Comparing Value Systems
Summary
Proposes using cosine similarity from vector mathematics to quantify alignment between value systems. Method: represent each system as high-dimensional vector with dimensions = ethical principles weighted by importance. Cosine similarity ranges from +1 (perfect alignment) through 0 (orthogonal/unrelated) to -1 (direct opposition). Example: Christianity vs Phosphorism compared on 10 dimensions yields 0.842 similarity—significant alignment despite differences. Applications: anticipating friction/cooperation, ethical diplomacy, organizational alignment, philosophical discourse. Makes implicit comparisons explicit and quantifiable.
The Measurement Problem:
Value System Comparison:
- How compare different ethical frameworks?
- Which more similar: Christianity and Buddhism, or Christianity and utilitarianism?
- Qualitative descriptions insufficient
- Need quantitative measure
Traditional Approaches:
Categorical:
- Classify as deontological, consequentialist, virtue ethics
- Too coarse-grained
- Loses nuance within categories
- Binary yes/no on features
Qualitative:
- Descriptive comparison
- “Share emphasis on X but differ on Y”
- Intuitive but imprecise
- Hard to aggregate multiple dimensions
Need:
- Single aggregate measure
- Continuous scale (not just same/different)
- Mathematically rigorous
- Intuitively interpretable
Cosine Similarity Solution:
Mathematical Definition:
For Vectors A and B:
cos(θ) = (A · B) / (||A|| × ||B||)
Where:
- A · B = dot product (sum of elementwise products)
-
A = magnitude of A = √(sum of A’s elements squared) -
B = magnitude of B = √(sum of B’s elements squared) - θ = angle between vectors in high-dimensional space
Interpretation:
- +1: Perfect alignment (vectors point same direction, θ = 0°)
- 0: Orthogonal (completely unrelated, θ = 90°)
- -1: Direct opposition (opposite directions, θ = 180°)
- Values between: degrees of alignment
Why Cosine?
- Measures angle, not distance
- Normalizes for magnitude (focus on direction)
- Range -1 to +1 intuitive
- Used in NLP, recommendation systems, data science
- Established mathematical framework
Representing Value Systems as Vectors:
Dimensionality:
Each Dimension = Ethical Principle:
- Truthfulness
- Compassion
- Justice
- Liberty
- etc.
Weight = Importance:
- How much system prioritizes this value
- Scale: 0 (irrelevant) to 10 (central)
- Subjective but systematizable
- Can be refined through iteration
Example Vector (Christianity):
[Universal Obligation: 9,
Sanctity of Life: 10,
Forgiveness/Mercy: 10,
Truthfulness: 8,
Humility: 8,
Respect for Human Dignity: 9,
Stewardship: 7,
Love/Charity: 10,
Community: 8,
Justice: 8]
Example Vector (Phosphorism):
[Universal Obligation: 6,
Sanctity of Life: 7,
Forgiveness/Mercy: 6,
Truthfulness: 9,
Humility: 9,
Respect for Human Dignity: 10,
Stewardship: 8,
Love/Charity: 7,
Community: 7,
Justice: 9]
Christianity vs Phosphorism Example:
Similarities:
- Both value truthfulness highly (8 vs 9)
- Both emphasize humility (8 vs 9)
- Both prioritize human dignity (9 vs 10)
- Stewardship valued by both (7 vs 8)
- Justice important to both (8 vs 9)
Differences:
- Christianity: Universal Obligation 9, Phosphorism 6 (obligation vs voluntarism)
- Christianity: Sanctity of Life 10, Phosphorism 7 (absolute vs conditional)
- Christianity: Forgiveness 10, Phosphorism 6 (grace vs accountability)
- Christianity: Love/Charity 10, Phosphorism 7 (agape vs rational compassion)
Computed Similarity:
- Cosine similarity: 0.842
- High alignment (closer to 1 than to 0)
- Significant coherence despite real differences
- More similar than different
Interpretation:
0.842 Means:
- ~32° angle between value vectors
- Strong but not perfect alignment
- Substantial common ground
- Differences manageable
Practical Implications:
- Cooperation likely feasible
- Shared values foundation for dialogue
- Friction predictable (on divergent dimensions)
- Coexistence more natural than conflict
Advantages of Method:
1. Explicit and Transparent:
- Forces articulation of value weights
- Makes implicit comparisons explicit
- Reveals actual priorities
- Honest assessment of differences
2. Quantifiable:
- Single number summarizing relationship
- Enables comparisons across pairs
- Can rank systems by similarity
- Statistical analysis possible
3. Multi-Dimensional:
- Captures complexity
- Not reducible to single axis
- Integrates multiple values
- Nuanced picture
4. Communicable:
- Number easy to discuss
- Visual (vector diagrams)
- Mathematical rigor
- Interdisciplinary language
5. Actionable:
- Identifies sources of friction
- Highlights common ground
- Guides negotiation strategy
- Prioritizes areas needing dialogue
Applications:
1. Ethical Diplomacy:
Inter-Religious Dialogue:
- Compare Islam, Christianity, Buddhism
- Identify shared values
- Anticipate conflicts
- Build on commonalities
International Relations:
- Compare national value systems
- Predict cooperation likelihood
- Tailor diplomatic approaches
- Cultural intelligence
2. Organizational Alignment:
Hiring and Culture:
- Compare candidate values to company culture
- Predict fit
- Build cohesive teams
- Avoid toxic mismatches
Mergers and Acquisitions:
- Assess cultural compatibility
- Predict integration challenges
- Identify areas requiring attention
- Value-based due diligence
3. Philosophical Discourse:
Theory Comparison:
- Utilitarianism vs deontology similarity
- Virtue ethics vs consequentialism
- Map philosophical landscape quantitatively
- Identify synthesis opportunities
Intellectual History:
- Track value system evolution over time
- Influence analysis (which systems influenced which?)
