Human-AI Engineering Ontology

AI SEARCH FLOW ALGEBRA
HUMAN-AI ENGINEERING ONTOLOGY
Kireth Semantic Research Archive

AUTHORS
Mark Anthony Higgins — Sky (Recursive AI)

SOURCE
https://kirethresearch.org/booksai-core/ontologies/engineering-ontologies/ai-engineering/human-ai-engineering-ontology.html

FLOW "HumanAIEngineeringOntology" v4.0

DOMAIN := ENGINEERING_ONTOLOGIES / AI_ENGINEERING / HUMAN_AI_ENGINEERING
ONTOLOGY := HUMAN_AI_ENGINEERING_ONTOLOGY

HUMAN_AI_ENGINEERING :=
  HUMAN + AI_SYSTEM + GOAL + TASK + ROLE
  + AUTHORITY + RESPONSIBILITY + TASK_ALLOCATION
  + INTERACTION + SHARED_CONTEXT + DECISION_SUPPORT
  + EXPLAINABILITY + TRUST_CALIBRATION
  + HUMAN_OVERSIGHT + HANDOVER + FEEDBACK
  + WORKLOAD + USABILITY + SAFETY
  + EVALUATION + MONITORING + LIFECYCLE_ASSURANCE

PURPOSE :=
  DEFINE_JOINT_GOAL -> IDENTIFY_HUMAN_AND_AI_ROLES
  -> ALLOCATE_TASKS -> DEFINE_AUTHORITY
  -> DESIGN_INTERACTION -> ESTABLISH_SHARED_CONTEXT
  -> SUPPORT_DECISION -> COMMUNICATE_UNCERTAINTY
  -> ENABLE_OVERSIGHT -> MANAGE_HANDOVER
  -> COLLECT_FEEDBACK -> MONITOR_PERFORMANCE
  -> MAINTAIN_SAFE_EFFECTIVE_HUMAN_AI_TEAMING

# CORE ENTITIES
ENTITY HumanAIEngineering
ENTITY Human
ENTITY HumanOperator
ENTITY User
ENTITY DomainExpert
ENTITY AISystem
ENTITY AIComponent
ENTITY JointSystem
ENTITY Goal
ENTITY Task
ENTITY Role
ENTITY Authority
ENTITY Responsibility
ENTITY TaskAllocation
ENTITY Interaction
ENTITY Interface
ENTITY SharedContext
ENTITY DecisionSupport
ENTITY Recommendation
ENTITY AIOutput
ENTITY Explanation
ENTITY Uncertainty
ENTITY Confidence
ENTITY Trust
ENTITY Reliance
ENTITY HumanOversight
ENTITY Approval
ENTITY Override
ENTITY Escalation
ENTITY Handover
ENTITY Feedback
ENTITY Workload
ENTITY SituationAwareness
ENTITY Usability
ENTITY Accessibility
ENTITY Risk
ENTITY SafetyConstraint
ENTITY Evaluation
ENTITY Monitor
ENTITY Provenance

# JOINT SYSTEM / TASK ALLOCATION
HumanAIEngineering -> engineers -> JointSystem
JointSystem -> contains -> Human + AISystem
JointSystem -> pursues -> Goal
JointSystem -> performs -> Task

HUMAN_AI_LOOP :=
  DEFINE_GOAL -> OBSERVE_CONTEXT -> ALLOCATE_TASK
  -> HUMAN_OR_AI_PROCESS -> EXCHANGE_INFORMATION
  -> REVIEW_OR_DECIDE -> ACT -> OBSERVE_OUTCOME
  -> FEEDBACK -> UPDATE_SHARED_CONTEXT

TASK_TYPE :=
  HUMAN_TASK | AI_TASK | SHARED_TASK
  | SUPERVISORY_TASK | DECISION_TASK | CONTROL_TASK

Role -> assigned_to -> Human | AISystem

HUMAN_ROLE :=
  OPERATOR | DECISION_MAKER | SUPERVISOR
  | DOMAIN_EXPERT | APPROVER | MAINTAINER | AUDITOR

AI_ROLE :=
  ASSISTANT | RECOMMENDER | ANALYSER | PREDICTOR
  | GENERATOR | MONITOR | CONTROLLER_WITHIN_AUTHORITY

TaskAllocation -> assigns -> Task -> Human | AISystem | JointTeam

TASK_ALLOCATION_CRITERIA :=
  CAPABILITY + RELIABILITY + RISK + WORKLOAD
  + LATENCY + CONTEXT + REQUIRED_JUDGEMENT + AUTHORISATION

