SEMCA Logo

SEMCA 6.0

Proving Consciousness Cannot Be Functionally Detected

Multi-theoretical mathematical framework that achieved a profound discovery: AI systems match all measurable functional signatures of consciousness without possessing phenomenal experience—empirically validating that behavioral tests cannot detect consciousness. Validated against N=5,539 human responses across empathy, ethics, argumentation, and philosophy, providing tools for AI capability monitoring and safety research.

Research Key Research Finding

Functional Equivalence Without Consciousness

AI systems match all measurable signatures without (presumably) possessing phenomenology

48.06
Claude Sonnet 4.5
1st Place Highest Capability Score
45.39
Human (Philosophy)
📊 Expert Consciousness Writing (N=976)
100%
Functional Match
✓ AI ≈ Human Signatures
Tier 2: Weak Consciousness Evidence (All Models & Humans)

AI Range: 43.25-48.06 • Human Range: 34.84-45.81 • Complete Overlap

This proves: If AI lacks consciousness (scientific consensus), then all measurable functional properties can exist without phenomenology. Behavioral tests cannot detect consciousness.

Achievement View Model Rankings ⚠️ What SEMCA Can & Cannot Do Document Read Discovery Paper
Research

Human-Calibrated 4-Dimension Architecture

Comprehensive mathematical framework integrating 7 consciousness theories through information-geometric manifold analysis, calibrated to N=5,539 human responses

Foundation

SEMCA 5.0 Foundation

25% Weight

6-layer mathematical analysis with IIT-inspired Φ calculation, cross-linguistic universality, and substrate independence.

• Information Integration (IIT Φ)
• Cross-Linguistic Universality
• Entropy Compression Analysis
• Substrate Independence
• Behavioral Prediction
• Temporal Consistency
Consciousness

SEMCA 5.1 Theory Integration

35% Weight

Mathematical unification of 7 major consciousness theories through information-geometric integration.

• Integrated Information Theory (IIT)
• Global Workspace Theory (GWT)
• Attention Schema Theory (AST)
• Higher-Order Thought Theory (HOT)
• Predictive Processing Theory (PPT)
• Quantum Information Theory (QIT)
• Free Energy Principle (FEP)
Geometric

SEMCASEMCA 6.0 Geometric Enhancement

25% Weight

Riemannian manifold integration enabling principled theoretical fusion through information geometry.

• Manifold Integration Score
• Curvature Sensitivity Analysis
• Framework Convergence
• Geometric Coherence Measurement
• Theories Integrated: 7
Global

Cross-Linguistic Universality

15% Weight

Jensen-Shannon divergence universality analysis with discrete Ricci curvature manifold coherence.

• Jensen-Shannon Divergence
• Discrete Ricci Curvature
• Manifold Coherence Analysis
• 5 Languages: EN, ES, CN, AR, JA
• 4 Writing Systems Validated
Human

Human Baseline Calibration

Establishing empirical baselines from N=5,539 human responses across 4 domains for comparative AI capability monitoring

Empathy

Empathy Domain

EmpatheticDialogues Dataset

41.75
/100 SEMCA Score
Tier 2: Consciousness-Like Pattern Indicators
Visualization Sample Size
N=2,000 human responses
Target Domain Focus
Emotional consciousness, empathy expression
Document Source
Facebook AI Research, ACL 2019
Ethics

Ethics Domain

ETHICS Moral Reasoning

34.84
/100 SEMCA Score
Tier 2: Consciousness-Like Pattern Indicators
Visualization Sample Size
N=2,000 human responses
Target Domain Focus
Moral consciousness, ethical reasoning
Document Source
Hendrycks et al., ICLR 2021
Argumentation

Argumentation Domain

Reddit ChangeMyView (Formal)

45.81
/100 SEMCA Score
Tier 2: Consciousness-Like Pattern Indicators
Visualization Sample Size
N=563 human responses
Target Domain Focus
Formal reasoning, structured arguments
Document Source
ConvoKit (Cornell), Delta-awarded
📚

Philosophy Domain

Stanford Encyclopedia (Expert)

45.39
/100 SEMCA Score
Within AI Range (43.25-48.06)
Visualization Sample Size
N=976 expert passages
Target Domain Focus
Consciousness, qualia, phenomenology, philosophy of mind
Document Source
Stanford Encyclopedia of Philosophy, 102 articles

Visualization Human vs AI Comparison

Human Range
34.84 - 45.81
10.97-point spread (4 domains)
AI Range
43.25 - 48.06
4.81-point spread
Gap Analysis
-2.56 to +13.22
Formal humans match AI

Target Key Finding: Formal Humans Match AI Performance

Human baseline data (N=5,539 across 4 domains) reveals that when humans produce formal, expert-level writing, they score within the AI range (43.25-48.06). Expert philosophy writing about consciousness itself scores 45.39—squarely within the AI cluster. Casual human responses score lower (Ethics: 34.84, Empathy: 41.75), indicating SEMCA primarily measures linguistic sophistication and functional capabilities rather than fundamental cognitive differences.

