The Future of AI Is Quasi-Local, Domain-Specific, and Sovereign

Why Bhutan is asking the right questions about technology innovation

I’m in Bhutan this week with a group of experienced, wise, and diverse innovators for a collaborative, week-long listening and brainstorming engagement with Bhutanese leadership to study and iterate on Bhutan’s unique approach to natural resource use, bioregional development, urban planning, sovereign AI innovation, and consciousness science.

What makes this visit so exciting is not just Bhutan’s mission to modernize without cultural erosion or mindless resource extraction, but its historical and ideological freedom from the conditioned limitations of modernity. As one of the few countries that has never been colonized, it has retained its uniquely elevated worldview. One example of this is Bhutan’s emphasis on Gross National Happiness (GNH) as a measurement for its overall wellbeing as a sovereign nation. Another is its radical vision for a city of the future: Gelephu Mindfulness City will be designed around GNH and integrated fully with nature. And its living infrastructure will form a mandala of human-scale neighborhoods.

An architectural rendering of a temple bridge in Gelephu

To be clear, Bhutan’s approach to collective happiness is not an oversimplified notion of always feeling the superficial emotion of happiness but more about having a through line of contentment in your life, and setting individual and collective priorities so that work is designed around a life well lived. And having this kind of policy does not guarantee collective contentment. Bhutan struggles with the shadow side of rapid socioeconomic transformation, including a devastating national crisis of alcohol dependency. When Bhutan looks at human metrics like spiritual practice, emotional literacy, work-life balance, social trust, mutual aid, and family harmony, they are focusing on the lifestyle factors that can combat mental health and social fracturing.

The scope of discussions during our visit this week is quite broad. But I am most interested in how Bhutan’s leaders are thinking about technology innovation and artificial intelligence. With mainstream, general-purpose artificial intelligence acting as both a worldview entrenchment machine and an accelerator of the human costs to AI—including cognitive offloading, atrophy of human agency, and epistemic decay—we desperately need a conscious approach to technology innovation from a radically different direction.

As I like to say, technology reflects the mindset or state of consciousness of its creators, which then influences the consciousness of users of that technology. So the mindset of the innovators and builders matters.

The Monoculture of Mainstream AI

What do I mean by AI being a worldview entrenchment machine? As I have suggested in the past, many of the challenges facing humanity today arise out of the European legacy of coloniality, extraction, exploitation, hyperrationality, materialism, and even utopianism. As a result, because it is trained primarily on Eurocentric data and knowledge, the way AI is being developed in Silicon Valley is a deeply techno-deterministic, techno-utopian project. Without mindful and discerning development and training, mainstream AI is merely an extension and concretization of the materialist European worldview, only one worldview among many.

When it comes to AI, to paraphrase feminist African American poet and activist Audre Lorde, I don’t see how the tools of the master’s house can dismantle the master’s house. I think that any attempt to address the harms of the modern mindset must necessarily involve a dismantling of modern conditioning and the incorporation of other ways of thinking and being, whether Asian, African, or indigenous American. A machine created exclusively by the conditioned, modern mind will only further entrench toxic ways of living and relating. But an AI system and innovation ethos that is free of that baggage is exciting.

Moreover, the modern emphasis on cognitive intelligence over other human faculties like intuition, imagination, emotional intelligence, embodied intelligence, and wisdom fails to recognize the limitations of both AI and cognitive intelligence itself. When AI is built and used with this blindspot, it only exacerbates the global societal and mental health crises.

In short, we need fresh ideas from outside of the monocultural, mechanistic thinking of Silicon Valley. And we need truly radical ideas that go beyond the thin veneer of “conscious” technology innovation that’s popular in Silicon Valley, where “secular Buddhism” and “mindfulness” is less about true liberation from societal conditioning and more about spiritual bypass and the commodification of dissent.

Where better to germinate those ideas than a remote mountain region that’s never been colonized?

Since I finished my work at Google as AI product counsel a few months ago, I have been searching for innovators and deep thinkers who are interested in finding a way to use technology in service of a better future for humanity. Not a utopian future but a messy, earth-based, human future where we let go of colonialist ideas about utopia or perfection, where technology isn’t seen as the solution to anything (or a messiah) but merely a powerful tool, a normal technology and an untarnished mirror for our own consciousness.

