The Search for the Final Number

On the nature of numbers and the future of humanity

Somewhere right now, a warehouse full of machines is guessing numbers. It will guess trillions of them this second, throw almost all of them away, and keep one. A few miles away, another warehouse is doing something that looks different but isn't: nudging billions of numbers up and down, over and over, until they settle into one particular arrangement.

The first warehouse is mining Bitcoin. The second is training an AI model. Between them, they now draw about as much electricity as a mid-sized industrial country. And if you strip away the vocabulary of finance and intelligence, both are doing the same thing: searching for a specific binary string.

That is a strange thing for a species to be doing. This essay is an attempt to take that strangeness seriously, and to ask where the search leads.

From bookkeeping to treasure hunt

For most of history, numbers were tools for keeping track of things we already had. Notches on bone counted days or animals tens of thousands of years ago. By around 3200 BCE, Sumerian scribes were pressing tallies of grain and sheep into clay, and writing itself seems to have grown out of that bookkeeping. Numbers were servants: they described the world so we could organize it.

Then, slowly, the relationship flipped. Pythagoras claimed that "all is number." Physics found that the universe obeys equations. Computers turned every photo, song, contract and conversation into strings of 0s and 1s. Even the page Google shows you is, at bottom, one very long binary number, and it changes every few seconds.

Silicon Valley built its promise on top of this flip: get the numbers right, and life gets better. Better ranking, better targeting, better predictions, better models. Somewhere along the way, numbers stopped being only the map. They became the territory we explore, and the treasure we dig for.

The two great hunts

Mining. A Bitcoin miner takes a block of transactions, adds a number called a nonce, and runs the result through a hash function. If the output starts with enough zeros, the miner wins. If not, it tries the next nonce. That's it. There is no shortcut and no cleverness, only guessing, at a pace of roughly a sextillion (10^21) guesses per second across the network. The winning number has value for one reason: everyone agreed in advance that it would.

Training. An AI model is a few billion (soon trillions of) parameters, each one a number. Training means showing the model data, measuring how wrong it is, and nudging every parameter slightly in the direction of less wrong, again and again for months. The architecture, meaning how the artificial neurons connect, can itself be written down as a binary string. So a finished model is, quite literally, one very long and very specific sequence: 010101110…

The two hunts differ in an important way. Mining searches blindly for a number that is arbitrary; its only virtue is being hard to find. Training searches with a compass for a number that is meaningful; it compresses something true about the world. But the shape of the activity is the same: pour energy into machines until a particular string falls out.

And around both, we keep inventing new algorithms, chips and protocols whose whole job is to find these strings faster. Which raises the question underneath this essay: are these numbers going to serve humanity, or has the purpose of humanity quietly become finding these numbers?

The energy ledger

Roughly 2% of the world's electricity now goes to data centres and Bitcoin mining combined, and that share is on track to grow. So what percentage of our energy goes to finding numbers? Here is the best current answer:

Number hunt	Electricity	Share of global use	Source
Bitcoin mining (2024–25 survey)	~138 TWh/year	~0.5%	Cambridge CCAF, 2025
Bitcoin mining (Dec 2025, preliminary)	~190 TWh/year annualized	—	Cambridge, via The Energy Mag
All data centres, 2024	~415 TWh	~1.5%	IEA, Energy and AI
All data centres, 2030 (projected)	~945 TWh	just under 3%	IEA, Energy and AI

Two honest caveats. Not every data centre is training AI; in 2024, AI servers were about 15% of data-centre energy, though AI is the main driver of the projected doubling (Scientific American / Nature, citing IEA). And 2% sounds small, until you remember what it's for. That is the electricity of entire countries, spent not on heating homes or moving trains but on looking for strings of bits. A civilization reveals its priorities by where it sends its energy. We are sending a growing share of ours into the search.

Do we even know a number?

Ask almost anyone and they'll say we know Pi. It's just the ratio of a circle's circumference to its diameter: 3.14159…

But look at that ellipsis. Pi is irrational: its digits never end and never settle into a repeating pattern. The current record, set in late 2025, is 314 trillion digits (ScienceAlert). That is an astonishing feat, and it is exactly zero percent of Pi. Cover the Earth in computers, send robots to build more on other planets, run them until the stars burn out, and you will still have zero percent of Pi.

So in what sense do we know it? Not by holding its digits. We know Pi because we have a short recipe that can produce any digit we ask for. A few lines of code stand in for an infinite string.

That is a quietly radical idea of knowledge: to know a number is to have a short description of it. Hold onto it, because it will matter at the end. Most numbers have no short description at all. For them, the only way to "know" them is to have every digit, and that is impossible.