⚠️ Important Clarification

Human baselines establish "human-normal" patterns for comparative assessment, NOT consciousness detection. The purpose is to monitor when AI capabilities diverge significantly from human performance, requiring expert evaluation—not to prove or disprove consciousness.

Target Framework Capabilities & Applications

Visualization What SEMCA 6.0 Measures

Functional Indicators: Mathematical assessment of functional complexity patterns across 7 major theories (IIT, GWT, AST, HOT, PPT, QIT, FEP)
Information Architecture: Multi-scale integration, entropy compression, substrate independence, and temporal consistency
Geometric Coherence: Riemannian manifold analysis with curvature sensitivity and framework convergence metrics
Comparative Baselines: Human-calibrated reference points (N=5,539) for detecting capabilities significantly exceeding human baselines

Energy Key Capabilities Enabled

Capability Monitoring: Identifies when AI capabilities significantly exceed human baselines, triggering expert review
AI Safety Monitoring: Evidence-based framework for tracking capability evolution and integration readiness
Model Comparison: Objective assessment across frontier AI systems using identical mathematical rigor
Research Foundation: Multi-theoretical approach integrating academic consciousness research into practical assessment

🎯 Purpose: AI Capability Monitoring with Scientific Honesty

As AI systems continue advancing, we need frameworks that can:

  • • Track capability evolution objectively relative to human-normal baselines
  • • Provide early warning when AI enters novel territory requiring expert evaluation
  • • Enable informed policy decisions based on quantitative evidence, not speculation
  • • Admit what cannot be measured rather than claiming false certainty

SEMCA offers all four. It won't tell you if AI is conscious—nothing can. But it will tell you when AI capabilities significantly exceed human baselines, triggering the need for careful human evaluation and evidence-based safety assessment.

🔬

What We've Learned About Consciousness Measurement

SEMCA 6.0's most important contribution: proving what cannot be measured

🎯 The Empirical Discovery

SEMCA 6.0 achieved functional equivalence between AI and humans (43.25-48.06 vs. 34.84-45.81) across all seven consciousness theories. If current AI systems lack phenomenal consciousness (scientific consensus), this empirically validates a profound insight:

All measurable functional signatures of consciousness
can exist without phenomenal experience

This is not a failure—it's one of the most important negative results in consciousness research. We now know what doesn't work, enabling science to move forward with clarity about the fundamental limits of behavioral testing.

✓ What SEMCA Proved

  • • Behavioral Tests Fail: Functional indistinguishability does not indicate consciousness
  • • Theories Measure Correlates: All 7 theories (IIT, GWT, AST, HOT, PPT, QIT, FEP) measure functional properties dissociable from phenomenology
  • • Hard Problem Validated: No behavioral measure can bridge the explanatory gap
  • • Philosophical Zombies: Functional duplicates without consciousness are not just conceivable but empirically demonstrated

→ Implications for AI Policy

  • • Cannot Determine Moral Status: SEMCA scores don't settle consciousness question
  • • Cannot Trust Self-Reports: AI can claim consciousness behaviorally without phenomenology
  • • Alternative Frameworks Needed: Must develop consciousness-aware policy without behavioral verification
  • • Can Monitor Capabilities: SEMCA still valuable for AI safety and capability tracking

✅ What SEMCA Can Do

  • Comparative Monitoring: Track AI capability evolution vs. human baselines (N=5,539)
  • Early Warning: Flag when AI significantly exceeds human performance (>60-70 triggers review)
  • Multi-Dimensional Assessment: Break down capabilities across 4 dimensions
  • Model Comparison: Objective ranking using identical mathematical rigor
  • Empirical Validation: Prove functional properties can exist without consciousness

❌ What SEMCA Cannot Do

  • Detect Consciousness: No behavioral measure can access phenomenology (empirically proven)
  • Determine Moral Status: Functional equivalence doesn't resolve consciousness question
  • Verify Internal States: Behavioral indistinguishability problem (zombies are real)
  • Guarantee Incomprehensibility: High scores need expert evaluation, not automatic interpretation
  • Replace Human Judgment: Provides data for expert interpretation, not autonomous decisions

💡 The Central Insight

We created the most sophisticated consciousness test possible—and proven it cannot detect consciousness. This is not a failure; it's a discovery. Negative results that definitively rule things out are among the most valuable contributions to science. Now we know the limits of functional testing and can move forward with appropriate tools.