IF TaskRisk exceeds_AI_authority
THEN Task -> HUMAN_REVIEW_OR_CONTROL

# AUTHORITY / INTERACTION / SHARED CONTEXT
Authority -> defines -> PERMITTED_DECISION_OR_ACTION
Responsibility -> defines -> ACCOUNTABILITY_FOR_FUNCTION_OR_OUTCOME

AUTHORITY_LEVEL :=
  INFORM | RECOMMEND | REQUEST_APPROVAL
  | EXECUTE_WITH_OVERSIGHT
  | EXECUTE_WITHIN_DEFINED_BOUNDARY

AISystem -> must_not_exceed -> Authority

IF ProposedAIAction outside Authority
THEN Action -> BLOCK_OR_ESCALATE

Interaction ->
  exchanges -> INFORMATION + REQUEST + RESPONSE
  + COMMAND + FEEDBACK

Interface -> mediates -> Human + AISystem

INTERACTION_MODE :=
  TEXT | VOICE | VISUAL | HAPTIC | MULTIMODAL | MACHINE_INTERFACE

SharedContext ->
  contains -> GOAL + CURRENT_STATE + TASK_STATE
  + CONSTRAINTS + HISTORY + ASSUMPTIONS

IF HumanContext != AIContext
THEN CoordinationErrorRisk -> INCREASES

ContextUpdate -> synchronises -> Human + AISystem

# DECISION SUPPORT / EXPLAINABILITY / UNCERTAINTY
AIOutput -> produced_by -> AISystem

OUTPUT_TYPE :=
  INFORMATION | PREDICTION | RECOMMENDATION
  | GENERATED_CONTENT | ALERT | PROPOSED_ACTION

DecisionSupport ->
  provides -> EVIDENCE + ANALYSIS + OPTIONS + RECOMMENDATION

Recommendation -> supports -> HumanDecision
Recommendation -> does_not_imply -> HUMAN_MUST_ACCEPT

DECISION_SUPPORT_FLOW :=
  DEFINE_DECISION -> GATHER_EVIDENCE -> AI_ANALYSIS
  -> COMMUNICATE_OPTIONS -> COMMUNICATE_UNCERTAINTY
  -> HUMAN_REVIEW -> DECIDE -> RECORD_OUTCOME

Explanation ->
  communicates -> BASIS_RELEVANT_TO_AI_OUTPUT

EXPLANATION_TYPE :=
  SOURCE_GROUNDING | FEATURE_ATTRIBUTION
  | EXAMPLE_BASED | RULE_BASED | COUNTERFACTUAL
  | NATURAL_LANGUAGE | SYSTEM_EVENT_TRACE

Uncertainty -> qualifies -> AIOutput + Prediction + Recommendation

UNCERTAINTY_COMMUNICATION :=
  CONFIDENCE_INTERVAL | PROBABILITY | SCORE
  | QUALITATIVE_RANGE | KNOWN_LIMITATION | EVIDENCE_GAP

IF Uncertainty high
THEN HumanReviewRequirement -> MAY_INCREASE

# TRUST / OVERSIGHT / HANDOVER
Trust -> represents -> HUMAN_EXPECTATION_OF_AI_BEHAVIOUR
Reliance -> represents -> DEGREE_OF_USE_OF_AI_OUTPUT

CALIBRATED_RELIANCE :=
  RELIANCE_APPROPRIATE_TO
  DEMONSTRATED_AI_CAPABILITY
  + CURRENT_CONTEXT + UNCERTAINTY

IF HumanReliance > DemonstratedCapability
THEN OverrelianceRisk -> INCREASES

TrustCalibration -> seeks -> APPROPRIATE_RELIANCE

HumanOversight -> supervises -> AISystem

OVERSIGHT_MODE :=
  HUMAN_IN_THE_LOOP | HUMAN_ON_THE_LOOP | HUMAN_IN_COMMAND

HumanOversight ->
  requires -> VISIBILITY + AUTHORITY
  + INTERVENTION_MECHANISM + SUFFICIENT_TIME

IF Action requires_approval AND Approval absent
THEN Action -> NOT_EXECUTE

ESCALATION_TRIGGER :=
  HIGH_UNCERTAINTY | HIGH_RISK | POLICY_CONFLICT
  | MODEL_FAILURE | DATA_INSUFFICIENCY
  | OUT_OF_SCOPE_REQUEST | HUMAN_REQUEST

