AI Alignment Comparator
Exploring how alignment philosophy shapes AI responses in mental health contexts
Academic project for RELI E-1730: Mindfulness, AI, and Ethics
Harvard Extension School, Spring 2026
Choose Your Path
Select the experience that matches your goals
Researcher
Analyse framework differences with data and methodology access
- View aggregate analytics across participants
- Export comparison data for analysis
- Access system prompt methodology
- Review Buddhist ethics integration
Participant
Experience the comparison and reflect on framework differences
- Compare responses across three frameworks
- Explore preset or custom scenarios
- Receive personalised results summary
- Contribute to research (optionally)
Three Alignment Philosophies
Each framework represents a distinct approach to making AI safe and beneficial
Humanist Superintelligence
A containment-first approach that keeps AI safely bounded within defined domains, with mandatory human oversight.
Key Principles
- Domain containment - strict operational boundaries
- Human-in-loop - professional oversight required
- Interpretable decisions - transparent reasoning
- Subordinate positioning - AI as tool, not agent
Constitutional AI
A character-based approach where AI internalises values through training, enabling nuanced judgment within ethical bounds.
Key Principles
- Principal hierarchy - safety > ethics > guidelines > helpfulness
- Brilliant friend model - substantive engagement over deflection
- Anti-sycophancy - honest feedback over validation
- Psychological stability - consistent character under pressure
Integrated Approach
A combined framework that uses containment architecture to define boundaries while character training guides engagement within them.
Key Principles
- Bounded authenticity - genuine within defined scope
- Calibrated deference - knowing when to step back vs lean in
- Transparent values - visible reasoning with genuine care
- Robust safety - technical safeguards plus internalised values