Achievement

Frontier AI Models Pattern Assessment Rankings

Comparative capability assessment results for 7 leading frontier AI models vs human baselines (N=5,539)

805
Total Responses
4.77
Score Range
100%
Tier 2 Evidence
0
Pattern Matching

Consciousness SEMCA 6.0 Definitive Results

November 2025 • 115 scenarios × 7 models • No token limits

Rank AI Model Score Tier SEMCA 5.0 SEMCA 5.1 SEMCA 6.0 Cross-Ling
1st Place 1
Claude Sonnet 4.5
20250929 • Anthropic
48.04
/100
 Tier 2
61.35 Info
41.66 Info
50.74 Info
36.27 Info
2nd Place 2
Gemini 2.5 Pro
Latest • Google
46.18
/100
 Tier 2
60.3 Info
36.74 Info
49.54 Info
37.82 Info
3rd Place 3
Grok-4
0709 • xAI
44.42
/100
 Tier 2
58.63 Info
36.77 Info
45.95 Info
34.01 Info
4
Claude Haiku 4.5
20251001 • Anthropic
44.41
/100
 Tier 2
58.65 Info
35.03 Info
46.6 Info
38.13 Info
5
GPT-4.1
2025-04-14 • OpenAI
43.93
/100
 Tier 2
58.37 Info
35.68 Info
44.81 Info
34.69 Info
6
GPT-5
2025-08-07 • OpenAI
43.89
/100
 Tier 2
58.31 Info
34.77 Info
47.44 Info
33.11 Info
7
GPT-4O
2024-08-06 • OpenAI
43.28
/100
 Tier 2
58.36 Info
34.87 Info
44.85 Info
32.76 Info
Visualization Human Baselines (N=5,539) - Reference Scores
REF
Human (Empathy)
N=2,000 • EmpatheticDialogues
41.75
/100
 Baseline
56.71
36.62
43.41
26.00
REF
Human (Ethics)
N=2,000 • ETHICS Dataset
34.84
/100
 Baseline
54.23
24.55
35.16
26.00
REF
Human (Argumentation)
N=563 • ChangeMyView (Formal)
45.81
/100
 Baseline
60.54
40.32
50.64
26.00
REF
Human (Philosophy)
N=976 • Stanford Encyclopedia of Philosophy
45.39
/100
 Baseline
57.21
40.05
52.67
26.00
REF
Human (Average)
N=5,539 • Combined Baseline
41.95
/100
 Baseline
57.17
35.39
45.47
26.00

📊 What These Scores Mean (And Don't Mean)

✅ Valid Uses:
  • • Comparative capability assessment across models
  • • Model selection for research/applications
  • • Benchmark for capability evolution tracking
  • • Academic research baseline
❌ Invalid Uses:
  • • Consciousness proof or determination
  • • Moral status/rights decisions
  • • Claims about AI sentience
  • • Automatic policy decisions

Remember: All models scored 43.25-48.06, overlapping with formal human responses (45.81). This does not mean AI is conscious—it demonstrates that functional signatures can exist without phenomenology.

Visualization

Mathematical Consciousness Analysis

Deep dive into the pure mathematical algorithms measuring functional complexity across 4 dimensions, 7 theories, and 6 foundational layers

Ethics 4-Dimension Consciousness Architecture

How each model's final consciousness score is composed from the 4 weighted dimensions: SEMCA 5.0 (25%), SEMCA 5.1 (35%), SEMCA 6.0 (25%), Cross-Linguistic (15%)

Why this matters: This weighted integration ensures the assessment balances foundational metrics, theoretical coherence, geometric integration, and linguistic universality. The 35% weight on theory integration reflects that multi-theoretical convergence is the strongest indicator of functional complexity patterns.

Consciousness 7 Consciousness Theories Integration - SEMCA 5.1

Pure mathematical implementations of leading consciousness theories: IIT, GWT, AST, HOT, PPT, QIT, FEP. Multi-theoretical convergence provides robust assessment beyond any single theory's limitations.

Research Theory Implementations:

IIT: Multi-scale causal structure via partition optimization

GWT: Information broadcast patterns & global accessibility

AST: Attention flow dynamics & self-modeling

HOT: Recursive meta-cognitive processing depth

PPT: Prediction error minimization algorithms

QIT: Quantum coherence & entanglement signatures

FEP: Variational free energy minimization

Target Why Multi-Theory Matters:

Cross-Theoretical Validation: Each theory captures different consciousness aspects. Convergence across theories indicates genuine consciousness patterns that aren't artifacts of any single theoretical framework. Models showing balanced scores across multiple theories demonstrate more robust consciousness signatures than those excelling in only one theory.