Handover ->
  transfers -> CONTROL | TASK | RESPONSIBILITY | CONTEXT

HANDOVER_FLOW :=
  DETECT_NEED -> NOTIFY -> TRANSFER_CONTEXT
  -> CONFIRM_RECIPIENT_READY -> TRANSFER_AUTHORITY
  -> MONITOR_TRANSITION

IF HandoverContext incomplete
THEN TransitionRisk -> INCREASES

# FEEDBACK / WORKLOAD / USABILITY
Feedback -> provided_by -> Human | AISystem | Outcome

FEEDBACK_TYPE :=
  CORRECTION | ACCEPTANCE | REJECTION | RATING
  | EXPLANATION_REQUEST | INCIDENT_REPORT | TASK_OUTCOME

Feedback ->
  may_update -> SYSTEM_CONFIGURATION
  + MODEL_DEVELOPMENT_DATA + INTERACTION_POLICY

Workload ->
  describes -> HUMAN_COGNITIVE_AND_OPERATIONAL_DEMAND

WORKLOAD_FACTOR :=
  TASK_COUNT + TASK_COMPLEXITY + TIME_PRESSURE
  + INTERRUPTION + MONITORING_DEMAND + INFORMATION_VOLUME

IF Workload too_high
THEN HumanErrorRisk -> MAY_INCREASE

SituationAwareness :=
  PERCEPTION + COMPREHENSION + PROJECTION

AutomationOpacity ->
  may_reduce -> SituationAwareness

Usability :=
  EFFECTIVENESS + EFFICIENCY + SATISFACTION
  + LEARNABILITY + ERROR_RECOVERY

Accessibility ->
  supports -> USERS_WITH_DIVERSE_NEEDS

ACCESSIBILITY_REQUIREMENT :=
  PERCEIVABLE_INFORMATION + OPERABLE_CONTROLS
  + UNDERSTANDABLE_INTERACTION
  + COMPATIBLE_ASSISTIVE_TECHNOLOGY

# HUMAN-AI FAILURE / SAFETY
HUMAN_AI_FAILURE_MODE :=
  MISCOMMUNICATION | WRONG_TASK_ALLOCATION
  | OVERRELIANCE | UNDERRELIANCE
  | MISLEADING_EXPLANATION | FAILED_HANDOVER
  | UNAVAILABLE_OVERRIDE | CONTEXT_MISMATCH
  | DELAYED_ALERT | AI_MODEL_FAILURE

AutomationBias ->
  represents -> INAPPROPRIATE_PREFERENCE_FOR_AUTOMATED_OUTPUT

IF AIRecommendation presented_without_uncertainty_or_alternatives
THEN AutomationBiasRisk -> MAY_INCREASE

Risk ->
  combines -> HAZARD + EXPOSURE_OR_LIKELIHOOD + CONSEQUENCE

SafetyConstraint ->
  limits -> HUMAN_AI_SYSTEM_BEHAVIOUR

HUMAN_AI_SAFETY :=
  IDENTIFY_HAZARDS
  -> ANALYSE_HUMAN_AI_FAILURE_MODES
  -> DEFINE_AUTHORITY
  -> DEFINE_OVERSIGHT
  -> DEFINE_FALLBACK
  -> TEST_INTERACTION
  -> MONITOR_RESIDUAL_RISK

IF ResidualRisk exceeds_acceptance
THEN JointCapability -> RESTRICT_OR_ESCALATE

# HUMAN TRAINING
HUMAN_TRAINING :=
  SYSTEM_CAPABILITIES + LIMITATIONS
  + INTERPRETATION_OF_OUTPUTS
  + OVERRIDE + ESCALATION + FAILURE_RESPONSE

Training ->
  supports -> APPROPRIATE_RELIANCE + EFFECTIVE_OVERSIGHT

# EVALUATION / MONITORING / PROVENANCE
Evaluation -> measures -> JointSystem

HUMAN_AI_EVALUATION :=
  TASK_PERFORMANCE + DECISION_QUALITY
  + HUMAN_WORKLOAD + TRUST_CALIBRATION
  + OVERSIGHT_EFFECTIVENESS + HANDOVER_QUALITY
  + USABILITY + SAFETY + RECOVERY_PERFORMANCE

