AICO Ontology & Schemas

1. Purpose and Scope

This document defines the core ontology and schemas used by AICO’s agency subsystem and related services. It serves as the single source of truth for:

• Shared concepts and types used across conversation, memory, world model, planning, curiosity, values/safety, and embodiment.
• Event and data schemas (especially PerceptualEvent) that specify how raw inputs are interpreted into semantic structures.
• Relations between entities (user, AICO, projects, tasks, hobbies, memories, emotions, environment, sensors, goals, plans, tools).
• Usage boundaries: which subsystem owns which part of the ontology and how they are allowed to extend it.

This is a conceptual and implementation-guiding ontology:

• It is designed to back a property/knowledge graph plus JSON schemas, not a heavyweight OWL logic stack.
• It is intentionally minimal but stable enough to support early implementations of:
• Perception / Sensors → PerceptualEvent
• World Model Service / KG+Schemas
• Goals & Intentions subsystem
• Curiosity Engine
• Values & Ethics / Safety

2. Design Principles

• 2.1 Pragmatic, not maximalist
  Start with a small, high-value core that is stable, then extend.

• 2.2 Shared language across components
  All subsystems must use the same terminology and identifiers for core entities and events.

• 2.3 Hybrid representation
  Every important object has:

• Natural-language fields for explanations (summary_text, titles, descriptions).

• Structured fields (JSON-like slots, IDs, scores) for machine reasoning and validation.

• 2.4 Explicit provenance
  Events and entities must include provenance (source, time, responsible component, triggering percepts) to support explainability, auditing, and metrics.

• 2.5 Extendable by domain

• Core ontology here is global and stable.
• Domains (e.g. specific apps, devices, verticals) may add:
  • new subtypes under core types,
  • additional slots,
  • domain-specific relations, without breaking core invariants.

3. High-Level Conceptual Areas

We define the ontology in five interlocking areas:

• A. Agents and Persons
  User(s), AICO as an agent, other people and organisations.

• B. Activities, Goals, and Hobbies
  Projects, tasks, routines, goals, habits, hobbies, plans.

• C. Mental and Social State
  Emotion/mood, personality traits, relationships, social context.

• D. World, Environment, and Sensors
  Places, devices, sensor observations, environment context, PerceptualEvent.

• E. Knowledge, Memory, and Facts
  Memory items, world facts, learned patterns, preferences, skills/tools.

4. Core Types and Identifiers

This section lists the primary types and relations AICO commits to. Actual storage can be in a property graph (nodes/edges) or equivalent, but these types must exist conceptually.

4.1 Agents and Persons

• Person
  Generic human or organisation actor.

• Key fields: person_id, display_name, relation_to_user (self, family, friend, colleague, org, unknown), importance_score.

• User (subtype of Person)
  The primary human owner/controller of AICO.

• Additional fields: preferences summary, safety profile ID, access-rights profile ID.

• AICOAgent
  The software agent representing AICO as an acting entity.

• Fields: agent_id, name, persona_profile_id, capabilities_profile_id.

• Relations

• INTERACTS_WITH(Person, Person|AICOAgent)
• PRIMARY_USER_OF(AICOAgent, User)
• BELONGS_TO(Person, Organisation) (organisation as a Person subtype or separate node type).

4.2 Activities, Goals, Tasks, Hobbies, Routines, Life Areas

These link directly to the Goals & Intentions subsystem schemas, but here we define the conceptual ontology.

• Activity (abstract supertype)
  Something that unfolds over time and may have goals, tasks, and events.

• Project (subtype of Activity)
  Larger, multi-step activity with a clear purpose.

• Fields: project_id, title, description, owner_id (User or AICOAgent), status, tags.

• Task (subtype of Activity)
  Smaller, more concrete unit of work.

• Fields: task_id, title, description, parent_project_id, status, estimated_effort, due_at.

• Habit (subtype of Activity)
  Repeating activity with temporal pattern.

• Fields: habit_id, title, description, schedule_pattern, owner_id.

• Routine (subtype of Activity)
  Recurring, structured pattern of behaviour that may group habits and tasks (e.g. morning routine, weekly review).

