An interoperability pipeline
AI Attributes™ is the umbrella initiative for a family of open interoperability specifications — Open Emotions™, Open Sensations™, and Open Morals™ — that let heterogeneous classifiers, analytical frameworks, and sensors produce standardized, provenance-preserving observability signals.
Normalization standardizes representation, not scientific meaning. A score from one classifier is not automatically equivalent to the same score from another. Domain-specific meaning stays with the child specification that produced it.
Rather than assuming heterogeneous scientific constructs are identical, AI Attributes™ owns the shared conventions — artifacts, event envelopes, provider identity, adapter manifests, capability discovery, provenance, versioning, and registries — while domain semantics remain with the child initiatives.
Candidate functional relationships
Real systems may use feedback, parallel processing, or different arrangements. Every arrow is a relationship to investigate, not a statement that the sequence is necessary or sufficient.
Potential attribute layers
Input & Sensing
Represent inputs and ask how context shapes what a system detects and prioritizes.
Internal State
Explore persistent and changing internal variables, and their influence on behavior.
Self & Cognition
Study how systems represent their own capabilities, retain continuity, and revise models.
Agency
Compare how systems select actions and pursue bounded goals under constraints.
Embodiment
Investigate the relationship between a system, its internal signals, and its environment.
Toward an AI Attributes Knowledge Graph
The planned architecture would connect Open Sensations™, Open Emotions™, Open Morals™, and future domains through shared identifiers and explicit relationships. It is not presented here as a deployed or complete graph.
Each relationship should carry provenance, evidence status, source references, context, uncertainty, and version history. Competing interpretations should coexist with their qualifications intact.
An illustrative relationship
A sensor signal may be associated with several candidate interpretations depending on context. A useful graph would retain those conditions and the supporting sources rather than declare a single interpretation universal.
Read the evidence model →From Knowledge to Experiments
Future controlled sandbox research could compare a baseline system with a version that adds one defined attribute. Comparative agent experiments would vary persistent memory, emotional state, sensory input, self-model, motivation, or environmental constraints.
Human-speed and accelerated agents are possible comparison conditions. Experiments would need comparable tasks and measurements, bounded resources, stopping conditions, and human oversight. Changes in behavior would not, by themselves, establish subjective experience.
Explore the experimental method →