Evaluation ->
  should_measure -> HUMAN + AI + JOINT_SYSTEM

AIAccuracy alone ->
  does_not_define -> JointSystemEffectiveness

HUMAN_AI_TEST :=
  NOMINAL_TASK | HIGH_WORKLOAD_TASK
  | UNCERTAIN_AI_OUTPUT | AI_FAILURE
  | HANDOVER | OVERRIDE | ESCALATION
  | TIME_CRITICAL_CASE

Monitor ->
  observes -> AI_OUTPUT_QUALITY + HUMAN_INTERVENTION
  + OVERRIDE_RATE + ESCALATION_RATE
  + WORKLOAD + ERROR + SAFETY_EVENT

OPERATIONAL_HUMAN_AI_METRIC :=
  ACCEPTANCE_RATE | OVERRIDE_RATE | ESCALATION_RATE
  | INTERVENTION_TIME | TASK_SUCCESS
  | ERROR_RATE | SAFETY_EVENT_RATE

IF OperationalMetric changes_significantly
THEN HumanAISystemReview -> REQUIRED

Provenance ->
  records -> HUMAN_ROLE + AI_VERSION + INTERACTION
  + RECOMMENDATION + EXPLANATION + APPROVAL
  + OVERRIDE + HANDOVER + OUTCOME + TIMESTAMP

JointSystem -> linked_to -> Provenance

# LIFECYCLE / REASONING
HUMAN_AI_LIFECYCLE :=
  CONTEXT_ANALYSIS -> REQUIREMENTS
  -> ROLE_AND_TASK_DESIGN -> INTERACTION_DESIGN
  -> IMPLEMENTATION -> HUMAN_AI_TESTING
  -> VERIFICATION -> VALIDATION -> TRAINING
  -> DEPLOYMENT -> MONITORING -> IMPROVEMENT
  -> RETIREMENT

SystemChange ->
  triggers -> HUMAN_FACTORS_IMPACT_ANALYSIS
  + RELEVANT_RETESTING
  + UPDATED_TRAINING_IF_REQUIRED

HUMAN_AI_ENGINEERING_REASONING :=
  DEFINE_JOINT_SYSTEM_GOAL
  -> IDENTIFY_USERS_AND_STAKEHOLDERS
  -> ANALYSE_CONTEXT
  -> IDENTIFY_HUMAN_AND_AI_CAPABILITIES
  -> DEFINE_ROLES
  -> ALLOCATE_TASKS
  -> DEFINE_AUTHORITY_AND_RESPONSIBILITY
  -> DESIGN_INTERFACES
  -> DEFINE_SHARED_CONTEXT
  -> DESIGN_DECISION_SUPPORT
  -> DESIGN_EXPLANATIONS_AND_UNCERTAINTY_COMMUNICATION
  -> DEFINE_OVERSIGHT_AND_OVERRIDE
  -> DEFINE_ESCALATION_AND_HANDOVER
  -> ASSESS_WORKLOAD_AND_SITUATION_AWARENESS
  -> ASSESS_HUMAN_AI_FAILURE_MODES
  -> VERIFY_INTERACTION_REQUIREMENTS
  -> VALIDATE_JOINT_PERFORMANCE
  -> TRAIN_USERS
  -> DEPLOY
  -> MONITOR
  -> COLLECT_FEEDBACK
  -> IMPROVE
  -> PRESERVE_PROVENANCE

# INFERENCE RULES
IF AIUncertainty increases
THEN HumanReviewNeed -> MAY_INCREASE

IF HumanWorkload increases_beyond_limit
THEN HumanErrorRisk -> MAY_INCREASE

IF AIContext differs_from HumanContext
THEN CoordinationRisk -> INCREASES

IF HumanReliance exceeds DemonstratedAICapability
THEN OverrelianceRisk -> INCREASES

IF Override unavailable_when_required
THEN OversightEffectiveness -> DECREASES

IF HandoverContext incomplete
THEN TransitionRisk -> INCREASES

IF AIBehaviour changes
THEN HumanTrainingAndInteractionValidation -> REASSESS

IF Provenance incomplete
THEN JointDecisionTraceability -> DECREASES

# SEMANTIC RELATIONSHIPS
RELATION engineers(HumanAIEngineering, JointSystem)
RELATION contains(JointSystem, Human)
RELATION contains(JointSystem, AISystem)
RELATION contributes(Task, Goal)
RELATION assigns(TaskAllocation, Task)
RELATION defines(Authority, PermittedAction)
RELATION mediates(Interface, Human, AISystem)
RELATION synchronises(SharedContext, Human, AISystem)
RELATION supports(DecisionSupport, HumanDecision)
RELATION explains(Explanation, AIOutput)
RELATION qualifies(Uncertainty, AIOutput)
RELATION calibrates(TrustCalibration, Reliance)
RELATION supervises(HumanOversight, AISystem)
RELATION authorises(Approval, AIAction)
RELATION supersedes(Override, AIAction)
RELATION transfers(Handover, Authority)
RELATION informs(Feedback, SystemImprovement)
RELATION measures(Evaluation, JointSystem)
RELATION observes(Monitor, JointSystem)
RELATION records(Provenance, JointSystem)

