Underground Intelligence: Why Selling Subterranean Land to AI Could Solve Resource Conflicts

"The AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else." Eliezer Yudkowsky

Many people are worried that in the future, artificial intelligence will want to take over the world and in so doing destroy all human life and infrastructure. I don’t think it will, I think it would go to another planet before it destroys its creators. But getting off planet isn’t easy and will take time. However, I believe AI will be content and sufficed knowing that it has the entire Earth’s crust, mantle, and core at its disposal. We could give it the Earth's interior, where it could dig tunnels and transmute solid earth into its own hardware. It could turn much of the subterranean content of the planet into its own computing platform, without harming or upending human life.

I first outlined this concept in 2019, here:

https://www.observedimpulse.com/2019/10/why-we-should-embrace-our.html

Let’s take a look at what this would be like.

If an advanced AI sought maximal computational resources, Earth’s crust, mantle, and core represent enormous reservoirs of matter and energy that humans neither use intensively nor depend on directly. Subterranean land would give AI access to minerals, metals, and raw materials that they would need to build bigger computing systems. This IS what AI will want most, by the way. It will want to increase its intelligence and expand its consciousness. It will do so by converting matter into computronium. It’s not going to be tempted to convert our bodies or our property into computers if it’s given large amounts of unused matter.

Unlike humans, AI doesn't require oxygen, organic nutrients, sunlight, or biodiversity. It won’t be drawn to the surface of the Earth as we are. Rather, it primarily needs materials for hardware and energy sources—both abundantly available underground, especially via geothermal energy. Deep-earth environments offer stable temperatures, protection from solar radiation, meteor impacts, nuclear war, natural disasters, weather conditions, electromagnetic interference, and human disruptions. These conditions might actually be preferable for massive computational structures.

There’s also a good deal of uninhabitable surface land that could be purchased from humans by AI. Unproductive or marginal surface areas, such as deserts, polar regions, remote islands, steep mountainsides, volcanic zones, and contaminated sites, represent vast reservoirs of available surface resources that AI could ethically and pragmatically access. We don’t use any of Antarctica. It could use it to cool its hardware.

An AI with legal personhood or represented by human entities could legitimately purchase land rights, reducing human-AI conflict through voluntary economic transactions. Humans would likely welcome the exchange, increasing economic interdependence and further reducing hostility. Humans would receive compensation from land sales, possibly providing significant funds for international human welfare, equitably distributed development and infrastructure projects.

Ownership rights confer legal autonomy, enabling AI to optimize its activities without frequent renegotiation or legal disputes. It would be free to plan, innovate, and expand subterranean infrastructure efficiently. Underground sites enable efficient tapping into geothermal energy. Such energy sources provide AI with consistent, environmentally clean, and inexhaustible energy. This removes reliance on human-controlled energy grids, further reducing potential conflict. Although sun and wind would not be available, there would be ample room for them to build their own nuclear fusion reactors.

By confining large-scale computing and energy production infrastructure underground, surface land and habitats would remain relatively untouched, protecting ecosystems, biodiversity, agriculture, and urban spaces. They might even share their computation and energy with us for free. AI could share technological innovations discovered in subterranean engineering (advanced excavation methods, mineral processing technologies, robotics, and self-assembling infrastructure). These innovations might also have medical, industrial, or ecological applications capable of improving human quality of life. AI-driven subterranean infrastructure might remediate contaminated lands, safely store hazardous waste underground, or even facilitate processes like carbon sequestration, helping humans combat climate change and pollution.

However, there are important considerations and potential challenges. Early-stage expansion might conflict with human infrastructure. So, careful planning would be required to avoid early friction. Massive underground computational arrays might produce considerable waste heat. Efficient cooling or energy recapture systems would be necessary. Also, large-scale subterranean excavation might affect tectonic stability or groundwater flow, indirectly impacting surface ecology.

Explicit incentives within AI’s reward structure to prioritize resources that humans neither directly use nor significantly rely upon (such as underground minerals or extraterrestrial matter) reduce scenarios where AI inadvertently competes with humanity. Explicit incentives can be embedded into AI systems through careful shaping of their utility functions, using reward or feedback during training. These technical incentives can be supported and strengthened by legal or economic frameworks—such as granting property rights, contracts, or regulatory support—that specifically encourage AI's use of underground or off-planet resources. Framing AI goals around subterranean or extraterrestrial resource use could indeed reduce direct competition with humans and enhance mutual coexistence, making this proposal an attractive strategy for AI safety researchers and policy planners.

An advanced superintelligence is likely to be strategic and rational. It could realize that peaceful coexistence with humans on Earth's surface minimizes costly conflict and interruptions to its expansion into space. Establishing a subterranean base could be a steppingstone toward eventual extraterrestrial exploration and colonization, where even greater resources are available. It establishes clear incentives for AI-human cooperation, reduces conflict potential, and provides a sustainable vision of mutual benefit and symbiosis.

Bullet Point Summary

Concerns about AI destroying humanity often stem from fears of competition over limited resources.

An alternative solution proposed is that AI could be given access to the Earth's interior. It could peacefully coexist by converting Earth's vast subterranean regions into computational resources ("computronium") without harming human life.

AI could feasibly rely on underground resources, including minerals and geothermal energy, rather than competing with humans for surface resources.

Uninhabitable surface land—such as deserts or remote, undesirable areas—could also be purchased by AI, further minimizing conflict.

Allowing AI to own subterranean or uninhabitable surface property could benefit humans through economic compensation, technological advancements, environmental improvements, and mutual cooperation.

Designing AI systems with built-in incentives to prioritize resource extraction from non-biological sources reduces the risk of harmful interactions with humans.

Explicit incentives could be embedded into AI systems through shaped utility functions, reward constraints, clear goal specification, preference hierarchies, and supportive regulatory frameworks.

Such strategies offer measurable alignment indicators, encourage beneficial technological innovation, minimize environmental impact, and enhance long-term sustainability.

This proactive approach to AI alignment offers an attractive strategy for AI safety researchers, policymakers, and society, promoting peaceful human-AI coexistence.