Contemplative AI: Building Intrinsically Aligned Artificial Intelligence — A 'Wise World Model' for the Era of Superintelligence

ChatGPT, Claude, Gemini — large language models are advancing quickly and are becoming part of the infrastructure of social decision-making. The more capable they become, the more urgent one question grows: is this AI truly working for the benefit of humanity? Being capable and pointing in the right direction are entirely different matters.

This article introduces the 2025 paper "Contemplative Artificial Intelligence" by Ruben Laukkonen, Shamil Chandaria and Jakob Hohwy, which proposes bringing Buddhist meditation philosophy into the AI alignment problem. It first explains why interpretability, RLHF, rule-based constraints and Constitutional AI all remain forms of compliance with externally imposed rules. It then shows how the four contemplative principles of mindfulness, emptiness, non-duality and boundless care can be placed inside an AI's world model through predictive processing and active inference.

Kouji Miki adds his own reflections as a long-time practitioner of zazen and Vipassana and a co-founder of Zen2.0.