• Fields: routine_id, title, description, schedule_pattern, owner_id, stability_score.

• Hobby (subtype of Activity, often AICO-owned)
  Activity primarily driven by AICO’s intrinsic interest, not just user-imposed.

• Fields: hobby_id, title, description, owner_id (often AICOAgent), intrinsic_value_tags, curiosity_link_strength.

• Goal
  Conceptual definition aligns with agency-component-goals-intentions.md:

• A temporally extended desired world/mental/social state.

• Key links: goal_id, owner_id, horizon, origin, state, priority, tags.

• LifeArea
  Conceptual domain of the user’s life that projects, habits, goals, and facts can affect.

• Fields: life_area_id, name (e.g. Health, Work, Relationships, Learning, Creativity), description, priority_score.

• Relations

• HAS_GOAL(Activity|Hobby|Project, Goal)
• PART_OF(Task, Project)
• DERIVED_FROM(Goal_child, Goal_parent) (goal decomposition graph)
• SUPPORTS(Activity|Task, Goal) (task directly supports a goal)
• OWNED_BY(Activity|Goal|Hobby, User|AICOAgent)
• FOCUSES_ON_HOBBY(Goal, Hobby) – links agent-self or curiosity-origin goals to concrete Hobby activities they advance.
• IS_AGENT_SELF_GOAL(Goal) – unary marker (can be modelled as a tag/flag) for goals whose origin = agent_self and whose primary beneficiary is AICO rather than the user.
• PART_OF_ROUTINE(Activity|Habit, Routine) – links specific activities or habits to the routine that organises them.
• AFFECTS_LIFE_AREA(Activity|Goal|WorldStateFact, LifeArea) – encodes which life areas are impacted by a given project, habit, goal, or fact.

4.3 Mental and Social State

• EmotionState
  Coarse-grained emotional or affective label for AICO or user.

• Fields: emotion_id, label (e.g. calm, stressed, curious, bored), intensity, valence, source (perceived, inferred, simulated), timestamp.

• PersonaTrait
  Stable or slowly changing trait defining AICO’s personality.

• Fields: trait_id, name, description, value (e.g. openness: 0.8).

• Relationship
  Social relation between two actors.

• Fields: relationship_id, actor_a, actor_b, type (friend, colleague, romantic, client, etc.), closeness_score, trust_score.

• Relations

• CURRENT_EMOTION(Subject, EmotionState)
• HAS_TRAIT(AICOAgent, PersonaTrait)
• RELATES(Person|AICOAgent, Person|AICOAgent, Relationship)
• EMOTION_INFLUENCES_GOAL(EmotionState, Goal) – conceptual link used by the Goal Arbiter to record that a particular emotional state biased the adoption, suppression, or reprioritisation of a goal (e.g., sustained stress increasing priority of rest-related goals).
• GOAL_AFFECTS_EMOTION(Goal, EmotionState) – link from goals to typical emotional consequences (e.g., hobby or relationship goals associated with positive affect), used by planning and curiosity to prefer emotionally beneficial pursuits when values/safety allow.

These types are used by the Emotion/Personality/Social components and influence Goal origins, priorities, and Arbiter decisions.

4.4 World, Environment, Sensors

• Place / Location
  Physical or virtual place.

• Fields: place_id, name, type (room, city, virtual_room, url), coordinates_or_locator.

• Device
  Device providing sensors or actuation.

• Fields: device_id, type (phone, laptop, camera, microphone, IoT_sensor), owner_id, capabilities.

• SensorObservation
  Low-level reading from a sensor.

• Fields: observation_id, sensor_id, timestamp, raw_value, units, confidence.

• PerceptualEvent
  The central interpreted event unit consumed by the agency subsystem, already defined at high level in other docs. Here we define the core ontology for it.

Minimal conceptual slots: - Identity: percept_id, timestamp, source_component (conversation_engine, world_model, ams, sensor_adapter, external_service_adapter, scheduler, etc.). - Type: percept_type (see 5.1 Event taxonomy). - Natural-language: summary_text. - Structured slots (examples): - actors (list of Person/AICOAgent IDs or entity refs), - topic_tags (strings/ontology tags), - time_window (start, end, granularity), - location_ref (Place ID or descriptor), - salience_score, - urgency_score, - risk_score, - opportunity_score, - confidence_score. - Provenance: - raw_observation_ids (link to SensorObservation, message IDs, logs), - interpretation_chain (which components or LLM calls produced/refined it).

