RESEARCH
Empirical Validation Program for Empathy Systems Theory and HEART Framework
OVERVIEW
The Heart AI Foundation is committed to open science. EST and HEART rest on testable hypotheses, not faith-based assertions. Constitutional principles governing emotional AI require empirical demonstration of effectiveness. We will publish findings positive and negative. We will share data and methods. We will welcome replication and critique. We will adapt our frameworks when evidence demands. The validation pathway spans 15 years with explicit decision points for progression, revision, or abandonment. Multiple pathways lead to framework abandonment if predictions fail—this falsifiability commitment distinguishes scientific theory from ideology.
VALIDATION PATHWAY
Phase 1A/B (Years 0-2): Foundation
- Sociopathy natural experiment
- CAEI four-factor validation
- Trust operationalization
- Falsifiable predictions testing
Phase 2-3 (Years 2-6): Mechanism Testing - CEOP cascade sequence studies
- SNIA-generativity studies –
- Infrastructure-outcome correlations
Phase 4 (Years 4-8): Intervention Effectiveness - Infrastructure restoration trials
- NES coordination experiments
- Repair vs. skill training comparisons
Phase 5 (Years 8-15): Cross-Cultural Validation - CAEI universal vs. cultural specificity
- CAEI-D module validation across populations
- Scope boundary determination
PREREGISTERED STUDIES
Sycophancy Natural Experiment
Status: Preregistered (January 2026)
Tests the Knowing-Feeling Dissociation principle: Does cognitive awareness of AI sycophantic behavior protect users from calibration accuracy degradation?
Three conditions compare Sycophant-Unaware, Sycophant-Aware, and Calibrating AI. Primary outcome is self-assessment calibration accuracy measured against human evaluator ratings.
If confirmed, behavioral architecture mandates in HEART Framework are empirically justified. Disclosure alone is insufficient for user protection.
Assessment instrument: Emotional Precision Sycophancy Battery (EP-S) v1.0
Emotional Precision Natural Experiment (Turing-Blind)
Status: Protocol specified; preregistration forthcoming
Tests whether Emotional Precision requires ongoing calibration from experiencing beings, not merely intact infrastructure.
Turing-blind design: Participants do not know whether they interact with human or AI. Compares Standard AI (non-NES compliant) vs. HEART-compliant AI effects on Emotional Precision with human targets post-exposure.
If confirmed, three-layer EST architecture is validated and NES governance is empirically grounded.
CAEI Validation Battery
Status: Protocol specified
Multi-phase psychometric validation of the Comprehensive Assessment of Empathy Infrastructure (CAEI) including factor structure and internal consistency, convergent and discriminant validity, known-groups validity comparing clinical and normative populations, test-retest reliability, cross-cultural invariance, and sensitivity to change in intervention
Critical test: Advanced contemplative practitioners should show high infrastructure scores alongside low narrative construction, confirming the content-neutrality principle.contexts.
Natural Experiments
Sociopathy Natural Experiment
Individuals with sociopathic presentations can simulate empathic behaviors but cannot sustain simulation under conditions that infrastructure-enabled processing sustains.
Five distinguishing signatures: differential sustainability under extended demands, coordination signature comparing parallel vs. sequential processing, neural substrates comparing limbic vs. prefrontal recruitment, cognitive load response under dual-task conditions, and phenomenology comparing intuitive vs. calculated processing.
If sociopathic simulation matches genuine empathy, EST infrastructure necessity claim is falsified.
Burnout Cascade Signature
Burnout progression should show a specific cascade signature rather than random decline.
Early-stage burnout shows Core Authenticity impairment with preserved downstream components. Late-stage burnout shows full cascade completion.
This tests the CEOP mechanism specification in naturalistic workplace contexts.
Jurisdictional Comparison Studies
Population-level outcome comparisons between HEART-compliant and non-compliant jurisdictions as adoption spreads.
Metrics include mental health service utilization, AI companion-related incidents, and emotional infrastructure population indicators.
Research Partnerships
The Heart AI Foundation welcomes collaboration with academic institutions, clinical research programs, and AI ethics research centers.
Partnership models include independent validation studies where the Foundation provides frameworks and researchers conduct independent testing, coordinated multi-site studies with shared protocols and distributed data collection, and adversarial collaboration where researchers design studies to falsify EST and HEART claims.
Current partnership interests include CAEI cross-cultural validation with non-Western research teams, clinical outcome studies in burnout and trauma populations, AI interaction effects on adolescent emotional development, and Guardian assessment inter-rater reliability studies.
Commitment to Transparency
All preregistrations are publicly accessible via OSF.
Negative results are published alongside positive findings.
Data and analysis code are shared upon publication.
Framework revision is documented when evidence demands change.
Three possible outcomes advance understanding. Full validation integrates EST and HEART with clinical practice and AI governance. Partial validation requires architectural revision with scope constraints. Falsification means the framework is abandoned and alternative mechanisms are identified.
This is not failure. This is science working as it should.
Last updated: January 2026
Heart AI Foundation
empathyethicist.ai