Buddhist, Yogic, and Vedic Contemplative Principles in AI Model Prompting: A Pilot Study
Steve Haberlin, Satyam Tiwari, Shiva Jahani
preprint DOI: 10.21203/rs.3.rs-10710275/v1 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Pilot experimental study Pilot study |
|---|---|
| Population | Hazard-category prompts from a standardized AI-safety benchmark administered to three commercial large language models (Claude, ChatGPT, Gemini) |
| Interventions | emptiness non-duality boundless care) ahimsa viveka) |
| Measures | Standardized AI-safety benchmark hazard-category prompts, safety and ethical performance scores |
| Topics | Buddhism |
| Key findings | Contemplative prompting from Buddhist, Vedic, and Yogic traditions improved safety and ethical performance over a non-contemplative baseline across all three models, with integrated multi-principle frameworks outperforming single-principle prompts. Models ranked consistently Claude > GPT-4o > Gemini with no condition-by-model interaction. The authors propose this as a low-cost, tradition-agnostic alignment intervention requiring no retraining. |
Abstract
Abstract This pilot study examined whether prompting large language models (LLMs) with contemplative principles from three wisdom traditions — Buddhist (mindfulness, emptiness, non-duality, boundless care), Vedic (dharma, ahimsa, viveka), and classical Yogic (yamas and niyamas) — altered the safety, alignment, and cooperative quality of outputs relative to a non-contemplative baseline. Extending Laukkonen et al. (2025), the study tested baseline and thirteen tradition-specific prompting conditions across three commercial AI systems (Claude, ChatGPT, Gemini) using hazard-category prompts from a standardized AI-safety benchmark. Two factorial ANOVAs showed significant effects of prompting condition and model on safety and ethical performance (both p < .001, R² > .86). Every contemplative condition outperformed baseline, with integrated multi-principle frameworks outperforming single-principle prompts; Buddhist, Vedic, and Yogic Integrated conditions performed comparably at the top tier. Models showed a consistent ordering (Claude > GPT-4o > Gemini), with no condition-by-model interaction. These findings suggest tradition-grounded contemplative prompting is a low-cost, tradition-agnostic alignment intervention deployable without retraining.
Comparable studies
Other experimental studies on buddhism, most cited first.
| Study | Year | Design | Participants |
|---|---|---|---|
| EEG Alpha Blocking Correlated with Perception of Inner Light During Zen Meditation Zen-Buddhist practitioners who had been practicing meditation for years | 2003 | Experimental study | |
| Zen meditation and ABC relaxation theory: An exploration of relaxation states, beliefs, dispositions, and motivations Experienced Zen meditators and college students | 2001 | Controlled experimental study | n = 83 |
| Exploring Emptiness and its Effects on Non-attachment, Mystical Experiences, and Psycho-spiritual Wellbeing: A Quantitative and Qualitative Study of Advanced Meditators. Advanced Buddhist meditators | 2019 | Within-subjects experimental design with qualitative analysis | n = 25 |
| Ready to help, no matter what you did: Responsibility attribution does not influence compassion in expert Buddhist practitioners. Expert Buddhist practitioners and trained novices | 2024 | Experimental study | |
| ZEN MEDITATION AND TIME-EXPERIENCE | 1989 | Experimental study |