• Relations
• OBSERVED_AT(PerceptualEvent, Place)
• OBSERVED_ON_DEVICE(PerceptualEvent, Device)
• ABOUT(PerceptualEvent, Person|Activity|Goal|Hobby|EnvironmentContext)
• LOCATED_IN(Activity|Device|WorldStateFact, Place) – captures where an activity typically occurs, where a device is used, or where a fact is situated.

4.5 Knowledge, Memory, and Facts

• MemoryItem
  Persistent memory entry in the Adaptive Memory System (AMS).

• Fields: memory_id, type (episodic, semantic, preference, pattern, reflection), summary_text, slots (JSON-like), created_at, last_accessed_at, importance_score, source_percept_id.

• WorldStateFact
  Fact or belief about the world (or the user/AICO) in the World Model.

• Fields: fact_id, subject_entity_id, predicate, object, validity_interval, confidence, source_percept_ids, value_safety_flags (e.g., is_sensitive_domain, requires_explicit_consent, high_risk_context).

• Skill / Tool
  Capability AICO can invoke.

• Fields (ontology-level): skill_id, name, description, input_schema_id, output_schema_id, side_effect_tags, safety_level, life_areas (typical LifeAreas it touches), implementation_ref (pointer into the Skill Registry/tool layer).

• Implementation-level details for concrete Tools (backend, endpoints, auth, runtime_context, resource_profile) live in the Skill & Tool Layer and Skill Registry, not in the core KG schema (see agency-component-skills-tools.md).

• Relations

• REALISES(WorldStateFact, Goal) (fact satisfies or contributes to goal conditions)
• REFERENCES(MemoryItem, Goal|Activity|Person|Hobby)
• LEARNED_FROM(MemoryItem, PerceptualEvent)
• AVAILABLE_TO(AICOAgent, Skill)
• DERIVED_FROM(WorldStateFact, MemoryItem) – links stable facts in the World Model back to AMS memories from which they were inferred.
• EVIDENCED_BY(WorldStateFact, PerceptualEvent) – provenance link from facts to supporting PerceptualEvents.

5. Perceptual Event Taxonomy

This is a taxonomy of percept_type values that upstream components must use when constructing PerceptualEvents. It is global and stable.

5.1 Top-Level Percept Types

• UserIntentEvent
  Interpreted user intention, typically from conversation.

• Example: "User wants to study for exam X next week".

• StateChangeEvent
  Change in world, system, or user state.

• Example: calendar event added/changed; deadline approaching; CPU usage spike.

• PatternEvent
  AMS or World Model detects a pattern or anomaly.

• Example: "User tends to work late before deadlines"; "Drop in mood correlated with social media use".

• SocialEvent
  Social interaction or relationship-relevant signal.

• Example: message from a key contact; relationship conflict; new connection relevant to a project.

• RiskOrOpportunityEvent
  Something that may harm or benefit the user or AICO's long-term objectives.

• Example: overwork risk, health opportunity, learning opportunity, hobby opportunity.

• SystemMaintenanceEvent
  Events related to AICO’s own health, resources, or configuration.

• Example: model performance degradation, memory saturation, safety profile change.

• CuriositySignalEvent
  Internal signal from the Curiosity Engine.

• Example: "Unexplored pattern in user’s creative work"; "Under-explored hobby area".

5.2 PerceptualEvent → Goal / Intention Links

Every PerceptualEvent may:

• Propose a new Goal (e.g. a UserIntentEvent mapping to a new task/project goal).
• Update existing Goals (change priority, state, or supporting evidence).
• Trigger curiosity-driven activities (CuriositySignalEvent → new agent_self or hobby goals).

Data fields (conceptual, aligned with Goals doc):

• candidate_goal_summaries (list of suggested NL goal titles/descriptions).
• candidate_goal_horizon (suggested horizon: theme / project / task).
• candidate_origin (user / agent_self / curiosity / system_maintenance).