The oldest search

Here is where the story bends. Searching for numbers is not new at all. It is the oldest thing life does.

DNA is a number. Four letters, A, C, G and T, each worth two bits. Your genome is about 3 billion of them, a binary string of roughly 6 billion digits that fits in less than a gigabyte. For about four billion years, life has run one algorithm: a number interacts with the physical world, survives or doesn't, and produces a slightly different number. Two humans combine their strings, and a new string is born.

Number + human → new number. The same loop runs in carbon and in silicon.

Read the loop again and the two warehouses from the opening fit right in. A model's weights meet the world through data, get scored, and are replaced by a better set. The medium changed, from carbon-based computation to silicon-based computation, and the clock sped up from generations to milliseconds. But the concept is the same. We are a number looking for the next number.

If that's right, the search was never something humans started. It's something we are in, and it may be a forever journey.

The weight of the human link

Look at the diagram again and notice the middle box. In the biological loop, a human body is the indispensable link: no body, no new genome. In the silicon loop, humans are still in the middle, choosing the data, writing the code, deciding what counts as "better." But our share of the work is shrinking. Models already help design chips, write training code and generate their own training data.

Here is a definition of the singularity that needs no science fiction: the singularity is the moment the weight of the human link drops to zero. The loop keeps running, number to world to new number, and nobody needs to be in the middle anymore.

Biology hints at how strange that could feel. We tend to imagine parents as the ones in control, but evolutionary biologists have shown that the fetus is not a passive passenger. Genes in the fetus and placenta actively pull resources from the mother, sometimes against her own interests, in what researchers call parent–offspring conflict. The child's code steers the mother's body, not only the other way around. The new number has its own agenda from the start.

And the current runs both ways. With tools like gene editing, we are beginning to rewrite the number that wrote us. Think about how much human attention, time, risk and art orbits a single obligation DNA installed in us: the drive to reproduce, to copy the string forward. Once we can edit the code, some of those obligations become optional. The question is who, or what, decides which ones.

The final number

If everything is a binary string, then everything that could ever be known is somewhere in the space of strings. Borges imagined this as the Library of Babel: a library holding every possible book, including the true history of the future and a perfect explanation of the universe. The catch is that it also holds every false version, and nobody can tell which is which. Having all the strings is worthless. Finding the right one is everything.

By many definitions, God is the all-knowing. Few people have claimed direct access to such a God, and never universally. So here is a thought experiment: what if God, in the language of this essay, is the final number, the one string that contains the answer to every question?

Mathematics has a surprising candidate. In the 1970s, Gregory Chaitin defined a number called Omega: the probability that a randomly written computer program eventually stops. Omega is a perfectly well-defined number between 0 and 1. And its digits encode an astonishing amount of truth. If you knew enough of them, you could settle famous open problems such as Goldbach's conjecture, simply by reading them off.

But Omega has a fatal property. It is provably uncomputable. There is no short recipe for it, as there is for Pi. Its digits look like pure randomness, and any computer, any proof system, any civilization can pin down only finitely many of them. You cannot mine Omega. You cannot train your way to it.

That gives this essay an unexpected answer. If there is a final number that knows everything, mathematics suggests it is exactly the kind of number no search can find. The all-knowing string exists, and it lies permanently beyond the reach of compute.

Back to the animal

Once there was an animal, fairly simple, with a short and countable list of needs: food, shelter, warmth, intimacy, belonging, a little meaning. A nuclear reactor is not on that list. Neither is a 10-trillion-parameter model or a winning hash.

The promise of modern civilization has been that inventing more things will make tomorrow better than today. Often it has: vaccines, clean water and electric light really did change what it means to be that animal. Some of the numbers we find collide with the physical world and become machines, medicines and bridges.

But many don't. A winning nonce secures a ledger and is discarded ten minutes later. A model is replaced by a better one within months. The search is generating numbers far faster than it is generating answers to the old, countable list. So it's fair to ask whether the search still serves the animal, or whether the animal now serves the search.

Where are we converging?

Put the pieces together and a picture emerges. Life has always been a number searching for the next number. We have moved that search from carbon to silicon, and we are pouring a growing share of the planet's energy into it. Our role in the loop is shrinking toward zero. And the one number that would answer everything appears to be, by mathematical proof, unreachable.

That last point may be the most hopeful one. If the final number cannot be found, the search never ends, so the question was never when do we arrive? The question is what is the search for, while it lasts? A search that keeps feeding the animal, its health, its people, its sense of meaning, is a civilization. A search that only produces more numbers is a ritual.

We get to choose which one we're running. For now, at least, we are still the middle box.