Overview
Thoughtseeds presents a hierarchical and agentic framework for investigating thought dynamics in meditative states. The framework models thoughts as self-organizing agents (thoughtseeds) that compete and cooperate within a cognitive landscape, using Active Inference to formalize the dynamics of attention, awareness, and thought emergence during meditation.
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- Four meditation states (breath_control, mind_wandering, meta_awareness, redirect_breath) emerge from thoughtseed interactions rather than being hard-coded.
- Simulated experts sustain control dominance of focused attention, whereas novices show more frequent, prolonged mind_wandering episodes.
- Learned weights show stronger breath_focus activation during breath_control in experts (0.98) than novices (0.78).
- In the interaction network, experts show strong facilitation from breath focus to self-reflection, absent in novices.
- The authors note the model is not a comprehensive theory of consciousness and needs validation against real neuroimaging or behavioral datasets.
Methods
- Three-level agent-based model of focused-attention Vipassana meditation — Levels are knowledge domains, a Thoughtseed Network of five competing thoughtseeds, and a meta-cognition layer regulating state transitions.
- Novice and expert state-transition matrices drawn from prior empirical research — Reference transition probabilities among four meditation states for experts and novices were derived from published focused-attention research.
- Rule-based hybrid learning of attractor weight matrices — Rule-based constraints optimize thoughtseed weights and interaction patterns, avoiding manual parameter tuning.
- Granger causality extraction of thoughtseed interaction network — Pairwise Granger tests (max lag five timesteps, chi-squared significance) define directional thoughtseed influences, then scaled and thresholded.
- Modified Wilson-Cowan network dynamics with winner-takes-all dominance — Competing thoughtseed populations are simulated with facilitatory/inhibitory connections and top-down meta-awareness; the most active thoughtseed dominates each timestep.
Summary sources: Paper metadata and evidence · Extracted source text.
Document downloads
PDFs are hosted here; supporting sources may use pinned archives.
- 2025_Thoughtseeds.pdf PDF · 5.78 MiB · extracted-text source
Citation
Citation metadata follows the unified bibliography.
Catalog details and resources
Related works
Ranked by recorded methods, topics, contributors and associated software, with contributor roles and domain-only matches labelled. Metadata overlap is not a citation or a scientific equivalence.