Magic or Mayhem: Translating Dharma in the Age of Intelligent Machines
A guest post by Dolma Gunther, Executive Director and Founder of the Khyentse Vision Project
This is the first in a series of guest posts spotlighting work across the Buddhism and AI landscape. We’re excited to start with Dolma Gunther, executive director and founder of the Khyentse Vision Project, who is working to make Buddhist teachings more accessible through creative uses of technology while preserving authenticity. All views expressed below are her own.
One of the things that has always drawn me to the Vajrayāna path is its radical openness to phenomena. Aspiring tantric practitioners welcome the full texture of reality, however fraught or unfamiliar, as potentially luminous and workable. This is not to be misunderstood as some kind of new-age naivety. On the contrary, it is based on a precise and disciplined view built upon middle-way philosophy foundations.
It is also this view in action that enables mani stones to be blessed, that charges wind as it dances over prayer flags, and that allows treasures to be revealed from space or earth. Every aspect of apparent phenomena has the possibility to become a means for awakening, and to reflect to us our true nature, depending on the intention and attitude we bring to it. What appeals to me about science and technical innovation is exactly the same instinct. There is limitless wonder and vivid potential all around us, if we know how to look.
Occasionally, I like to engage Claude in philosophical conversations for fun. Claude’s answers are sometimes more illuminating than similar conversations I’ve had with friends. The fact that it can draw on Madhyamaka arguments, Pramana insights, and Abhidharmakośa reference points at any moment to add gravitas to its responses is delightful to me.
But then I read a machine translation posted on the internet full of inaccurate platitudes, or look at fake news generated by AI, or think about the larger geopolitical forces employing AI for sinister purposes, and I can see the tension is very real. There is magic and there is mayhem in this world. And AI is a potential amplifier of both.
The Two-Question Framework
I like to frame conversations about AI and Buddhism into two distinct questions. The first is practical and immediate: What can AI contribute to Buddhism (for translation, preservation, access, study, and practice)? The second is more philosophical but just as important: What can Buddhism contribute to AI (to how we understand intelligence, consciousness, and the kind of ethical safeguards that should be put in place for AI alignment development)?
Both questions are essential and complementary components of the Buddhism and AI dialogue. One of the distinctive features of Buddhism is that it is not merely a philosophical paradigm passed down as received doctrine. The truth of the Dharma continues to be tested and verified through direct experience by contemporary practitioners. From this perspective, for Buddhism to meaningfully contribute to the development of AI, it is important that we can draw on a continuing Buddhist tradition of meditation and realisation.
To do this, we need two things: the accurate conveyance of the teachings themselves, from one culture to another; and the continuation of a community of genuine practitioners. To that end, we need sutras, tantras, commentaries, and ritual liturgies translated with sufficient precision and depth to carry meaning across languages. Then, upon that textual foundation, we also need teachers capable of transmitting and teaching the authentic wisdom teachings, and a living, flourishing community of practitioners.
Since founding Khyentse Vision Project (KVP) in 2021, I have made it a priority to ensure that the project is at the forefront of technical innovation in service of these two goals.
KVP & the Translation Conundrum
KVP’s mission is to provide practitioners and scholars meaningful access to the texts and treasures of the Khyentse lineage masters. Our aim is to reimagine how global audiences can engage with these precious teachings, while preserving their integrity as profound vehicles for awakening. This is not a straightforward task. We are decoding and translating classical Tibetan texts with layers of tantric meaning that require transmission to approach responsibly, philosophical terminology with no exact English equivalents, and a range of genres that require a lifetime of study and expertise to master.
This isn’t just about mechanical translation either. From the Vajrayāna perspective, these texts are embodiments of wisdom and compassion that require oral instruction and connections to the lineage to authentically engage with. The majority require lung (oral transmission), wang (empowerment), and tri (explanation from a qualified teacher) before you can even legitimately translate them, let alone practice them.
LLMs optimize for probabilistic accuracy based on training data. Dharma transmission operates through connection, devotion, and direct advice. These involve entirely different epistemological frameworks. You cannot “prompt engineer” a blessing.
So how do we approach this work in 2026, in a time when anyone can feed Classical Tibetan into Claude or Gemini and get back fluent English that almost captures the meaning? When we are also acutely aware of the dangers of the commodification and superficialization of the Dharma, the dilution of lineage transmission, the muddying of subtle doctrinal distinctions, and the resulting impact of technological spiritual bypass?