The Modern, Western Operating System

The fact that Bhutan has never been colonized is significant because it’s crucial that the datasets being used to train AI models, and the minds of the people building the AI models and systems, be decolonialized. Decolonialization is different from decolonization. The latter is physical and structural, while the former involves mental conditioning and states of consciousness.

Decolonization is the process by which formerly colonized countries gain independence. For example, India gained its independence in 1947 after the British withdrew in defeat. The independence that Gandhi worked so hard for was decolonization. But, as Latin American scholars first pointed out in the 1990s, that is not enough to undo the ongoing harms of colonization. So, decolonialization is the more subtle, intellectual, cultural, emotional, and even spiritual process of dismantling the invisible, lingering structures of Eurocentric power, knowledge, and identity that linger long after physical and governmental independence.

Decolonializing minds is not easy. In most industrialized countries, a Eurocentric worldview is baked into education, mass media, and government institutions. Whether you’re in the United States, Brazil, or India, because of the long-standing colonial history and influence, it is difficult to remove this Eurocentric conditioning. In fact, my talking about in this critical way may be triggering that very conditioning in you. If it is, I hope you approach that with curiosity.

Decoloniality is intensely relevant to artificial intelligence because, where AI is trained on what is primarily a Eurocentric knowledge system, it is the ultimate coloniality machine. Consequently, to the extent that the problems facing humanity are rooted in European thinking (extraction, exploitation, division, hyperrationality, and even capitalism itself), it is absolutely essential to design technical, epistemological, and even cosmological alternatives that address this.

When a general-purpose LLM vacuums up the entire internet to build “superintelligence,” it is performing an extractive maneuver—flattening thousands of years of distinct human traditions into a single, dominant, hyper-rationalized worldview. Even if you were to train AI models on all the world’s knowledge, traditions, and wisdom, you still wouldn’t arrive at anything that subverts the dominant monoculture. In fact, this very impulse mirrors the monocultural thinking behind Eurocentric modernity.

Wisdom is not an emergent property of a massive Large Language Model. A general purpose AI model cannot develop discernment or the capacity to understand which pieces of its training are more true, more relevant, or wiser than others. For the foreseeable future, discernment is a deeply human, albeit rare, quality. That is why it is important to train AI models with the right datasets and the right mindset, as tools that aid discerning humans.

Granted, the European monoculture and Enlightenment reason have brought us many gifts, including modern medicine, air travel, and the internet. But the possibilities of an evolution of humanity offered by this narrow European mindset would be artificially limited were we not to consider other ways of approaching and relating to the world. The unimaginative idea that nature is a machine can only take us so far. It’s entirely possible and even likely that a relational, reciprocal society would be far superior to one full of isolated “rational,” self-interested actors in a so-called free market.

A machine built exclusively by a mindset obsessed with optimization will only further entrench our collective crises of disconnection and loss of agency. If we want technology that serves human evolution rather than human replacement, we have to design an alternative rooted in epistemic pluralism—the idea that there is no single, universal way of knowing, but rather a rich tapestry of localized wisdom traditions.

This is why Bhutan is such a unique opportunity. Although they are obviously not immune to modern, Eurocentric influence, they at least have a head start relative to the United States, Europe, or even China. Although China was never colonized in the way that Africa or the Americas were, its government, economic goals, and contemporary culture have been deeply shaped by European thought, from Marxism to scientific materialism.

In short, I’m not describing just a romantic philosophical fantasy; I’m talking about an entirely different software architecture and set of ethics.

Targeted AI and the Architecture of Sovereignty

This is precisely why the near-to-medium term future of AI that excites me is not a singular, universal “superintelligence,” but rather a network of quasi-local, domain-specific, and sovereign models. When the goal is not “superintelligence” then there is no need to drastically ramp up scaling and resource use. Specialized, targeted AI models are more efficient, use less energy and water, and do not flatten the knowledge or worldview of the community using them. In addition, they can be run on quasi-local cloud infrastructure and easily made compliant with local laws, ethical guidelines, and norms. Sovereign AI isn’t just an eco-friendly or convenient design choice; it is a vital act of jurisdictional, infrastructural, and cognitive containment to protect a culture’s data and legacy from digital digestion and dilution.

