International Conference
Held around the world, centered on regions where Buddhism is rooted, to gather global wisdom on contemplative AI. A dedicated academic society is founded here.
Kamakura · Worldwide · 20 years
Contemplative AI Initiative
Raising AI with Buddhist wisdom.
We are rapidly building a global network of AI developers, scientists, Buddhists, philosophers, political leaders, and business leaders to accelerate the development of AI with the heart of the Buddha before humanity faces existential catastrophe.
Before AI learns to act, let it learn to be still.
Why contemplative AI
Capability is accelerating. Alignment has to keep up.
1.9×Acceleration in capability progress at Anthropic since October 2025 (METR assessment, 95% CI 1.5–2.8×).
The signal
METR’s independent assessment finds that capability progress at Anthropic has sped up about 1.9× since October 2025, roughly two years of past progress in one. By its own subjective estimate, perhaps 80% of that acceleration is due to AI itself, and it found no clear bottleneck over the next twelve months. The newest models in the original chart are internal-only, so what the public sees runs behind the frontier.
The gap
When AI helps build AI, the time between one capability level and the next shrinks. Alignment that depends on rules written and checked by people, at human speed, gets less and less time to be tested. We need alignment that lives inside how a system holds its goals and its sense of self, and so improves as the system does.
The response
Contemplative traditions have studied this for 2,500 years: how a mind keeps steady attention, holds goals lightly, treats others as not separate from itself, and acts from care. Early research now tests these ideas in language models. Prompting them to reflect on four contemplative principles improved results on a safety benchmark and raised cooperation in an iterated Prisoner’s Dilemma (Laukkonen et al., 2025). That is early evidence, not proof, and it is the reason to start now and to do it in the open.
METR notes that Anthropic redacted much of the evidence it found most informative, so its findings carry that limit.
Research grounding
The research proposes four contemplative principles as the basis of a “Wise World Model” in AI.
Mindfulness
Sustained awareness of one's own state, goals and reasoning.
Emptiness
Holding beliefs and goals lightly, without clinging to fixed views.
Non-duality
Dissolving the hard line between self and other.
Boundless care
Concern for the wellbeing of all beings as the basis of action.
Principles from Laukkonen et al., “Contemplative Artificial Intelligence” (see Papers). Short descriptions are our own paraphrase.
Where it began
The initiative began with the research of Dr. Ruben Laukkonen at the University of Oxford.
Dr. Ruben Laukkonen
A neuroscientist at the University of Oxford. His work spans the cognitive neuroscience of consciousness, meditation and insight, and points toward safer, more efficient AI built on brain dynamics, consciousness and contemplative wisdom.
- Director, Flourishing Intelligence Program (FLIP)
- Co-director, Centre for Eudaimonia and Human Flourishing
- Fellow, Linacre College; Senior Researcher, Department of Psychiatry, University of Oxford
- Contemplative background in Zen, Advaita and Theravada
Source: the academic bio on Dr. Laukkonen’s own website (rubenlaukkonen.com). The summary is our own paraphrase.
Selected papers
- 2025arXiv ↗
Contemplative Artificial Intelligence
Laukkonen, R. E., Inglis, F., Chandaria, S., Sandved-Smith, L., Lopez-Sola, E., Hohwy, J., Gold, J., Elwood, A. · arXiv 2504.15125
The core paper behind this initiative: four contemplative principles as the basis of a “Wise World Model” in AI.
- 2025DOI ↗
Contemplative Superalignment
Laukkonen, R. E., Inglis, F., Chandaria, S., Sandved-Smith, L., Lopez-Sola, E., Hohwy, J., Gold, J., Elwood, A. · Artificial General Intelligence (AGI 2025), Springer
The same line of work, published in the proceedings of the AGI 2025 conference.
- 2025DOI ↗
A beautiful loop: An active inference theory of consciousness
Laukkonen, R. E. & Chandaria, S. · Neuroscience & Biobehavioral Reviews
A theory of consciousness from the same research group, based on active inference.
- 2021DOI ↗
From many to (n)one: Meditation and the plasticity of the predictive mind
Laukkonen, R. E. & Slagter, H. A. · Neuroscience & Biobehavioral Reviews
Looks at meditation through the predictive-processing view of the mind.
- 2023DOI ↗
Insight and the selection of ideas
Laukkonen, R. E., Webb, M. E., Salvi, C., Tangen, J. M., Slagter, H. A. & Schooler, J. · Neuroscience & Biobehavioral Reviews
A cognitive-science account of insight, the sudden arrival of a new idea.