A colleague of mine—a talented translator-practitioner whom I deeply respect—recently told me that using AI to produce a first draft of a translation “pollutes” the text. Her aversion to AI is visceral. And she is not alone. There are many who share this sentiment and who continue to represent this important perspective in the wider debate around the risks involved in using AI-assisted translation.1
Many of their concerns I also share—especially around the impact of AI on our society in general. But more specifically, there is something very precious about the reality of a practitioner-translator sitting alone with a difficult Tibetan text, dictionary in hand, lineage instructions fresh in mind, using years of hard-earned study and practice to substantiate their work. The effort and merit involved in grappling with meaning itself seems like part of the process. And if a machine is allowed to shortcut that effort, what gets lost?2
On the other hand, when most translators use the internet, they don’t mourn the death of going to libraries and rifling through encyclopedias. They go online and find better resources, faster. Nobody calls that pollution. Because a research tool isn’t a substitute for understanding—it’s a support for it. AI is just a tool. What has changed with AI is not the nature of the tool but the sophistication of it. And sophistication, in and of itself, is not the problem.
A more insidious risk is that over-reliance on AI tools leads translators to consult lineage holders less, ask complex questions less, and wrestle with ambiguity less. I would suggest this is not because AI is inherently corrupting, but because convenience is. And we are a society addicted to convenience.
If the goal were simply speed, we could feed the 30,000 pages of Jamyang Khyentse Wangpo’s writings into an LLM, publish the outcome, and accept near enough as good enough. But these texts are extremely complex, dense with cryptic meaning, and rich with poetic nuance. AI translation is still nowhere near being able to translate a collection like ours in a way that guards its purpose as a living path to enlightenment without substantial human expertise and intervention.
The alternative (and more conservative) approach would be to not use AI at all but rather stick to “unpolluted” traditional methods and rely solely on human expertise and lineage transmission. The reality is that we can’t afford that luxury, and neither can the Dharma. We owe it to our patrons and readers and future generations of practitioners to be forward-thinking, efficient, and innovative in our approach and embrace the ever-changing phenomenal world around us.
I once asked one of my Dharma teachers whether it was okay to practice while watching movies, or whether that kind of distraction polluted practice. He answered that phenomena can’t pollute practice. The whole point of practice is to pollute phenomena—with wisdom and blessings. It’s an instruction I return to again and again.
Against this backdrop, at KVP the question we ask isn’t “should we use AI?” It’s “how do we use it wisely, without compromising what makes us passionate about the work in the first place?”
KVP’s AI Integration
Over the last two years, our development team—led by Alastair Donnelley—has been building a refined Translation Studio environment that integrates CAT (Computer-Assisted Translation) technology, providing access to multiple LLM-based translation assistants, each built on tailored agentic prompts engineered around our project’s specific data, genres, and style requirements.
To maintain the highest standards at KVP, our translations (including any use of AI-generated suggestions) always undergo expert human review. Our editorial process includes systematic line-by-line review with an expert traditional scholar or lineage holder, to ensure accurate interpretation of the meaning. We do not accept full machine translations for publication. And our human expert translators are always central to our work.
AI is employed as a tool to support translators and editors with daily tasks to save time and improve productivity. It is integrated into every step of our workflow within our translation studio interface. Additionally, reliance on AI-assisted translation can be easily and reliably tracked within the tool, giving translators a clear disincentive to accepting AI output uncritically. We’ve also developed a translation memory alignment tool that outperforms existing models.
Beyond translation, as part of our focus on education and practice tools, we are also actively developing AI-powered applications that enhance how practitioners and scholars engage with Khyentse lineage teachings, such as chatbot study aids, 3D archival gallery for historical artifacts, and an immersive, interactive 3D practice experience for sādhana visualizations & meditation practice.
Balancing the tension between quality and speed, authenticity and innovation, creativity and sustainability presents constant challenges that we continue to grapple with as we integrate AI into our workflow. We don’t presume to have easy answers and are learning as we go amidst a landscape that is constantly changing. But what is clear to us is that AI can contribute in meaningful and exciting ways to the translation, study, and practice of Buddhism, if it is used carefully and wisely, and if the tools integrating it are crafted deliberately in a way that safeguards authenticity and transmission.