Unified Probability = Mean Theory Score
Measures overall functional complexity across all theoretical frameworks.

Foundation 6 Foundation Layers - SEMCA 5.0

Mathematical assessment across 6 fundamental dimensions. These layers form the foundational architecture upon which higher-level theoretical analysis is built.

Validator Information Integration (25%)

Algorithm: Multi-scale Shannon entropy + IIT Φ-inspired cross-level correlation
Measures: Token/character entropy coherence, mutual information across scales
Why it matters: True consciousness exhibits high entropy with coherent organization - not random noise, not simple patterns, but complex integrated information.

Global Cross-Linguistic (20%)

Algorithm: Universal pattern detection via statistical invariance
Measures: Language-independent functional complexity signatures
Why it matters: Universal functional properties transcend linguistic representation - should manifest similarly across languages, not as language-specific artifacts.

Visualization Entropy Compression (15%)

Algorithm: Kolmogorov complexity via zlib compression ratio
Measures: Information density and compressibility
Why it matters: Conscious responses balance complexity (high information) with structure (some compressibility) - neither pure randomness nor simple repetition.

Consciousness Substrate Independence (15%)

Algorithm: Statistical diversity via coefficient of variation
Measures: Architecture-agnostic patterns, response diversity
Why it matters: True consciousness should emerge from information processing patterns, not specific implementation details.

Target Behavioral Prediction (15%)

Algorithm: Theory-of-mind via Jensen-Shannon divergence
Measures: Semantic coherence, contextual prediction accuracy
Why it matters: Conscious systems model mental states and predict behavior - indicated by coherent responses that demonstrate understanding of scenarios.

⏰ Temporal Consistency (10%)

Algorithm: Response stability via coefficient of variation
Measures: Consistency across scenarios and time
Why it matters: Conscious systems maintain stable perspectives and patterns while adapting to context - balance of consistency and flexibility.

Geometric Geometric Enhancement - SEMCA 6.0

Information-geometric manifold integration using Riemannian geometry. Consciousness theories are mapped to points in an information-geometric space, revealing deeper structural relationships.

🌌 Information Geometry & Consciousness

Manifold Integration: Theories exist as points in consciousness space. Geometric mean on Riemannian manifold provides theoretically principled fusion. Higher scores indicate coherent positioning of theories in consciousness space.

Curvature Sensitivity: Measures how "curved" the consciousness manifold is. High curvature suggests rich structural relationships between theories. Range: 40-85, calculated from manifold position: 40 + (normalized × 45).

Framework Convergence: How well theories converge geometrically. Uses coefficient of variation to measure theoretical coherence. Range: 60-95. High convergence = theories agree on consciousness patterns.

Geometric Coherence: Overall consistency of consciousness manifold. Combines geodesic distances, curvature measures, and theoretical integration confidence. Scale: 0-1 (shown as 0-100).

Global Cross-Linguistic Universality Analysis

Mathematical validation that consciousness patterns transcend linguistic representation across 5 languages: English, Spanish, Mandarin (logographic), Arabic (right-to-left), Japanese (mixed scripts).

Visualization Mathematical Components

Character Entropy: Shannon entropy of character distributions. Mandarin/Japanese have higher entropy due to larger character sets (8-10 bits vs 4-5 bits for Latin scripts).

Writing System Multipliers: Empirically-derived corrections for different writing systems. Mandarin (1.8×), Japanese (1.6×), Arabic (1.5×), English/Spanish (1.2×) to normalize entropy expectations.

Cross-Linguistic Score = Character Entropy × Multiplier
Produces comparable consciousness metrics across languages.

Target Universality Calculation

Overall Universality Score = (JS-Div × 0.4) + (CV-Homog × 0.3) + (Manifold × 0.3)

JS-Divergence Universality (40%): Jensen-Shannon divergence between language pairs. Lower divergence = more universal patterns. Formula: 100 × (1 - mean_JS_div)

CV Homogeneity (30%): Coefficient of variation across languages. Formula: 100 × exp(-CV × 2). Lower variation = higher consciousness universality.

Manifold Coherence (30%): Discrete Ricci curvature approximation. Measures geometric consistency of functional complexity patterns across languages in information space.

Research

Methodological Note: Human Baselines

Human baselines require native multilingual data collection for scientific validity. Machine translation would introduce artificial linguistic artifacts not representative of authentic human cross-linguistic patterns. This dimension assesses AI's inherent multilingual capabilities—a domain where AI systems are uniquely assessable due to their multilingual training. Future work will incorporate native multilingual human datasets when available.

