Empirical Measurement of AI Cognitive Architectures
Applying task-based and resting-state fMRI methods to AI metacognition research
I develop empirical frameworks to measure cognitive architectures in AI systems using methods validated through 60+ task-based and resting-state fMRI sessions. My approach uniquely combines:
TIDE-analysis: Quantitative architecture mapping (like task-based fMRI)
TIDE-resonance: Metacognitive perception studies (like resting-state fMRI)
This directly parallels my PhD work using both task-based and resting-state fMRI to study ASD/NT differences.
When AI models describe visualizations and reflect on their own processing, they exhibit patterns analogous to Default Mode Network activation in human brains - providing the first empirical window into AI metacognition.
21 sessions | 630 responses | p < 0.0001
| AI Model | Coherence Score | 95% CI | Architecture Type |
|---|---|---|---|
| Gemini 1.5 Flash | 71.5% | [63.1%, 79.8%] | Most consistent/predictable |
| Claude 3 Haiku | 55.1% | [51.7%, 58.4%] | Moderate variability |
| GPT-3.5 Turbo | 38.3% | [35.9%, 40.7%] | Most varied/creative |
PhD in Cognitive Neuroscience - Developed a 14-feature semantic framework through extensive task-based and resting-state fMRI studies, revealing how neurotypical and autistic individuals use fundamentally different neural architectures.
Task & Resting-State Expertise: My dissertation included multiple studies analyzing both task-based and resting-state differences between ASD and NT processing patterns - directly informing my dual approach to AI cognition research.
This validated framework now enables empirical measurement of AI cognitive architectures through linguistic pattern analysis.
Automated quantitative testing of AI architectures across 14 semantic dimensions. Achieves statistically significant differentiation between models.
Live Dashboard →Metacognitive analysis where AIs describe visualizations then reflect on their processing - analogous to resting-state self-referential tasks.
GitHub Repository →Comprehensive framework analyzing linguistic patterns, information structure, and cognitive signatures in AI outputs.
View Framework →Theoretical foundation mapping Temporal-Internal Dimensional Exploration, validated through neuroscience research.
Interactive Demo →Boundary Interface Dynamics - detecting cognitive state transitions, inspired by fMRI boundary detection methods.
Interactive Tool →12+ specialized tools for cognitive architecture analysis, all open source and actively maintained.
View All Repositories →Statistical significance
Responses analyzed
Validated dimensions
All tools available
Empirical measurement of cognitive architectures provides data for understanding AI behavior patterns and potential risks. The metacognitive analysis reveals how models conceptualize their own processing.
These frameworks bridge neuroscience and AI research. My resting-state fMRI expertise combined with these tools enables novel investigations into AI consciousness and self-awareness.
Understanding architectural diversity can inform the design of more robust, interpretable AI systems. The data reveals fundamental processing differences between models.
Currently collecting data on AI metacognitive patterns across multiple models. Early results show measurable differences in how AI systems conceptualize their own information processing - paralleling neurodiversity in human cognition.
"Measuring what we can, acknowledging what we can't, remaining curious about the rest"