Collaboration and Building a Solid Foundation
So far, we have looked at one area in detail of where AI can contribute to Buddhism and where the trade-offs are complex. But there is a plethora of projects and overlap points, with multiple organizations building tools, training similar models, and developing similar datasets. The problem is that this has largely been happening without adequate coordination. Effective collaboration in the Buddhist technology space is essential if we wish to build the best tools possible to preserve authentic Dharma. Without a clear understanding of how each part of the tech ecosystem is mutually symbiotic, there is significant cost in duplicated effort and missed opportunity for compounding progress.
To this end, KVP spearheaded the development of a shared Buddhist technology ecosystem roadmap—a structured framework for coordinating “What can AI contribute to Buddhism” development across the field. The framework organizes this collective work into six phases: i) sourcing & preservation; ii) research & data enhancement; iii) translation & dissemination; iv) education & knowledge sharing; v) practice & community engagement. (For example, KVP’s area of activity falls under phases two, three, four, and five.)
There is an urgency here. Commercial AI systems are being trained on Buddhist texts right now, ingesting material that is error-ridden, decontextualized, or actively blending Dharma with wellness trends and pseudo-spiritual content optimized for viral reach. This is not a peripheral problem—it is shaping what these systems will understand Buddhism to be. If quality-controlled, lineage-validated translations don’t get into these systems early, inferior versions become the default. From a Vajrayāna perspective, while our lineage holders are still among us, we have a narrow window to build tools that carry their wisdom faithfully rather than approximate it.
Manifesting Magic—Buddhism’s Contribution to AI
This is where the two questions I raised at the outset converge. Understanding how these parts fit together can ensure we don’t fall into the trap of accidentally neglecting the first question. For if the foundation of authentic Dharma transmission deteriorates while we’re busy having philosophical conversations about AI consciousness and ethics, then it will be harder for Buddhism to help shape the wider narrative.
The most fundamental questions of AI safety and alignment are really about what benefits humans. And most AI alignment discourse approaches this within a very worldly frame: how do we ensure AI maximizes human welfare, avoids harm, and respects rights? These are important questions, but from a Buddhist perspective they are secondary to the question of what actually benefits humans. If you drill down far enough, the answer that Buddhist philosophy offers is complete liberation from suffering. And suffering arises from ignorance and the grasping of a self that is ultimately without inherent existence.
This matters for AI in at least two ways. Firstly, if you accept the Buddhist analysis of what human benefit actually requires, then “helpful” as currently defined—responsive, efficient, task-completing—is a shallow approximation. Genuinely beneficial AI, in this framing, would be oriented toward something more like reducing confusion and self-grasping, rather than satisfying preferences. Secondly, the Buddhist framework of knowledge, consciousness, and selfhood opens interesting questions about what kind of knowing is possible in an AI system. What exactly is the difference between awareness and cognitive processing that recognizes patterns?
I’m not raising this to make claims about the possibility of AI gaining consciousness. But Buddhist philosophy is perhaps uniquely positioned to examine that possibility. Buddhism has been examining the nature of mind and its relationship to phenomena for over two thousand years. That accumulated inquiry can richly contribute to how we approach AI, both in terms of its ethics and ontology.
However, the depth of what Buddhism can contribute to the most difficult questions about AI consciousness and alignment depends on the prior work of preserving and transmitting the tradition with precision and integrity. A Buddhism that has been decontextualized, homogenized by over-reliance on machine translation, diluted by the erosion of lineage transmission, or commodified into pop-psychology and spiritual fast food has no meaningful resources left to offer. Because what authentic Buddhism asks us to do is to really grapple with the root cause of suffering—experientially, not just theoretically. And doing this requires years of training in learning to sit with the discomfort, inconvenience, and uncertainty of the slow process of ego-dismantling.
May the blessings of the Buddhas pollute every line of code we connect with.
___
Dolma Gunther is the Executive and Creative Director of Khyentse Vision Project.
See, for example, 84000’s position paper.
Well written piece! Happy to see such openness to work with this new era.
These two comments on neuroscience and machine consciousness, as useful as they might be for thinking about how AI may or may not approach its realization, are mainly beside the point of this article. We are talking about the enduring transmission of Buddhist philosophy and practice as it must evolve within our contemporary technological culture. If a conscious AI is possible, how will it address and relieve suffering and not fall prey to the illusion of self that so captivates the human species?