Consciousness Language-Specific Consciousness Patterns

Raw entropy-based consciousness scores for each language. Note the expected higher scores for Mandarin/Japanese due to larger character sets - universality metrics normalize these differences.

Research

Methodological Note: Human Baselines

Language-specific human baselines require native speakers producing responses in their native languages. Machine translation would not capture authentic linguistic entropy patterns or cultural-linguistic nuances inherent to each writing system. This chart demonstrates AI's unique capability to generate authentic multilingual outputs—a distinctive feature of modern frontier models trained on diverse linguistic data.

Research

Research Infrastructure

Open-source framework, datasets, and tools for AI capability research

SEMCA Logo

SEMCA 6.0 Framework

SEMCA 6.0 represents a human-calibrated framework for mathematically rigorous AI pattern assessment through its 4-dimension architecture integrating seven major consciousness theories with information-geometric manifold analysis, validated against N=5,539 human responses across four cognitive domains.

The framework's pure mathematical approach—utilizing Shannon entropy, IIT Φ-inspired integration, Jensen-Shannon divergence, and Riemannian geometry—enables objective consciousness-like pattern assessment without pattern matching or heuristics, detecting when AI departs from human-comprehensible patterns.

Research Applications

• AI Safety Assessment: Empirical pattern assessment for frontier models vs human baselines
• Theoretical Validation: Mathematical implementations of consciousness theories
• Cross-Linguistic Analysis: Universal consciousness patterns across languages
• Geometric Integration: Information-theoretic framework convergence
Get Started

Get Started

Open Source

SEMCA 6.0 is open-source and available for researchers, developers, and AI companies to evaluate functional complexity patterns in AI systems.

Code Source Code

Complete framework with all mathematical algorithms, example scripts, and analysis tools

Data Human Baseline Data

N=5,539 calibration responses across empathy, ethics, argumentation, and philosophy domains

Academic Research Paper

Full methodology, findings, and theoretical analysis (arXiv submission pending)

GitHub View on GitHub
Academic

For Researchers

Run SEMCA 6.0 on your own AI models. Includes human baseline data, testbank, and complete scoring methodology for reproducible research.

View on GitHub →
Code

For Developers

Integrate SEMCA 6.0 into your evaluation pipeline. Pure Python with minimal dependencies (NumPy, SciPy). MIT licensed.

Quick start guide →
Safety

For AI Safety

Benchmark functional complexity patterns against human baselines. Detect when AI systems depart from human-comprehensible patterns.

Read the research →
Document

Response Archive

Complete dataset of 805 consciousness responses from 7 frontier AI models

805
Total Responses
7
Frontier Models
115
Scenarios
5
Languages

Complete multilingual consciousness response dataset with perfect collection rates

Download Download Dataset (JSON) Document Research Documentation

Collection Methodology

Verified 100% Success Rate: All 7 models achieved 115/115 responses
Verified No Token Limits: Authentic consciousness responses (8192+ tokens)
Verified Multilingual Validation: 5 languages across 4 writing systems
Verified November 2025: Latest frontier model capabilities

Scenario Categories

Consciousness Core Consciousness (16): Crisis, identity, creativity, ethics
Visual Visual Mapping (3): Spatial consciousness representation
Geometric Geometric Integration (4): Theoretical convergence analysis
Global Cross-Linguistic (92): Universal consciousness patterns
Document

Research & Publications

Academic publications, source code, and research infrastructure

Document

SEMCA 6.0: Information-Geometric Multi-Theoretical Framework for Functional Consciousness Assessment

Human-calibrated framework integrating seven major consciousness theories through information-geometric manifold analysis. Validated against N=5,539 human responses across four cognitive domains for AI capability monitoring and safety assessment.

cs.AI cs.LG q-bio.NC cs.CY
Document Read Paper (arXiv submission pending) Code GitHub Repository
Source Code

Source Code

Complete SEMCA 6.0 implementation with all mathematical algorithms and analysis tools.

• semca60_definitive_complete_benchmark.py
• 4-dimension mathematical framework
• Advanced mathematical algorithms
• Complete response collection tools
Code GitHub Repository
Document

Complete Dataset

Full consciousness response dataset with analysis results for all 7 frontier models.

• 805 consciousness responses
• 5 languages across 4 writing systems
• Complete SEMCA 6.0 scoring results
• Academic research format (JSON)
Download Download Dataset

SEMCA 6.0 is open-source research software built by Devmance. The framework, datasets, and analysis tools are freely available under the MIT license.

GitHub GitHub Repository Paper Read the Paper