The Absurd #3: The Game That Beats Every Game

Dear reader,

I like capitalism. It built the world I live in, the tools I use every day, and the company I am building. And still, I think one of the hard lessons of capitalism after AGI, the AI that can do most thinking work as well as a person, will be that capitalism is the worst model we have for handling it. My reason fits in two sentences. Knowledge does not follow the laws of economics, because it is not scarce. So a capitalism geared to harness knowledge gain, or intelligence proficiency, will beat every other capitalist game out there, and a game that one player always wins is no longer a game.

Capitalism is a game played under scarcity

Economics, by its most famous definition, is about scarcity. The economist Lionel Robbins put it in 1932 as the study of "human behaviour as a relationship between ends and scarce means which have alternative uses." Prices, wages, profit and competition all exist because there is not enough of something to go around.

Every player in that game hits limits. A worker has only so many hours. A factory has only so many machines. A farm has only so much land. Economists since David Ricardo, writing in 1817, have called the last of these diminishing returns: each extra unit of the same input gives you less than the one before.

Those limits are what keep the game open. No single player can make everything, serve everyone, or be everywhere at once. So there is always room for a rival, and the rivals keep each other honest on price and quality. That is the bargain I like about capitalism: everyone chases their own gain, and competition turns it into something useful for the rest of us.

Knowledge has no limit to hit

Knowledge breaks every one of those limits.

It is not used up. Thomas Jefferson put it well in a letter in 1813: "He who receives an idea from me, receives instruction himself without lessening mine; as he who lights his taper at mine, receives light without darkening me." Economists call this being non-rival: my using an idea does not stop you using it.

It costs almost nothing to share. Finding something out can be expensive. Using it again, or giving it to the next person, costs close to nothing.

And it compounds. Every thing you learn makes the next thing cheaper to learn. Paul Romer won a Nobel prize in 2018 for showing what this does to growth: because ideas are non-rival, they give increasing returns, the opposite of Ricardo's diminishing returns.

That is what I mean when I say knowledge defies the laws of economics. The two rules every other input obeys, scarcity and diminishing returns, simply do not apply to it.

The game that beats every game

Now put that inside capitalism.

Most advantages in business are local. A better factory wins one market. A better shop wins one street. A rival can always build a better factory or open a better shop.

Intelligence is not local. The same ability to think can write the contract, design the product, code the app, check the accounts and plan the next move. So a player who harnesses knowledge gain is not competing in one market. It is competing in every market that runs on thinking, which is nearly all of them.

And each win makes the next win cheaper. More customers bring more money and more experience, which buy more learning, which makes the product better, which brings more customers. The economist Brian Arthur showed, from the late 1980s on, that markets with increasing returns tend to tip toward a single winner. With knowledge gain, it is not one market that tips. It is all of them, one after another.

This is the thesis of this entry. The most capitalist move there is, owning the engine of knowledge gain, beats every other capitalist game. And by winning every game, it ends the competition that made capitalism work in the first place. The reward for winning is the end of the game.

Neither labor nor capital

Eric Weinstein, the mathematician and podcaster, comes at the same problem from another side.¹ His point, as I understand it, is that our economy runs on the relationship between capital and labor, and AI does not fit neatly into either.

Adam Smith, in The Wealth of Nations in 1776, split the price of everything into three parts: wages paid to labour, profit paid to stock (his word for capital), and rent paid to land.² People did the work, machines made them faster, and the owners of each got paid.

AI does the work a person was paid a wage to do: writing, coding, analyzing, designing. But it is not a person. It takes no wage, does not tire, and can be copied as many times as there is hardware to run it. It is owned the way a machine is owned. So whoever owns it collects what used to go to labor and what used to go to capital, at the same time.

Weinstein's own proposal, as reported, is to treat a person's place in the job market as something like property, which a company would have to license before replacing that person with AI. Whatever the right answer turns out to be, his question and mine meet in the same place. Capitalism was built on players with limits, and AI gives one player a way past them.

Capitalism after AGI: knowledge should not be a company's product

Knowledge, knowledge gain, and the experiments that produce it should not be turned into products that companies own and sell.

Selling something that is not scarce as if it were scarce means turning people away from a thing that would cost almost nothing to give them. And owning the engine of knowledge gain is not like owning a factory. It is owning the one advantage that keeps growing, in every market at once. Making it a product hands that advantage to whoever is already ahead.

There is a second reason, and it is the auditor in me speaking. In my own field, AI alignment research, a safety result kept inside one company cannot be checked by anyone outside it. You cannot audit what you are not allowed to see. The experiments that tell us whether these systems are safe are exactly the knowledge that most needs to be shared.

To be fair to the system I like: capitalism paid for a great deal of this. Someone had to build the data centers, pay the researchers and run the training, and profit is a strong reason to do hard, expensive things. And we know knowledge can be produced in the open. The Human Genome Project was paid for mostly by governments and a charity, and in 1996 its scientists agreed in Bermuda to release every sequence they read within a day.

Why it is absurd

The absurd, for Camus, is the gap between what we expect from the world and what it gives back.

I expected capitalism to stay a game with many winners, because every player runs into limits. What I am watching instead is a resource with no limits entering the game, and the system rewarding whoever owns it more and more each round. I am building a company and I sell a product, Cheap Recorder, in that very market.³ I like the game. And I can see that its best move ends it.

The Stoic split helps me hold both. Who owns the engines of knowledge gain is not in my control. What I publish openly, what I check, and what I share from my own research are.

Questions I am living with

• If one player can win every market that runs on thinking, is there still a market?

• What is a fair price for something that costs almost nothing to share?

• Who should own the experiments that tell us whether AI is safe?

• What would it take to run knowledge gain as something we all share, the way the genome was?

• And if capitalism's best move ends the game, what game do we play next?

Frequently asked questions

Why does knowledge defy the laws of economics?

Economics is built on scarcity and diminishing returns. Knowledge is not used up when shared, costs almost nothing to pass on, and makes the next discovery cheaper, so neither rule applies to it.

Why would a capitalism built on knowledge gain beat every other capitalist game?

Most advantages work in one market, but intelligence works in every market that runs on thinking, and each win makes the next one cheaper. A player who owns knowledge gain keeps pulling ahead everywhere at once, which ends the competition capitalism depends on.

What does Eric Weinstein mean when he says AI is neither labor nor capital?

As I understand him, our economy runs on capital and labor, and AI fits neither: it does the work of a person, but it is owned like a machine. Whoever owns it collects what used to go to both.

If you like capitalism, why call it the worst model for AGI?

Because it works through players who all run into limits, and AI removes the limits for whoever owns it. Capitalism rewards that owner so well that the competition it runs on disappears.

Until the next absurdity,

Miguel