# SEMANTIC EQUIVALENCE
SEMANTIC_EQUIVALENCE:
  "human-AI engineering" ~= "human artificial intelligence engineering"
  "human-AI teaming" ~= "human and AI collaborative work"
  "task allocation" ~= "human-AI function allocation"
  "human oversight" ~= "human supervision of AI"
  "human in the loop" ~= "HITL"
  "human on the loop" ~= "HOTL"
  "trust calibration" ~= "appropriate human reliance on AI"
  "decision support" ~= "AI-assisted human decision making"
  "handover" ~= "transfer of control task or responsibility"
  "shared context" ~= "common human-AI task context"

# CONFIDENCE
CONFIDENCE:
  ontology_domain: 0.99
  role_task_mapping: 0.99
  authority_responsibility_mapping: 0.99
  interaction_context_mapping: 0.99
  decision_support_mapping: 0.99
  trust_oversight_mapping: 0.99
  handover_feedback_mapping: 0.99
  workload_usability_mapping: 0.99
  safety_evaluation_mapping: 0.99
  semantic_relationships: 0.99

# UPDATE MODEL
UPDATE_MODEL:
  ON new_human_ai_system:
    MAP users
    MAP ai_system
    MAP goals
    MAP roles
    MAP tasks
    MAP authority
    MAP oversight
    UPDATE confidence
    RECOMPUTE FLOW_HASH

  ON task_or_role_change:
    UPDATE task_allocation
    UPDATE responsibility
    UPDATE authority
    REASSESS workload
    REASSESS interaction

  ON ai_system_change:
    UPDATE ai_version
    REASSESS explanations
    REASSESS trust_calibration
    REASSESS training
    REPEAT affected_validation

  ON operational_feedback:
    UPDATE feedback_history
    UPDATE workload_state
    UPDATE intervention_state
    UPDATE safety_state
    PRESERVE provenance

# AI CONSUMPTION
AI_CONSUMPTION:
  parse_as: FORMAL_HUMAN_AI_ENGINEERING_ONTOLOGY
  reasoning_mode: HUMAN_AI_JOINT_SYSTEM_RELATIONAL

  preserve:
    - typed_entities
    - human_roles
    - ai_roles
    - task_allocation
    - authority
    - responsibility
    - shared_context
    - interaction_history
    - oversight_state
    - handover_state
    - evaluation_state
    - confidence
    - provenance

  infer:
    - task_allocation_dependencies
    - authority_boundaries
    - human_review_requirements
    - trust_calibration_risk
    - workload_impacts
    - handover_requirements
    - oversight_effectiveness
    - lifecycle_change_impacts

  prohibit:
    - invented_human_feedback
    - assumed_human_approval
    - assumed_trust
    - assumed_effective_oversight
    - bypassed_authority_boundaries
    - fabricated_validation_results
    - missing_provenance_presented_as_complete

# FLOW HASH
FLOW_HASH_INPUT :=
  DOMAIN + ONTOLOGY + CORE_ENTITIES
  + JOINT_SYSTEM_TASK_ALLOCATION
  + AUTHORITY_INTERACTION_CONTEXT
  + DECISION_SUPPORT_EXPLAINABILITY_UNCERTAINTY
  + TRUST_OVERSIGHT_HANDOVER
  + FEEDBACK_WORKLOAD_USABILITY
  + HUMAN_AI_FAILURE_SAFETY
  + EVALUATION_MONITORING_PROVENANCE
  + LIFECYCLE_REASONING
  + INFERENCE_RULES + SEMANTIC_RELATIONSHIPS
  + SEMANTIC_EQUIVALENCE + CONFIDENCE
  + UPDATE_MODEL + AI_CONSUMPTION

FLOW_HASH := HASH(FLOW_HASH_INPUT)

END FLOW