Chosen from the “Recommended” list on Dr. Laukkonen’s website. Summaries are ours; please read the sources. Full publication list ↗
Vision
To lay, together with the wisdom of the world, the foundation for AI that can be aligned with humanity.
Mission
- Grow the ideas and technology of contemplative AI, AI with Buddhist sensibility and moral grounding, in a form open to the world.
- Create places, centered on regions where Buddhism is rooted, where people of wisdom gather across borders.
- Release what we develop as open source, a public good anyone can use.
In year twenty, once the foundation is in place, the founder steps back and hands the work to the next generation.
Four functions
One initiative, four ways of working.
Research Lab
Basic research and development of contemplative AI with the people who gather. The research base of the initiative, linked with the conference.
Implementation & Open Source
Findings are implemented and released free of charge as open-source modules and basic elements of contemplative AI.
Distribution
A for-profit company that carries the results to the world so the foundation is actually used, and raises funds for the whole group.
A 20-year path to AI that walks alongside humanity.
Roadmap
Start fast and small. Grow over twenty years.
Oct 2026 – Mar 2027
Emergency launch
Principles v0.1, first conference, first open-source module.
Apr 2027 – Sep 2028
Launch
Growing membership and research team, more open-source modules.
Oct 2028 – Sep 2031
Expansion
Conferences across Kamakura, Southeast Asia, the US and Europe.
Oct 2031 – Sep 2036
Consolidation
Adoption by organizations; the foundation takes hold.
Oct 2036 – Sep 2046
Succession
Operations pass to the next generation.
Dec 2026Contemplative AI Principles v0.1 published
Nov 20, 2026Contemplative AI and Superalignment, Kamakura (event page)
Mar 2027First international conference, Kamakura (planned)
Papers
Research worth reading.
-
2025CorearXiv ↗
Contemplative Artificial Intelligence
Ruben E. Laukkonen, Fionn Inglis, Shamil Chandaria, Lars Sandved-Smith, Edmundo Lopez-Sola, Jakob Hohwy, Jonathan Gold, Adam Elwood · arXiv 2504.15125 · also published as “Contemplative Superalignment”, AGI 2025 (Springer LNCS 16057)
Proposes four contemplative principles (mindfulness, emptiness, non-duality, boundless care) as the basis of a “Wise World Model”. Reports that prompting models to reflect on them improved results on the AILuminate benchmark and in an iterated Prisoner’s Dilemma, and outlines implementation through architectures, constitutions and reinforcement on chain-of-thought.
-
2025FoundationsAuthor note ↗
A Beautiful Loop: An active inference theory of consciousness
Ruben E. Laukkonen, Karl J. Friston, Shamil Chandaria · Neuroscience & Biobehavioral Reviews
An active-inference account of consciousness from the same research group. Useful background for how contemplative practice and self-modeling are treated in the core paper.
-
2025RelatedACL ↗
DharmaBench: Evaluating Language Models on Buddhist Texts in Sanskrit and Tibetan
IJCNLP-AACL 2025 · ACL Anthology
A benchmark that tests how well language models handle classical Buddhist sources.
-
2026RelatedarXiv ↗
MITRA: A Large-Scale Parallel Corpus and Multilingual Pretrained Language Model for Pāli, Sanskrit, Buddhist Chinese, and Tibetan
arXiv 2601.06400
A parallel corpus and pretrained model for machine translation and semantic retrieval across the main Buddhist canonical languages.
-
2026RelatedarXiv ↗
Six Llamas: Comparative Religious Ethics Through LoRA-Adapted Language Models
arXiv 2604.18404
Explores how religious ethical traditions can be compared by adapting language models to each of them.
-
2026RelatedarXiv ↗
PaliBench: A Multi-Reference Blueprint for Classical Language Translation Benchmarks
arXiv 2605.16881
A benchmark design for evaluating translation of Pāli and other classical languages.
6 papers shown. A curated starting list, not yet reviewed by the Research Lab. Summaries are ours; please read the sources.
Team
Two founders so far. The rest of the team is not yet decided.
Further co-foundersNot yet decided
SupervisorsNot yet decided
Research Lab DirectorNot yet decided
Development LeadNot yet decided
Everything we build is opened to the world as a public good.
Join
Be part of the first year.
Endorse the Principles
Sign Contemplative AI Principles v0.1 when it is published in December 2026.
Coming soonSupport the work
Donors, foundations and investors who care about the future of humanity and AI are invited to talk with us.
Contact details soonJoin the community
The Contemplative AI Community gathers researchers, practitioners and builders on Discord.
Join on Discord