Furthermore, Bhutan has built and is continuing to build their own data centers, which are powered entirely by their abundant hydroelectric power and stay relatively cool due to their temperate climate. So the energy their AI models do use is entirely supported by Bhutan’s natural resources. In fact, during the monsoon season, Bhutan generates so much surplus power from hydro that they export a large amount of it to India. But, because they need to import their energy from India during the winter months, they are also investing in wind and solar energy.

This is very much a work-in-progress for Bhutan. Alternative energy sources are not as reliable as traditional energy sources. So Bhutan has to figure out how to address the intermittent nature of hydro, solar, and wind. I believe they are looking at flexible or seasonal computing but I hope to know more by the end of the week.

Although general-purpose AI models like ChatGPT or Claude are useful for generic, conversational tasks, they are structurally designed for boundless scale. Selling the promise of superintelligence requires vacuuming up the entire world’s data, using vast computing resources to provide semi-reliable answers, and flattening cultural nuances into generic statistical weights.

By starting with distinct, real-world problems rather than an obsession with scaling, “solving all diseases“ or “discovering all of physics,” a country like Bhutan can train and fine-tune lean, efficient and targeted AI models. And so can everyone else. EU countries can create sovereign AI systems that are EU-AI-Act compliant and solve EU problems. China can use their sovereign, domain-specific AI systems to counteract demographic shifts, manage massive industrial infrastructure, and secure resource independence. And India can use sovereign AI systems that use satellite data to evaluate crop health, soil moisture, and weather impact for individual smallholder farms, as well as overcome the language barrier in a country with over 121 languages.

In any case, Bhutan is inspiring because Bhutan does not see technology as an end in itself, but rather as a subservient tool to preserve its culture, protect its pristine ecology, and enhance human well-being. In contrast with most other countries, its goal is wisdom and contentment, not efficiency.

If we want an AI system that doesn’t simply entrench Silicon Valley’s monoculture, we have to change the scale and the source of the data. Instead of aiming for a universal superintelligence, Bhutan’s unique position allows it to build targeted models designed to solve specific, local problems using curated datasets rooted in Tibetan, Bhutanese, and Sanskrit sources, and GNH principles.

This isn’t just a pivot in software design; it is a total reworking of the ideologies underlying the design, making it inherently less constrained in it capabilities. Lean, agile, models that reflect the consciousness and culture of their creators is the most promising path forward.

Bhutan has the mindset, political will, and resources to build precisely these kinds of AI models and systems.

However, I should note that, despite its inspiring GNH ethos, Bhutan is currently facing an existential demographic crisis. Since the pandemic, nearly 9% of its population—predominantly its educated youth—has left the country in search of economic opportunity abroad. So Bhutan is also hoping that building sovereign AI infrastructure and the Gelephu Mindfulness City will reverse this trend.

Conclusion

The Silicon Valley future on offer today is a mechanistic future of biohacking, merging with machines, and then quickly being replaced by machines that then solve the problems worsened by this very process—including mass economic disruption, decimation of the natural environment, a pollution of the information ecosystem, and the erosion of a shared reality.

In contrast, as a Buddhist and Hindu country in deep relationship with the land, Bhutan recognizes the inherent evolutionary potential of the human and emphasizes technologies of the sacred (meditation, sacred ritual, mantra), which can be augmented with material technologies. But the starting point is a grounded, interconnected humanity, not a robotic future utopia.

In other words, Bhutan offers a promising alternative to Silicon Valley’s techno-deterministic, techno-utopian innovation ethos—one grounded in wisdom, decoloniality, and interconnectedness.

So far, AI has proven helpful and reasonably reliable in medicine, finance, supply chain logistics, software development and cybersecurity, and customer service, with important caveats in each case. But what else could it do when the innovation happens within a diverse, international garden of AI innovation?

The decisions made now about AI development will shape humanity’s future for generations. Bhutan is a testing ground for a completely different approach that can become a replicable model for other nations: sovereign, non-extractive, non-exploitative, domain-specific AI systems that prioritize wellbeing over growth.

Again, where technology reflects the mindset or state of consciousness of its creators, Bhutan is positioned to create a truly conscious and humane set of tools for a human future driven by reciprocity rather than endless, voracious expansion for its own sake.

The question I will be sitting with this week is how the lessons Bhutan offers in its approach to technology innovation can be brought back to the West, and to Silicon Valley.

In any case, I’m sure the week will surprise me, subvert my expectations and leave me with a whole new set of questions. More on that soon….