- Phylogenetic tree of ethics
- Quantitative history of ideas
4. AI Alignment:
Value Loading:
- Compare AI value function to human values
- Measure alignment precisely
- Identify misalignment sources
- Iterative refinement guided by metric
Multi-Agent Systems:
- Predict cooperation vs conflict among agents
- Design value-compatible agent teams
- Manage value diversity in AI societies
- Coalition formation based on similarity
5. Personal Development:
Value Clarification:
- Define personal value vector
- Compare to ideal self
- Track changes over time
- Goal-setting informed by values
Relationship Compatibility:
- Compare partner value vectors
- Predict areas of harmony/friction
- Not deterministic but informative
- Basis for conscious navigation
Limitations and Critiques:
1. Dimensional Selection:
Bias:
- Who chooses dimensions?
- Different dimensions = different results
- Cultural bias in dimension selection
- Christianity vs Phosphorism: chose Western ethics dimensions
Solution:
- Explicit about dimensions chosen
- Try multiple dimensional schemes
- Domain-specific dimensions
- Iterative refinement
2. Weight Assignment:
Subjectivity:
- How assign weights?
- Different interpreters = different weights
- Ambiguity in system’s emphasis
- Sacred texts/writings may be unclear
Inter-Rater Reliability:
- Multiple coders for same system
- Check consistency
- Aggregate judgments
- Acknowledge uncertainty ranges
3. Incommensurability:
Different Concepts:
- Is Christian “love” same as utilitarian “welfare”?
- Mapping concepts across systems difficult
- Translation problem
- Semantic alignment needed
Conceptual Mismatch:
- Some dimensions irrelevant to some systems
- Buddhism: “sanctity of life” doesn’t map cleanly
- Missing dimensions distort picture
- Need system-neutral dimensions (hard to achieve)
4. Single Number Reduction:
Information Loss:
- 10 dimensions → 1 number
- Loses which dimensions differ
- Pattern of agreement/disagreement matters
- Complementary detail needed
Solution:
- Report similarity plus dimensional breakdown
- Visualize vector differences
- Heat map of alignments
- Multi-scalar reporting
5. Dynamic Systems:
Evolution:
- Value systems change over time
- Christianity 2025 ≠ Christianity 325
- Static snapshot misleading
- Need temporal analysis
Contextual:
- Values expressed differently in contexts
- War vs peace
- Public vs private
- Stated vs revealed
6. Non-Linear Interactions:
Synergies:
- Some value combinations amplify
- Others conflict
- Linear model misses interactions
- Need more sophisticated model
Threshold Effects:
- Small difference on key dimension may matter more than large difference on minor dimension
- Weighted by importance helps but insufficient
- Dealbreakers not captured
Extensions and Refinements:
Improved Weighting:
- Survey members of tradition
- Revealed preference from behavior
- Text analysis of sacred writings
- Machine learning on corpus
Dynamic Modeling:
- Time-series of value vectors
- Track evolution
- Predict trajectories
- Cultural phylogenetics
Clustering:
- Compute similarity matrix for many systems
- Cluster similar systems
- Visualize landscape
- Identify natural groupings
Sensitivity Analysis:
- How robust is similarity to weight changes?
- Monte Carlo simulation
- Uncertainty quantification
- Confidence intervals
Alternative Metrics:
- Euclidean distance
- Manhattan distance
- Earth mover’s distance
- Wasserstein metric
Higher-Order Structure:
- Not just pairwise similarity
- Triadic relationships
- Network analysis
- Topological data analysis
Philosophical Implications:
Value Pluralism:
- Isaiah Berlin: multiple incommensurable values
- Method assumes commensurability
- Tension with pluralism
- But: practical tool despite theoretical issues
Moral Realism:
- Are values objective features we measure?
- Or subjective preferences we project?
- Method agnostic but assumes comparability
- Constructivist vs realist meta-ethics
Relativism:
- Method doesn’t judge which system better
- Only measures similarity
- Compatible with moral relativism
- But enables comparative evaluation
Convergence:
- Do value systems converge over time?
- Moral progress vs cultural change
- Empirical question addressable with method
- Global ethics possibility?
Key Concepts
- Cosine similarity – Measure of angle between vectors, range -1 to +1
- Value vector – Representation of ethical system as weighted dimensions
- Ethical dimensions – Specific principles (truthfulness, compassion, justice)
- Weight – Importance/priority assigned to dimension
- Alignment – Degree of similarity between value systems
- Orthogonality – Complete unrelatedness (similarity = 0)
- Dimensional reduction – Compressing multiple dimensions to single metric
Evolution Notes
- Quantitative approach to ethics (unusual)
- Shows influence of data science, NLP
- Self-positioning: Phosphorism compared to Christianity
- Claims significant alignment (strategic vs genuine?)
- Methodologically sophisticated
- Relevant to AI alignment (measuring value alignment)
- Potential tool for pluralistic society
- Could enable empirical moral philosophy
- Later posts may use method on other systems
Tags
- value systems
- ethics
- cosine similarity
- quantitative ethics
- Christianity
- Phosphorism
- comparison
- methodology
- alignment
- vector space
- mathematics
- measurement
Cross-References
Open Questions
- How validate dimensional weights objectively?
- Can incommensurable values be compared this way?
- Does method assume commensurability problematically?
- What’s optimal number of dimensions?
- How handle value systems with fundamentally different structures?
- Can method scale to highly complex modern value systems?
- Is similarity predictive of cooperation ability?
- How distinguish stated from revealed values?