so; the September communication restored S’s judgement.Source: §10.4 · p. 17

10.5 Limitations

  • The rankings are reconstructed, private summer communications rely on two accounts, Hollywood studios are omitted, and each camp and the TWG are treated as a single DM.The private communications are reported second-hand or retrospectively; the camps’ preferences are also reconstructed.Source: §10.5 · p. 17

11 Discussion

The discussion interprets reduction operators as distinct vulnerabilities to manipulation and identifies boundaries on the framework, including monotone evidence accumulation and omitted epistemic or coalition structures.

Interpretation

  • Reduction operators differ in how evidence shifts vulnerability: δNY can increase caution but not embolden, δYN can embolden but not increase caution, and the other two allow either direction.Under monotone accumulation, contradiction is absorbing, so contradictory evidence is effective only once and ambiguity from hedging persists.Source: §11 · p. 17

Limitations and further work

  • The analysis leaves open how conditions for preference manipulation relate to those for assessment manipulation, and assumes accumulated evidence cannot be retracted or corrected.Retraction, correction, or discrediting sources could resolve contradictory assessments and would require a different treatment.Source: §11 · p. 17
  • The framework excludes changes in assessments caused by physical moves, higher-order assessments, and coalitional stability; it also leaves action sequences and conflict anticipation to companion work.A further epistemic layer for actions perceived as physical-state changes would model feints in the narrow sense.Source: §11 · p. 18

A.1 States and rankings

The appendix presents the DVD model’s feasible states and reconstructed preference rankings, then specifies the principles and comparisons used to derive its stability results.

A.1 States and rankings

  • Table 6 lists feasible states in option form and each decision maker’s ranking, with rank 1 most preferred and tied numbers indicating equal preference.The five TWG columns represent candidate types, with θU identified as the true type.Source: Table 6 · p. 18, §A.1 · p. 18

A.2 Construction of the rankings

  • The rankings were reconstructed rather than elicited, using classes based on the camps’ positions and the TWG’s statement, then ordinal principles applied symmetrically to both camps.Table 7 presents the resulting camp ranks by situation from each camp’s viewpoint.Source: §A.2 · p. 19, Table 7 · p. 19
  • The camp rankings encode support, first-mover hope, unification, format war, and concession; some placements, including unification relative to an unendorsed launch, are modelling assumptions.The sources report pressure on a hold-out camp but not the camps’ valuations of concession outcomes.Source: §A.2 · pp. 19–20
  • The TWG’s five candidate types describe what a camp may believe about its preferences, not evidence of its actual preferences.The types distinguish blocking, enforcing, SD-fallback enforcer, and the two partisan profiles.Source: §A.2 · p. 20
  • The status-quo stability results depend on a limited set of camp and TWG comparisons; other ranking choices, including the exact position of unification and concession principle, do not affect them.The anticipatory analysis in uses further comparisons, notably the unification and concession principles.Source: §A.2 · p. 20
Loading 2610.11690v1…