My paper on cognitive MRI for AI conversations has been accepted to Complex Networks 2025 in New York.
Presentation scheduled for December.
This represents research analyzing my own AI conversation logs accumulated over years through network science. Here’s what we found.
The Core Idea
AI conversations appear as linear text logs. But buried in that linearity is topological structure—a network of conceptual relationships that reveals how knowledge exploration actually happens.
We introduce a cognitive MRI for AI conversations: network analysis that makes thought topology visible.
The Method
- Start with conversation logs (my own AI interactions over months)
- Construct semantic similarity networks using user-weighted embeddings
- Identify knowledge communities through network clustering
- Find bridge conversations that enable cross-domain flow
The key innovation: user-weighted embeddings. Not just semantic similarity, but similarity weighted by what matters to the user’s knowledge exploration.
Key Finding: Heterogeneous Topology
Different domains have different network structures:
Theoretical domains (math, CS theory, philosophy):
- Hub-and-spoke structure
- Central concepts with many connections
- Radiating explorations from core ideas
Practical domains (coding, tools, applied work):
- Tree-like hierarchies
- Branching problem-solving paths
- Directed flow from problem to solution
This wasn’t expected. The topology reflects how knowledge is organized in each domain.
The Three Bridge Types
We identified three distinct types of bridge conversations that facilitate knowledge integration across communities:
- Concept bridges: Link related theoretical ideas across domains
- Application bridges: Connect theory to practice
- Meta bridges: Enable reasoning about the exploration process itself
These bridges are where cross-domain insight happens.
Why This Matters
Linear conversation logs hide structure. The network reveals:
- Knowledge communities: What conceptual clusters have formed
- Bridge points: Where different domains connect
- Exploration patterns: Hub-and-spoke vs. tree-like reasoning
- Integration opportunities: Where communities could connect but don’t yet
It’s a map of how understanding develops through conversation.
Personal Context
This research came from analyzing my own AI interactions:
- Hundreds of conversations over years
- Topics spanning math, CS, philosophy, cancer research, code
- Real knowledge exploration, not synthetic data
The network structure emerged from actual thinking, not benchmarks.
Connection to My Research
This connects several threads:
- Complex networks: Network science applied to cognition
- AI understanding: How conversation enables knowledge building
- Knowledge representation: Semantic embeddings as cognitive maps
- Meta-learning: Understanding how you learn through AI
It’s network science for personal knowledge exploration.
The PhD Context
This is PhD-level work:
- Novel methodology
- Clear empirical results
- Accepted at peer-reviewed conference
- Opens new research directions
Whether I complete the PhD or not, this work exists and contributes.
Stage 4 Context
Doing this research while managing stage 4 cancer has interesting implications:
The work documents my own knowledge exploration during a compressed timeline. The networks map real intellectual work under real constraints.
It’s fitting: network analysis of thought, while time runs out.
Complex Networks 2025
New York. December 2024.
I’ll present this work alongside researchers studying:
- Social dynamics
- Biological networks
- Infrastructure systems
- Economic flows
Showing them: AI conversation has network structure we can analyze.
And that structure reveals how knowledge actually develops.
Revealing thought topology hidden in linear conversation logs. Complex Networks 2025, New York, December.
Discussion