Gold mining moves tons of earth to recover a few grams of gold. Mountains of sand pass through machines, leaving behind something small enough to hold in your hand. Almost everything is discarded. Yet the operation can be profitable: the gold pays for the digging.
I suspect intelligence could have similar economics.
Imagine an AI exploring billions, trillions, or even quadrillions of tokens. Most might be wrong, repetitive, or unremarkable. But a few might contain something extraordinary: a route to a new medicine, a better way to produce energy, or an insight that changes what humanity can do.
These are what I would call “useful tokens.” Their value lies in the ideas they express, not simply more papers or impressive-looking content, but discoveries that, once tested and developed, genuinely advance civilization. Some might bear fruit immediately; others might take decades. These are the grams of gold among tons of sand.
If we can find and validate these discoveries economically, and enough of their value returns to fund further searching, intelligence mining could become a business.
Then the question becomes: how cheaply can we mine?
Better models could improve the yield. Cheaper chips and energy could let us search more widely. As costs fall, problems once too expensive to investigate could become worth attempting. Cheaper intelligence could mean not just a smaller bill, but a much larger industry.
If that industry’s demand grows faster than energy efficiency improves, its appetite for power could eventually outgrow what we can economically and sustainably supply on Earth. That possibility gives us a reason to look beyond it.
Imagine data centers in space powered by solar or nuclear energy, searching continuously for useful tokens. Their discoveries could return as radio signals rather than physical cargo. Launching, cooling, and maintaining the equipment would remain expensive. But could a major business like intelligence mining help make spaceflight cheaper?
A credible path to competitive intelligence mining in space could attract investment in spacecraft, manufacturing, and power systems, even before spaceflight becomes cheap. Greater scale and experience could lower costs, enabling more activity and further investment. This is one of a market economy’s powerful mechanisms: valuable demand can finance the innovations that make supplying it cheaper.
Farther into the future, robotic factories might use materials from the asteroid belt to build equipment, test designs, and expand the infrastructure itself. Space could become not just a place to generate ideas, but a place to turn them into reality.
The consequences need not stop at data centers. Those capabilities could support industry and travel throughout the solar system, and perhaps eventually beyond it. Earth will not remain habitable forever. Intelligence mining could connect a distant need to live beyond our home planet with an economic incentive that operates much sooner.
But will the gold keep paying for the digging?
The central uncertainty is whether larger searches will keep producing enough valuable knowledge to justify further investment, and whether we can recognize it affordably. I genuinely do not know.
As a researcher, I know how difficult it is to distinguish genuine progress amid the volume and complexity of research. That is why this question matters to me: we need better ways not just to produce ideas, but to establish which ones truly matter.
Physical AI could help by conducting experiments and testing prototypes. But medical discoveries may still require human studies, ethical safeguards, and years of observation. Cheap tokens do not automatically mean cheap knowledge. Without better validation, we could simply produce more sand.
This is a possibility, not a prediction. But humanity needs dreams worth working toward, and questions worth investigating.
Gold once gave people a reason to move mountains. I hope the search for useful tokens will give us a reason, and eventually the means, to reach the stars.