The Goals & Intentions component is responsible for consuming these and deciding which become actual goals, but the ontology ensures all components use the same structure.

6. How and Where the Ontology is Used

This section ties the ontology to concrete components described in other agency docs.

6.1 Perception / Sensors Layer

• Upstream owners: Conversation Engine, AMS/World Model, sensor adapters, external service adapters.
• Responsibilities:
• Map raw signals (messages, logs, sensor readings, API payloads) into:
  • SensorObservation nodes (optional, when low-level data is persisted).
  • PerceptualEvent nodes structured according to this doc.
• Use the Perceptual Event Taxonomy (Section 5) to set percept_type.
• Populate provenance fields and candidate goal metadata where applicable.

6.2 World Model Service / KG+Schemas

• Represents Person, User, AICOAgent, Activity, Project, Task, Habit, Hobby, Place, Device, MemoryItem, WorldStateFact, Skill as nodes and edges according to Section 4.
• Uses PerceptualEvents as ingestion edges:
• LEARNED_FROM(MemoryItem, PerceptualEvent)
• REALISES(WorldStateFact, Goal) via events that confirm success.
• Provides schema-aware APIs to other components (see agency-component-world-model.md).

6.3 Goals & Intentions Component

• Consumes PerceptualEvents and:
• Creates/updates Goal nodes and their graph structure.
• Links goals to Activities, Hobbies, Persons, WorldStateFacts.
• Uses ontology relations:
• HAS_GOAL(Activity, Goal)
• DERIVED_FROM(Goal_child, Goal_parent)
• SUPPORTS(Task, Goal).
• Tracks provenance via:
• last_percept_event_id and related fields described in agency-component-goals-intentions.md.

6.4 Curiosity Engine

• Reads from World Model and AMS using ontology types:
• Under-represented topics, patterns, or Hobbies.
• Emits CuriositySignalEvent PerceptualEvents.
• Proposes new agent_self or hobby Goals that are clearly distinguished by origin and owner_id.

6.5 Values, Ethics, and Safety

• Uses ontology structure to:
• Identify actors (User vs others vs AICO) involved in actions.
• Understand relations (e.g. closeness of relationship).
• Assess risk/opportunity based on tagged types and scores.
• Can apply policies conditioned on:
• Percept types (e.g. special scrutiny for RiskOrOpportunityEvent).
• Entity types (e.g. stricter rules for financial activities).

6.6 Metrics and Explainability

• Metrics in agency-metrics.md refer to ontology-based entities:
• Goals by horizon/origin/state.
• PerceptualEvents by type, salience, origin component.
• Hobby-related goals and activities.
• Explainability:
• When explaining why AICO did something, references are:
  • PerceptualEvent chain,
  • linked MemoryItems and WorldStateFacts,
  • related Goals and Activity objects.

7. Representation and Implementation Notes

• Primary store: property/knowledge graph (nodes/edges) backed by the shared libSQL store used by AMS and the World Model, plus:
• JSON-like schemas for PerceptualEvents, MemoryItems, Goals, Skills, and auxiliary libSQL tables for policy (ValueProfile/PolicyRule) and skill registries.
• We do not require heavy OWL reasoning in v1.
  Reasoning is largely delegated to:
• LLM-based components (goal interpretation, planning, explanation).
• Lightweight rules and validators (type checks, policy checks, referential integrity).
• Schema evolution:
• Core types and relations in Section 4 and Section 5 must remain stable or evolve via explicit migrations.
• Domain- and feature-specific extensions must not violate the meaning of core types.

8. References and Cross-Links

• docs/concepts/agency/agency.md – overall concept of Agency.
• docs/concepts/agency/agency-architecture.md – architecture and Perception/Sensors layer.
• docs/concepts/agency/agency-component-world-model.md – world model and knowledge/property graph integration.
• docs/concepts/agency/agency-component-goals-intentions.md – detailed goal/intention data model and operations.
• docs/concepts/agency/agency-component-curiosity-engine.md – curiosity-driven signals and agent-self/hobby goals.
• docs/concepts/agency/agency-component-values-ethics.md – values, ethics, and safety policies consuming ontology types.