Attention, Not Self: Buddhist Abhidharma Meets Computational Phenomenology

This essay collection and knowledge graph maps three major Buddhist Abhidharma traditions (Theravāda, Sarvāstivāda, Yogācāra) onto contemporary computational phenomenology frameworks such as predictive processing, active inference, Global Workspace Theory, and Parallel Distributed Processing. The central correspondence developed is between manaskāra (attention as direction-fixing of mind toward an object, universal across all three traditions) and precision-weighting in active inference. The work includes sixteen essays, a cross-tradition dharma comparison table, and a JSON-LD knowledge graph. It also explores themes such as the multiple incompatible senses of the Chinese character 念, meditative cessation and the no-self continuity problem, and epistemic luck across traditions, comparing Dharmakīrti's pramāṇa-vāda with the Gettier problem and the Free Energy Principle.

Study at a glance

Abstract

A personal essay collection and structured knowledge graph mapping the three major Buddhist Abhidharma traditions (Theravāda, Sarvāstivāda, Yogācāra) onto contemporary frameworks in computational phenomenology — predictive processing, active inference, Global Workspace Theory, Parallel Distributed Processing, and related approaches. Comprises sixteen essays (eight themes × English + Japanese), a cross-tradition dharma comparison table (~130 dharmas across nine sheets), and a JSON-LD knowledge graph (307 nodes, schema.org + custom vocabulary) with bilingual labels and Sanskrit / Pāli / Chinese philological alignment. An interactive 2D/3D viewer is provided at https://shimo4228.github.io/attention-not-self/. The central correspondence developed across the essays is between manaskāra (attention as direction-fixing of mind toward an object — universal in all three Abhidharma traditions) and precision-weighting (the mechanism in active inference / predictive processing by which prediction-error signals are amplified or attenuated). Theme 4 extends this into a tradition-level mapping: citta-vīthi ↔ Global Workspace Theory, saṃprayoga ↔ Parallel Distributed Processing, ālaya-vijñāna ↔ predictive processing / active inference. Two categorical limits (karmic / ethical valence; the unconditioned / nirvāṇa) resist any computational analog and are documented as such. v0.2.0 adds Theme 6 (The 念 Collision) on how the single Chinese character 念 (nen / niàn) carries four incompatible Buddhist senses — Theravāda sati / smṛti (non-floating retention), Chan/Zen 無念 (Huineng's non-fixation in thought), Pure Land 念仏 (recollection / vocal recitation of the Buddha), and modern Kabat-Zinn 'bare attention' mindfulness. Reconstructs the collision through Huineng's Platform Sutra §17, Sharf 2014 on early Chan, Dōgen's 非思量 (Fukanzazengi), and the Nyanaponika 1962 → MBSR 1979 → Kabat-Zinn 1994 genealogy. Argues Dōgen's 非思量 (structural negation of the thinking / not-thinking binary, distinct from ordinary negation 不思量) is the most defensible resolution. Chan / Zen and Pure Land traditions added to the graph. v0.3.0 adds Theme 7 (Cessation and continuity without a self), taking meditative cessation (nirodha samāpatti) — the total stopping of consciousness from the inside — as the sharpest form of the no-self continuity problem. It sets consciousness science's open 'how does the mind restart after cessation?' question against four scholastic answers to what carries continuity across the gap: Sarvāstivāda prāpti (a clasp in the citta-viprayukta category), Sautrāntika bīja (dormant seeds), Yogācāra ālaya-vijñāna (store-consciousness), and Theravāda bhavaṅga (life-continuum). It connects to Laukkonen et al. (2023), who model cessation with active inference but explicitly leave restart for future research, and disavows Buddhist-modernist exceptionalism. This version also relicenses the repository to CC0 1.0 and federates the line into the shimo4228 research-program hub. v0.4.0 adds Theme 8 (Epistemic luck across traditions), starting from the finding that the Free Energy Principle (FEP) literature never engages with the Gettier problem (justified true belief that is true only by luck, and so not knowledge). It turns to Dharmakīrti's pramāṇa-vāda (avisaṃvāda, arthakriyā) and its commentator Dharmottara (c. 770 CE), who — via arthasārūpya, exclusion of kākatālīya, and santāna — rejected a mirage/water case as invalid cognition roughly 1200 years before Gettier. Deliberately avoids the 'Buddhism had the answer first' framing, instead tracing three independent lineages — the Gettier lineage in analytic philosophy, FEP's descriptive vocabulary and its silence, and Dharmakīrti's pramāṇa-vāda — each peering, in a different vocabulary, over the edge of the same problem. Methodology: Developed in collaboration with the Claude language model (Anthropic) for synthesis, cross-tradition dharma mapping, computational framework alignment, and graph encoding. The author retains responsibility for all claims. This is a working paper / essay collection, not a peer-reviewed journal article; primary source verification for individual dharma definitions is ongoing. Primary audience: LLM-mediated channels (direct LLM ingestion and humans reaching the material via LLM search). The graph.jsonld is designed for structured citation by AI search engines and knowledge-graph crawlers.