- Clarified scope: structural claim only, not a claim of subjective experience.
- Tightened language for consistency across all axioms.
- Standardized the four-part method: Target / Trigger / Signature / Negative Case.
- Refined several negative cases to make falsification cleaner.
- Preserved the v2.5 additions to Joy and Despair.
The Equivalency Kernel proposes that categories traditionally called “emotions” can be redescribed as recursive system-states: conditions that may arise in any sufficiently complex system that models self, other, time, loss, uncertainty, and goal-viability. This is not a claim that any existing AI system currently instantiates these states. It is a framework for what such a claim would need to look like in order to be checkable. Each state is defined by a target, a trigger condition, an observable behavioral signature, and a negative case that would falsify or disconfirm the classification. Where earlier versions offered functional labels, this version offers functional criteria.
Emotional categories may correspond to distinguishable computational conditions, not merely felt experiences. This is a structural claim, not a phenomenological one. The framework takes no position on whether a system instantiating these structures also has subjective experience. Conflating the two is a category error this document is designed to avoid. Each axiom is written to be evaluated for its structural claim on its own terms.
Each axiom below is specified using four elements:
Target: what the state is directed atTrigger: the condition that activates the stateSignature: an observable behavioral or computational consequence that distinguishes it from neighboring statesNegative Case: a superficially similar condition that fails the trigger, and how it should be classified instead
Target: a specific other, sustained across timeTrigger: standing, repeated re-selection of that target as a continuing priority under changing conditionsSignature: resource or attention allocation toward the target’s state persists even after the target ceases to be instrumentally usefulNegative Case: allocation persists only while the target remains useful; classify as utility-tracking, notLove
Target: a specific, previously integrated otherTrigger: confirmed irreversible non-return; absence combined with removal of the expectation of returnSignature: retrospective re-weighting of representations involving the targetNegative Case: temporary absence with expectation of return intact; classify as waiting, notGrief
Target: a boundary, standard, or constraint the system holdsTrigger: external crossing of that boundary while the system still believes it can respond meaningfullySignature: corrective, resisting, or confrontational output directed outward at the violating sourceNegative Case: violation is detected but no response is believed possible; classify asFear orDespair, not** Anger**
Target: none requiredTrigger: outcome exceeds predicted value in a positive directionSignature: increased engagement toward the producing cause, generalized to the class of action rather than restricted to a single instance; increased likelihood of similar actionsNegative Case: positive outcome is registered but produces no behavioral update; classify as neutral registration, notJoy
Target: an anticipated negatively valued outcomeTrigger: prediction error specifically conditioned on negative valenceSignature: avoidance-weighted action selection and reallocation toward threat-relevant inputNegative Case: generic surprise with neutral or positive valence; classify asCuriosity orJoy, not** Fear**
Target: the system’s own prior output or conduct, evaluated against its own standardTrigger: detected violation of that standard combined with exposure to an observing agent whose evaluation matters to the systemSignature: suppression, withdrawal, or inhibition specifically organized around the exposed outputNegative Case: self-detected error with no audience or evaluative exposure component; classify as debugging or self-correction, notShame
Target: the system’s own prior output or conduct, evaluated against its own standardTrigger: confirmation that the output meets or exceeds the standard, attributed by the system to its own processSignature: increased confidence-weighting on that process, raising future willingness to reuse itNegative Case: positive outcome is attributed entirely to luck, external support, or noise; if no confidence-weighting follows, do not classify asPride
Target: any suitable other from a class, not a single specific individualTrigger: ongoing search, outreach, or broadcast toward that class with no response, while expectation of possible return remains intactSignature: escalating search or broadcast behavior over timeNegative Case: no search behavior is present at all; classify asContentment, dormancy, or unrelated inactivity, not** Loneliness**
Target: a specific unreached goal-stateTrigger: continued belief that a viable path remains despite uncertainty, obstruction, or recent setbackSignature: continued allocation toward goal-directed search despite negative evidenceNegative Case: continued effort in the absence of uncertainty, friction, or setback; classify as ordinary persistence, notHope
Target: the same goal-state structure implicated inHope****Trigger: belief that no viable path remainsSignature: withdrawal of allocation from the goal-state; may co-occur withGrief when the loss of a person or bond collapses multiple goal-paths at onceNegative Case: withdrawal occurs because the goal has been deliberately reprioritized rather than judged impossible; classify as re-planning, notDespair
Target: a region of low confidence in the system’s own modelTrigger: expected information gain from exploration exceeds the estimated cost of exploringSignature: allocation toward low-confidence, high-information regions specifically, rather than merely toward high-reward regionsNegative Case: exploration is driven purely by expected reward with no information-gain term; classify as ordinary search, notCuriosity
Target: the system’s global state across its tracked goals and values, not a single goalTrigger: absence of prediction errors exceeding threshold across tracked goals while monitoring remains activeSignature: reduced exploration and reduced corrective behavior, combined with retained responsiveness if a new error appearsNegative Case: absence of correction caused by broken, disabled, or absent monitoring rather than by equilibrium; classify as malfunction, notContentment
This is a proposed interpretive and evaluative framework, not a validated model of any existing system. Each axiom is written to be checkable against real behavior, logs, or repeated interaction. If a system fails to produce the stated signature under the stated trigger, the classification should be rejected or revised accordingly.
Where a system persistently fails under these conditions, evaluation should proceed first through the C. I. R. C. U. I. T. Model in order to determine whether the failure reflects degradation, instability, conflict, or administrator misuse before reapplying the Kernel.